This PDF file contains the front matter associated with SPIE Proceedings Volume 12079, including the Title Page, Copyright information, and Table of Contents.

We propose a pose detection algorithm for robot arm grabbing square building blocks in this paper. The main innovations of this method are:(a) A new method combining image segmentation and neural network is constructed. (b): A method for self-correction of posture detection results is proposed, and. (c) The image segmentation method is used to realize the function of automatically marking the position and angle of the target object. Compared with traditional methods that only use image segmentation or neural networks for detection, the method in this paper has absorbed their advantages, which has both the robustness of neural network and the accuracy of image segmentation. Using the method in this paper on the Xarm7 manipulator arm has achieved a grabbing success rate of 92%, which proves the effectiveness of the method in this paper.

Benefiting from the superiority visual capacity of infrared imaging system such as long operating distance and all-weather work, infrared target detection is widely applied in weapon guidance, security protection and other fields. Aiming at the characteristic of infrared targets, we propose an infrared detection algorithm based on feature fusion. Firstly, gaussian filter and background suppression are employed to improve signal-to-noise ratio and increase the contrast between target and background, respectively. Secondly, a feature fusion architecture is added to backbone for feature extraction, so as to obtain features with rich information. Finally, in order to solve the problem of imbalance between positive and negative samples, the network is trained by focal loss to improve training performance. We evaluate the proposed detection algorithm on public infrared image datasets. The results demonstrate that the proposed algorithm can achieve promising performance.

With the development of autonomous driving technology, the level of intelligent vehicles has also become higher and higher. However, in the process of upgrading the level of automatic driving, the upgrading of the intelligent level of vehicles cannot effectively solve a series of functions that need to be realized in the process of high-level automatic driving, such as global traffic intelligent planning and scheduling, and beyond visual range perception. Combining intelligent vehicles with cloud computing technology to develop a cloud data platform is a practical and feasible solution for the development of high-level autonomous driving technology. Based on the key projects of the Intelligent Connected Laboratory, this paper conducts research on the rapid implementation plan of the small standardized intelligent connected vehicle cloud data platform, using the Internet of Things protocol Message Queuing Telemetry Transport to realize the intelligent vehicle-cloud platform’s data communication, MySQL database storage analysis and web front-end data visualization construction plan for related system development and experimental verification.

Implementing accurate and precise brain tumor MRIs classification in diagnosis for patient treatment is a pivotal part of Computer-Aided Diagnosis (CAD) for medical applications. Our work focuses on glioma, meningioma, and pituitary tumors, three typical types. To extract latent features from brain MRIs more efficiently, we adopt transfer learning and use EfficientNet in our classification system, gaining a 4-category classification (3 types of tumors and no tumor) accuracy of about 98%. In the experiment, comparing some popular transfer learning neural networks, we find out that as the depth of neural network increases, the classification accuracy increases, which means neural networks with more convolutional layers have stronger discrimination. To make this work available for the public, we implement our method as an online application with more flexibility by developing a Speech Synthesis System and User Interface.

This paper proposes an algorithm for real-time detection of the user's finger pointing. The innovations of this algorithm are: (a) In order to enhance the robustness of the detection results, the Kinect data is comprehensively used to segment the hand, and the depth value is divided into three situations to better avoid interference ; (b) In order to improve the real-time performance of the detection results, the position of the fingertip is obtained by using the contour information of the hand, thereby narrowing the scope of the solution ; (c) In order to improve the accuracy of detection, the Hough line detection algorithm is used to detect possible straight lines in the contour, and the angle is compared with the straight line from the fingertip to the palm of the hand. The algorithm in this paper can realize real-time pointing detection and has achieved good results.

The degree of matching between the skills acquired by college students when applying for jobs and the skills required by companies for online recruitment of talents is low, resulting in a low employment rate. In response to this problem, it is proposed to apply the Spark-based parallel FP-Growth algorithm to the analysis of job requirements for recruitment information, extract the job skills corresponding to the job, and improve the matching degree between graduates and the recruitment market demand. Compared with the traditional FP-Growth algorithm, the Spark-based parallel FP-Growth algorithm has better mining for massive data sets.

Aiming at the difficulties and pain points of garbage quality identification in the transfer process, technical demonstrations are mainly carried out from the aspects of image preprocessing, deep learning algorithm selection, and model training. A model of dry and wet garbage abnormalities in complex scenarios was established. The specific garbage image preprocessing method, deep learning algorithm and model training technology used in this project are determined. A set of intelligent identification system for automatically detecting the quality of garbage classification has been developed, and all the identification indicators of the system are better than expected.

The Internet network system extends the connection objects to the entire economic entity system, the entire industry chain, and the entire value chain. It can realize the ubiquitous and deep interconnection of the entire chain of human, financial, and material from R&D and design to production, management, and service. Promoting the chain-network integration of blockchain and enterprise networks is expected to solve the dilemma of efficiency and risk. We adopt a comprehensive literature research method to focus on the theoretical evolution and application development of enterprise networks and blockchains, explain the main logic of the integration of blockchain and enterprise network theory and chain networks from a theoretical perspective, give the definition of enterprise networks, and innovate to establish chain- network integration model and correlation measurement model. It provides basic theoretical support and practical reference in order to open up the information, trust and credit barriers between the network and the network based on the blockchain technology for the industrial Internet, complex enterprise networks or hyper-networks, and to improve the overall operating efficiency.

In middle school intelligent chemistry experiments, experimenters often perceive the scene information through Kinect and other fixed cameras. There is a serious occlusion problem, which makes this method almost infeasible. Therefore, this paper puts forward a research idea of setting a micro camera on the beaker for the first time. The main contribution of this work is to propose a 3D reconstruction algorithm of experimental scene based on mobile camera. The main innovation lies in the process of obtaining two images of the same laboratory scene from different angles through a moving binocular camera, then using the point cloud to reconstruct the three-dimensional shape of the experimental instrument in the scene and restore the spatial position information of the object. In order to reconstruct the experimental instrument information better in the chemical experiment scene, this paper integrates the model of transforming the picture from RGB to HSV based on the traditional binocular 3D reconstruction algorithm, which is also another innovation of this paper.

Two different methods are used to calculate the air resistance of high-speed monorail pulley. The calculation of rocket sled Air Drag Coefficient is the key technology to realize rocket sled trajectory accurate control. The commonly used methods for calculating the Aerodynamic Drag Coefficient of rocket sled are the wind tunnel test method and the sled integral decomposition method. In Wind Tunnel Test, the wind tunnel wall influence, the test result error is big. The whole decomposition method is simple and the settlement result is more accurate. Firstly, the relationship between the pulley speed and the air drag coefficient is obtained by discretizing the measured air drag coefficient curve. Then, the wind resistance coefficient of the high-speed monorail pulley is calculated by the whole decomposition method. At last, the air resistance of pulley under different wind speed is calculated by finite element method. Compared with the finite element method, the error of the whole decomposition method is less than 10% . When the air resistance coefficient of the pulley increases by 0.02, the maximum speed of the pulley decreases by 30m/s, which provides a basis for the design of the high-speed monorail pulley. This method provides technical support for the design of high-speed monorail train, and calculates the wind resistance coefficient of the pulley by using the integral decomposition method and the finite element simulation method

The accurate regulation of the flow rate of the inflow control valve is the core technology of production fluid control in intelligent completion. The accurate monitoring of the inflow control valve opening is the key technology to realizing its intelligent closed-loop control. The existing hydraulic inflow control valve in China has no sliding sleeve position sensor and thus cannot accurately position the opening of the sliding sleeve and achieve the precise control of production. In this study, the position sensor of the inflow control valve was developed by adopting a concentric linear structure, which can accurately measure the position of the sliding sleeve of the inflow control valve, integrate a single-core cable power supply and signal transmission technology, and share single-core steel pipe cables with the downhole electronic monitoring systems of intelligent completion in China. In China, the position sensor of the inflow control valve can be simply and quickly connected with the hydraulic inflow control valve without the need to modify the structure of the hydraulic inflow control valve, which is convenient for quick use and has strong versatility. Field test results show that the position sensor of the inflow control valve realizes the accurate positioning of the hydraulic inflow control valve and achieves the purpose. The development of this tool bridges the gap in the opening measurement of the hydraulic inflow control valve in China and lays the foundation for realizing the closed-loop intelligent control of the inflow control valve of intelligent completion.

In order to realize the wide range transmission characteristic test of a set of gears at the same time, this paper proposes a wide range gear working characteristic test system. The test system can be used to test various gears of different specifications and types under quasi-static and dynamic conditions. In this paper, the test object, working range, test item, working principle, structure composition, driving loading principle and control testing principle of the test system are described; for the test piece, the finite element model of gear transmission system suitable for quasi-static loading tooth contact analysis is established and the loading contact analysis is carried out; the test bench for testing item verification is developed, and the tooth contact area is introduced Test principle of detection and transmission error detection; on the basis of verification test-bed, the tooth contact area and gear transmission error detection are carried out, and the test results are compared with the simulation results, and the analysis results show that the two are consistent. The test system developed in this paper can obtain the parameters of gear transmission in the whole work field. which will provide comprehensive test basis for gear design, manufacturing and assembly.

Special shaped stone products have the characteristics of individuation, diversity and artistry, and are mainly produced in single piece and small batch, which requires its processing equipment to have flexible production capacity. Traditional manual processing and simple stone equipment can not meet the high-speed production needs of special-shaped stone products. According to the special process of special-shaped stone, the structural topology optimization calculation and analysis of beam rib plate structure is carried out by using MATLAB programming language, and the sand cleaning hole of beam are optimized and analyzed by using topological opt module in ANSYS, and a new beam structure is obtained. The analysis results show that compared with the original scheme, the new scheme not only reduces the weight, but also effectively improves the static and dynamic performance of the beam, achieves a good optimization effect, and lays a theoretical foundation for practical production.

In order to optimize the delay of data transmission and the routing efficiency of routing search algorithm, an adaptive routing search algorithm for SRIO network based on chaotic ant colony algorithm is proposed. This paper analyzes the principle of SRIO network transfer transaction, and selects routing information by exchanging properties of SRIO network packets. Using chaotic ant colony algorithm, the LEACH protocol of SRIO network is established to search the best relay node adaptively, and the SRIO network routing adaptive search algorithm is designed. Experimental results show that the proposed algorithm can effectively reduce the delay of data transmission and optimize the routing efficiency.

The new generation of network security threats present a mixed and multi-stage attack mode. Using AI technology for accurate detection has gradually become a consensus in the field of network security. The paper briefly describes the background and research status of threat detection and traceability, puts forward a network security threat detection and traceability system based on AI, and designs the overall system architecture of network data acquisition and traffic processing, security threat detection engine, situational awareness and traceability evidence. According to different attack types, AI based detection engine is studied, and threat intelligent detection and location services for cyberspace and application system is developed by using cloud security service platform. Through AI detection, intelligence detection, anomaly detection and traffic gene detection, real-time monitoring, traceability analysis and emergency disposal of network security threats are realized. At the end of the paper, three typical threat detection application scenarios are recommended.

Music is an indispensable part of people's lives. It can make people relaxed and can become a way for people to socialize. However, it is a big challenge for ordinary people to generate music, because they do not have professional music knowledge. Deep learning can solve this. By using the LSTM model, we can convert music into music scores and express them with specific numbers. Then pass the data packet into the model, and the model will guess the next music based on the existing music data, so as to achieve music generation.

This paper sets up the research progress on the influence of user-side management on user preference elasticity, and classifies the influencing factors. Combing with sorts of user preference and new power system, terminal consumption behavior analysis based on "ubiquitous power Internet of Things"is carried out. This article uses behavior data from Jinan as sample to study demand response factors on user elasticity. These conclusion can be applied in consumers' consumption behavior based on "ubiquitous power Internet of Things.

To address the challenges of preparing C4 olefins, this study processes the data by using regression analysis, calculating its Spearman correlation coefficient, analyzing the test results of a given catalyst combination at 350 degrees at different times in an experiment, as well as analyzing various catalyst combinations and temperature to determine the effects of ethanol conversion rate and C4 olefin selectivity. Moreover, based on the machine learning XGBoost algorithm, this study establishes a regression model between the catalyst combination and temperature and the yield of C4 olefins. At the same time, this study further explores the variable values when the C4 olefins yield is as high as possible. For instance, the catalyst carrier, Co loading, total Co/SiO2, Co/SiO2 and HAP loading ratio, ethanol concentration, temperature, and six other elements are considered as the independent variables. Then, C4 olefin yield is considered as the dependent variable. The XGBoost algorithm is then used for integrated training. Then, the six most important ranking features are selected. Finally, the trained model is reverse-optimized and, to further optimize the catalyst combination and temperature values when the C4 olefin yield is as high as possible, the grid search method is applied.

Today, with the initial application of 5g technology, distributed storage system has been widely used in the storage management of big data because of its strong service ability, low cost and easy expansion. Firstly, this paper introduces two fault-tolerant mechanisms in distributed storage system, and points out the advantages of erasure code technology; Then the LDPC code and bit reversal decoding algorithm are briefly introduced; Finally, an efficient decoding mechanism is proposed and an example is analyzed.

In this paper, we enumerate perfect matchings of a special class of graphs by the Pfaffian method, which was discovered by Kasteleyn. Primary discussion for the skew energy of a Pfaffian digraph.

The raw materials of production enterprises are an indispensable factor to ensure the production capacity of enterprises. In this paper, we first select the evaluation indexes, obtain the weight vectors of all indexes through AHP, verify that they have passed the consistency test, and establish a multi index weighted grey target decision-making model to complete the evaluation of the importance of suppliers. Then we use 0-1 programming and dynamic programming algorithms to analyze the ordering and transportation of raw materials in production enterprises from multiple angles, and finally determine the best ordering strategy and the transportation scheme with the least loss.

Brain tumors are a hazardous type of tumor, and they build pressure inside the skull when they grow, which can potentially cause brain damage or even death. Attention mechanisms are widely adopted in state-of-the-art deep learning architectures for computer vision and neural translation tasks since they enhance networks' ability to capture spatial and channel-wise relationships. We offer an attention-based image segmentation model that outlines the brain tumors in Magnetic Resonance Imaging (MRI) scans if present. In the paper, we mainly focus on integrating Squeeze-and-Excitation Block and CBAM into the commonly used segmentation model, U-Net, to resolve the problem of concatenating unnecessary information into the decoder blocks and attempt to locate the tumor boundaries. Our research clearly shows the application of the attention mechanism in U-Net, incorporates the Squeeze-and-Excitation with CBAM, and improves the performance in the brain tumor segmentation task. The model is delivered on an app with additional text to speech and chatbot features provided.

This paper presents an intention understanding method for intelligent chemical experiment system. The innovations of this method mainly include, (a) Constructing an intention database according to the standardization of chemical experiment operation, (b) Quantifying and fusing the behavior information of different modes, constructing mathematical models, and calculate behavior intention probability of experimenters in real time, and (c) According to the predicted intention, the machine carries out active human-machine cooperation. The method in this paper can effectively obtain rich intention information from the experimenter's operation behavior which will help to establish a natural human-computer interaction environment and realize the active cooperative work between human and machine. The method in this paper has been evaluated and verified in an intelligent chemistry experiment system based on Unity.

In the actual battlefield, training and exercises for submarine and ASW operations are often costly and poorly repeatable. Using computer-generated forces (CGF) to interact and operate in a simulated battlefield environment is a repeatable, efficient, and low-cost means.The knowledge base and knowledge base editor of the MAXSim simulation platform provide the functions of generating forces, editing tactics, generating scenario and situation deduction. It can be used for the study of submarine and Anti-submarine platform combat tactics, and lays the foundation for the construction of large submarine and ASW platform simulation combat systems.

Path-planning as one of the most fundamental tasks of robots should cause less burden on calculation capacity and be efficient when the robots’ major tasks are not only about motion. Artificial Potential Field (APF) algorithm as an effective and simple path-planning method could meet such a requirement. However, the local minimum problem limits its application on robot’s path-planning tasks. To deal with this problem, this paper proposed a method named “Dynamic Repulsive Potential Factor” (DRPF). Compared to other existed improvements such as the sub-target strategy and combination algorithms with Rapidly exploring Random Tree (RRT), DRPF has the advantages of simple and effective. And the produced path of DRPF is shorter than the one produced by APF combing RRT, which is demonstrated in the experiment result. Meanwhile, the robot is continuing to move in the DRPF method while it is stopped when searching path in RRT, demonstrating the effectiveness on time aspect. The application of DRPF could meet the requirements of efficient and less calculation burden for path-planning function of robots whose major tasks are not only about navigation.

With the development of information technology and the change of the international situation, information security rapidly becomes the focus of national security, and autonomous control as the cornerstone of information security, can provide guarantee for the realization of information security. This paper expounds the relevant concepts of autonomous control and the significance of autonomous control evaluation, summarizes and compare the commonly used evaluation methods, takes some domestic basic hardware and software and typical systems as examples, summarizes the ideas and methods in the process of autonomous control evaluation, and prospects the next direction of evaluation.

This paper relates to a set of virtual simulation experiment system of the Zeeman effect developed according to the teaching requirements for the Zeeman effect experiment. The corresponding relationship among changes in the magnetic field, optical elements and Zeeman split spectral lines can be intuitively observed from the interface of the system. The system can help observe the Zeeman effect and study the atomic energy level structure so as to obtain better experimental teaching effect, shorter experiment duration and higher utilization rate and safety of the equipment.

With the development of science and technology, the smelting process of the metallurgical industry is constantly updated. The traditional temperature measurement method can no longer meet the process requirements. Therefore, the infrared on-line continuous temperature measurement method for AOD came into being. The mechanical transformation and temperature compensation of the bottom gun temperature measurement point make the temperature measurement system more perfect and improve the production efficiency of smelting.

The main research content of this paper is to improve and optimize the signal processing circuit of the electromagnetic current meter on the basis of the existing electromagnetic current meter technology to improve the measurement accuracy of the electromagnetic current meter. In order to collect and calculate data, we choose STM32 series MCU, which is an ultra-low power consumption 32-bit MCU with Arm Cortex M4 core. Its advantages are: strong performance, low power consumption, and fast frequency. Based on STM32L496, we carry out the signal processing circuit design of the electromagnetic current meter, the circuit includes: analog signal processing board, digital signal processing board. At the same time, we have completed the processing and debugging of the signal processing board of the electromagnetic current meter. After that, we completed the overall installation and joint adjustment of the electromagnetic current meter, laboratory tests, and conducted tank tests and marine tests, recorded and analyzed data records and analysis.

Limb deficiency is when individuals lose their limb’s normal function caused by amputations or congenital malformations. With the development of technology, people came out various method to deal with limb deficiency, and try to decrease its negative impacts as much as possible. In order to offer people a thoroughly guideline on what degree human have progressed on dealing with limb loss, this paper studies the limb prosthesis through various papers about different advancement in bionic limb technology. New materials, new design, and new techniques give people chances to upgrade bionic limbs. Looking through the bionic limb historically as well as functionally, this study found that the bionic limb is a prosthetic way people use to replace human original limb function by a artificial bionic limb.

High-order nonlinear optical fiber cross phase plays an important role in optical fiber applications. All optical logic circuits based on high order nonlinear optical fiber cross phase have the advantages of fast, accurate, reliable and distance free in the process of information transmission. In this paper, the principle of high order nonlinear optical fiber system and the basic concept of all-optical logic circuit are described, and the application of various serial decoding algorithms of all optical logic circuit in optical fiber communication is studied.

The cross-amplification modulation effect is a key technology for the optimization of highly nonlinear optical fibers. Based on this technology, the expression of multifunctional all-optical logic can be realized effectively. This paper introduces the cross-amplification modulation effect in the application of highly nonlinear fiber. From the perspective of all-optical logic, it is found that the impedance Z can be directly represented by the model, and it is a cross-amplification modulation commonly used in multifunctional analysis.

Mechanical vibration has a great influence on the design of mechanical system. Mechanical systems often need to design a reasonable vibration damping mechanism to maintain the normal operation of the mechanism the compliant mechanism has good guiding accuracy and cushioning stiffness. Through the analysis of the actual cushioning problem and the structural topology of the existing traditional, compliant unit, this paper selects Roberts four-bar mechanism as the primary design scheme of the respectful team, obtains the basic configuration of the extended compliant unit through the stroke expansion deformation of the selected mechanism, and carries out the stiffness modeling and analysis of the basic configuration unit of the compliant mechanism, to obtain the stiffness performance. Finally, the structural and element verification optimization is carried out according to the designed compliant mechanism. The central performance verification is carried out through FEA (finite element analysis) simulation. The results show that the designed mechanism has good stiffness buffer performance and meets the design requirements.

In order to study the reliability and safety of the multiple luffing mechanism of the full hydraulic drilling rig, the dynamic response characteristics of the multiple luffing mechanism under the maximum impact force were determined. Based on the theory of modal analysis, taking the ZDY2-3200LF full hydraulic drilling rig with multiple amplitudes as the research object, its natural frequency and various modes can be obtained through the ABAQUS finite element analysis software. On this basis, the harmonic response analysis is carried out on the maximum thrust of the full hydraulic drilling rig, and the displacement and stress spectrum characteristics at different positions are obtained. The research results provide a basis for the optimization design of other full-hydraulic drilling rigs with multiple luffing mechanisms, and also have reference value. The research results provide the basis, and also have reference value for the optimization design of multi-luffing mechanism of other full hydraulic drilling rigs.

Boiling water pot is simple structure, environmental protection and emission reduction, humanization, sturdiness and durability. In this paper, based on the analysis of the general scheme and leakage testing principle of leak testing machine of boiling water pot, the key issues of the design of the tire gear, the design of the vacuum system and the selection of special rubber materials were described. boiling water pot. This equipment is a new type of machining equipment with the advantages of high production efficiency, good reliability and use safety. The feasibility, reliability and stability of the design scheme have been proved in the practical application of the equipment in production, which can be used for reference by engineers and technicians engaged in the research in this direction.

Supercritical carbon dioxide (SCO₂) is widely used and studied in all over the world as a working fluid for thermodynamic cycles due to its excellent physical properties. In this paper, an elastic tilting pad journal bearing using SCO₂ as lubricating medium is designed. Based on the Navier-stokes equation and continuity equation, the theoretical model of SCO₂ hydrodynamic lubrication bearing considering turbulence effect is established by combining the nonlinear SCO₂ density and viscosity expressions, and the numerical method of SCO₂ hydrodynamic lubrication is proposed. The static and dynamic characteristics of SCO₂ lubricated tilting pad journal bearings are studied.

With the rapid development of China's economy, the design of bridges has also made considerable progress. Starting with the model requirements and competition rules of the comprehensive ability competition of engineering training, this thesis designs an arched truss bridge after comparing the concentrated bridge structures. The bridge has simple structure, clear and concise stress and beautiful appearance, and can give full play to the compression resistance of paper made from bamboo. The finite element analysis software ANSYS is used to analyze and calculate the strength, stiffness and stability of the bridge model. The experimental results prove that the designed bridge can bear the load of 20 kg.

Process parameters are an important part of the CNC machining system, and the reasonable selection of process parameters can significantly affect the energy efficiency of the machine tool. In order to reduce the energy consumption of the processing system and improve the energy efficiency, this paper proposes a method to optimize the process parameters of the processing system based on reinforcement learning. First, construct an objective function with process parameters as variables and energy efficiency and process time as constraints. Then it analyzes its effectiveness through actual turning processing cases. Finally, through comparative experiments, it is concluded that the method proposed in this paper can reduce energy consumption by 20% and the consumption of cutting fluid by 35%, which verifies its effectiveness and practicability.

A new type of turntable structure is optimized for the lightweight design objectives of aerial work vehicle turntable structure. First, the structural calculation model is established by the HyperWorks software, and the optimal material distribution of the turntable is obtained according to topological analysis based on the variable density method of topology optimization theory. Then, a new high-altitude work vehicle turntable structure is designed considering the actual application of the components. The rigidity of the designed structure is increased by 16.14% on the basis of a 10% reduction in weight compared to the original structure, which achieves the optimal design goal of weight loss of the high-altitude vehicle turntable structure.

Aiming at the problem of short service life of a rivet nut die for an automobile, this paper analyzes the problems existing in the existing production scheme and puts forward an optimized cold heading forming scheme, that is, to reduce the die stress at the corresponding station by transferring the deformation of important stations. The optimum combination of forming process parameters was obtained by orthogonal test design method. By means of numerical simulation and experiment, the cold heading scheme and process parameters were verified. The results show that the optimized scheme and process parameters can effectively reduce the die stress and prolong the die life.

Aiming at the RRT algorithm's strong randomness, the obtained path is not the shortest, and the vehicle differential constraints are not satisfied, this paper proposes an improved RRT smart car motion planning algorithm. In a priori information environment where the map is known, a reasonable metric function that satisfies the vehicle kinematics constraints is proposed, so that the path is smoother and facilitates the optimization of the back-end trajectory. In addition, a pruning optimization method that reduces path nodes and shortens the path length is also proposed. The simulation results show that the improved RRT algorithm has a smoother path than the basic RRT algorithm, with fewer path nodes and a shorter path length. Finally, a collision-free, smooth and continuous trajectory that satisfies the vehicle movement is obtained.

With the development of intelligent manufacturing technology, 3D printing technology is also developing rapidly.This paper mainly studies the thermodynamic changes of different nozzle materials in FDM melt deposition technology of 3D printer at a certain temperature.ANSYS software is mainly used to simulate and compare the heating of copper nozzle and aluminum alloy nozzle. Through comparison, it is found that the deformation, i.e. thermal expansion of aluminum alloy material after heating is greater than that of copper alloy material.

Under the background of the energy revolution, the energy structure is constantly optimized, the proportion of primary energy consumption such as fossil energy is continuing to decline, and low-carbon economy is an important trend of global economic development. However, coal power is the main power source positioning in the short term will not change. Therefore, we should constantly adjust the energy structure at the same time, strengthen the low-carbon use of primary energy, reduce carbon emissions. Reducing carbon emissions in the coal supply chain is an important measure under the "low carbon policy". Therefore, this paper mainly focuses on the coal supply chain carbon dioxide emissions research and analysis, discusses the main sources of carbon emissions and key factors, the construction of interpretation structure model for the analysis and ranking of nine major factors, and divide it into levels. Finally, according to the results of the analysis, some rationalization suggestions are put forward to the relevant enterprises in the coal supply chain to realize the low carbon development of coal supply chain network.

In the past decade, the amount of data in the petroleum industry has increased rapidly, and the demand for data mining has continued to increase. The method of big data analysis can more scientifically guide the exploration and development, oil refining and chemical industry, transportation, storage and marketing of the petroleum industry. This paper briefly describes the big data theory and the development status and problems of the petroleum industry, and puts forward an overall design architecture of big data platform in the petroleum industry, which is divided into five levels, including data source, acquisition technology, storage technology, processing and analysis technology, as well as big data applications in the upstream, midstream and downstream of the petroleum industry. This paper analyzes and summarizes the characteristics of big data in the petroleum industry, studies the common algorithms of supervised learning, unsupervised learning, semi-supervised learning and reinforcement learning in data mining technology, and finally gives six typical application scenarios of the petroleum industry chain.

At present, the analysis technology based on intelligent big data can realize the full collection of massive information, the in-depth mining of data relevance and intelligent analysis, and the assistance of upper-level application command decision-making. Many countries have evolved advanced intelligent data analysis technology as a strategic means. The status of "data-driven command and decision-making" is increasingly emerging. This paper constructs a user interest model based on hierarchical vector space, uses the TF-IDF classification algorithm to mine the user interest model, quickly and accurately obtains the required basic data, dynamic information and the latest intelligence, tracks the frontiers of science and technology, sorts out the development context, and grasps the key of the problem. It is the key to analyze the competitive landscape and development trend, battlefield simulation and planning analysis, to find breakthrough points, so as to occupy a dominant position in the war.

The main idea of the project is to utilize machine learning to try to make a diagnose of potential skin cancer. We used a traditional CNN model and later a ResNet Model to solve the problem. The model takes an image of the skin sample as input and tries to predict whether the abnormality on the surface of the skin is malignant or benign. The traditional CNN model resulted in an accuracy of 81.3% while the ResNet model resulted in an accuracy of 87.7%, which is a great leap due to the advancement in ResNet model compared to a traditional CNN model. However, we concluded that the accuracy of the model can still be increased by updating come factors of the model like the size of the dataset, the computing ability, and the structure of the model itself. Besides, the researchers implemented a GUI so that the users can utilize our research more conveniently.

The content of this simulation is the influence of rigidity and flexibility on the aerodynamic performance of airfoil. The research objects are drones and rotors. The focus is on rotors with different speeds of rigid rotors, different speeds of flexible rotors, and different elastic moduli (stiffness) of flexible rotors. Research on lift and torque.

Many existing methods have achieved the efficiency and speed of person re-identification deep neural models. However, due to the expensive cost of annotations, semi-supervised and unsupervised models have been treating seriously nowadays. The normal process of the unsupervised re-id model can be sum up into five steps and categorized into purely unsupervised learning person re-ID and unsupervised domain adaptation person re-ID. This paper will discuss some extant unsupervised re-id methods and address some existing problems about Person Re-identification. We also sort out the code of each paper mentioned and compare their performance on the existed dataset.

Object detection is an important research direction in the field of machine vision, which is widely used in industry, security and intelligent transportation. In the intelligent traffic gate detection, most of the visual detection is concentrated on face detection, which can not detect the behavior of people who follow or walk side by side and their height. In order to solve this problem, we proposed a visual detection algorithm based on Yolo V3 deep learning network model, and trained the network model for pedestrian detection; The tracking algorithm based on Yolo V3 sort adopts two-line detection, which can not only detect the following, side-by-side and other abnormal behaviors, but also effectively solve the problem of single line detection; At the same time, the target pedestrian is segmented by using Yolo V3 grabcut, and the segmented image is filtered and binarized. After traversing the pixels, the height detection is realized based on the optical axis aggregation model. The experimental results show that the improved network model realizes the detection of ticket evasion behavior such as tailing, side by side and height detection, improves the accuracy of pedestrian detection, and can meet the needs of intelligent transportation field.

In order to avoid too many Tibetan text errors from affecting the text quality, a method of Tibetan text error correction based on soft masked Bert is proposed. In this work, the definition and classification of Tibetan text error correction are pointed out, the embedding composed of word embedding, position embedding and paragraph embedding is used as the input of Tibetan text error detection model. The probability labels of characters in the text correspond to the output of the error detection model. The output of the model detects characters with high error probability.The final soft masked embedding result is obtained by the output of the weighted sum error detection model and the input embedding. The best model constructed by the encoder layer sets all soft masked embedding sequences as input, and the residual connection values are obtained by connecting the input embedding results and all hidden states of the encoder layer. The obtained residual connection values are input to the full connection layer, and the softmax function is used to obtain the probability that the output characters can be accurately corrected as candidate characters to complete Tibetan text error correction. Experimental results show that this method can achieve efficient Tibetan text error correction, and use the error probability of different candidate words to obtain the best error correction results.

The rapid development of deep learning has sparked a lot of interest in its application to medical-image processing issues. Cancer, as one of the most intractable diseases, needs high accurate and efficient adjuvant detection. Our study compared and developed a deep learning algorithm by using CNN models from multiple dimensions that can accurately detect breast cancer. One of our depthwise separable convolutional network methods for classifying breast cancer based on histopathologic scans of lymph node data attain excellent performance compared to the previous investigation to other applications. This method achieves 93.41% in test accuracy, 0.9568 in recall lastly. The remarkable recall performance shows the diagnosis is almost completely correct for benign tumors. However, there is still a small chance of misdiagnosis for malignant tumors so that no patient is spared. Such results are ideal for the adjuvant diagnosis of breast cancer. We also demonstrate that the depth of the network and the computation and number of parameters play critical roles. These findings demonstrate that deep learning approaches can be readily trained to achieve high accuracy and efficiency in breast cancer classification. They hold enormous promise for enhancing clinical diagnosis methods and being more portable.

Visual object tracking aims to capture the given target in the whole sequence, where the target will suffer challenges such as occlusion, motion blur, distractors, and scale variation. Several popular frameworks are proposed for improving accuracy and speed. This paper aims to review the existing tracking frameworks, such as correlation filter, Siamese network, and multi-domain network and their variants. We also give a brief comparison of the OTB benchmark. Our review can give an in-depth introduction to visual object tracking tasks and help beginners recognize this area efficiently.

B Investment into hazardous assets should always be made with caution, if not avoided completely by risk-averse groups, such as the older workforce. However, as society and technology advance, it becomes almost impossible for regulators to make effective enough laws to mitigate the risk of newly invented financial instruments, such as cryptocurrency. This paper analyzed how effective are modern data science techniques: supervised learning models, such as random forest, K Nearest Neighbors, decision tree, combined with bagging and stacking techniques, could be used to catch the notorious "pump and dump" activities in the Dogecoin market, which is the cryptocurrency that had a close to positive infinity return in May 2021 during the COVID-19 pandemic. This paper concluded that random forest algorithm, when trained with a five-folded cross-validation technique, could reach an out-of-sample testing accuracy of 100%. Furthermore, the F1 score of 0.84101, precision score of 0.94402, and recall score of 0.77700 could alleviate one's concern about overfitting. In conclusion, the model results suggest modern supervised learning techniques are quite effective in catching suspicious activities in modern financial instruments

Breast cancer has long been a life-threatening disease, and the application of digital pathological detection systems on breast cancer detection, which employs convolutional neural networks (CNN), is a milestone in the medical field. For medical testing, higher accuracy means a stronger possibility of saving ones’ life. However, the accuracy of the current study on mammography image classification is not satisfying enough. In this paper, we simplified and improved the current network structure's performance on classifying histopathological images by proposing a new Squeeze and Excitation network DenseNet based improved model (SE-DNI), which is a hierarchical multi-stage process that consists of six layers. We tested three different DenseNet models for feature extraction, and our improved model of DenseNet201 with attention layer (SENet) has the highest test score among all three models we tested, which achieved an accuracy of 99.82% compared with 96.15%, the accuracy of the network that uses a normal DenseNet201 model.

Breast cancer is a cancer which has the highest morbidity in women worldwide. Modern medicine uses imaging methods to analyze pathological images, which may cause the human error, and consume resources and manpower. Deep learning can help people and doctors to detect breast cancer, treat it better. This paper developed a breast cancer diagnosis system based on the Convolutional Neural Networks (CNN). We proposed a convolutional model consists of two convolutional layers and pooling layers. The Breast Cancer Histopathology Image Classification (BreaKHis) dataset was used to contain benign and malignant two classes. BCE loss was used as the loss function during our training process. The learning rate, optimizer, training epochs are 0.001, Adam and 11, respectively. In addition, we also developed a breast cancer diagnosis application using the Flask, which the users can use to detect breast cancer online and know where has cancerous by putting red boxes. Experimental results indicate that our cancer diagnosis system can provide good feedback to breast cancer patients.

Correlation filter-based target tracking is a simple and efficient method for visual tracking. However, the correlation filter tracker updates the target model with a fixed learning rate which cannot adapt to the change of target appearance for a real-time tracking, resulting in model drift and tracking failure. To deal with these issues, we develop an adaptive model updating method based on the peak-to-sidelobe ratio (PSR) of response map and the mean value of the frame difference. We adjust the learning rate according to different scenarios to improve the target tracking performance. Extensive experimental results conducted on OTB2013 and OTB2015 prove that the proposed FSAMF tracker superiority over others state-of-the-art trackers.

Data missing is one of the relatively difficult problems in the process of data mining,because the lack of part of the property value will seriously affect the results of data analysis. Most of the absence of missing value interpolation does not take into account the relationship between missing properties and other attributes in the sample. Aiming at this problem, this paper proposes a missing data based on KNN and random forest completion algorithm (RF-KNN) .Algorithm firstly chosen by KNN model of K value, makes it as building one of the parameters of the random forest. Then according to the characteristics of the random forest attribute, attribute classification prediction model was constructed, the missing attribute value for effective completion. Through model test and evaluation on the public data sets, the results show that compared with the traditional KNN algorithm and random forest, RF-KNN in more than one, this method is applicable to a variety of data missing value in the field of completion,which can effectively improve the efficiency of data mining analysis.

For enterprises, the ordering and transportation process of raw materials is a key link. Aiming at the problem of ordering and transportation of raw materials in enterprises, this paper establishes Topsis comprehensive evaluation model, entropy weight method, 0-1 integer programming model and so on, and uses Matlab, Lingo, Excel and so on to solve it. First of all, the collected data are classified according to A, B and C. Then four indicators are created and calculated: the cooperation rate between each supplier and the production enterprise, the demand satisfaction level of each supplier to the enterprise, the actual market share of each supplier to the same material, and the actual weekly distribution rate of each supplier. Then the Topsis comprehensive evaluation model is established, and the entropy method is used to modify the model, and the comprehensive weights are 0.0408, 0.0555, 0.4495 and 0.4542 respectively, thus the comprehensive score of each supplier is obtained, and the score is sorted, and the supplier data can be screened. The total supply of each supplier to each raw material within 240 weeks is calculated respectively, and the production capacity ratio of each raw material is obtained. Based on this, the rankings of three types of data accounting for the 50 most important suppliers are calculated. Then, with the minimum number of suppliers in the raw material ordering process as the objective function and the weekly production capacity standard as the constraint, a 0-1 integer programming model is established to get at least the number of suppliers. Then the goal programming is carried out around the lowest raw material procurement cost, and the demand of all kinds of raw materials is taken as the constraint, a linear programming model is established, and the results of the ordering plan for the next 24 weeks are shown in Annex A. Taking the minimum value of transport loss as the objective function, a linear programming model is established to solve the transport volume of all kinds of raw materials, and the transfer scheme is obtained. The linear programming model is modified by the loss capacity, and the dynamic programming model is established to get the optimal scheme.

In today’s most researches on autonomous driving, similar to industrial robot systems relying on sensor information to avoid obstacles, there is no prior knowledge to support logical reasoning analysis of their environment, which will make autonomous vehicles not receive enough information on traffic scenes and participate in traffic. Insufficient description of the contextual relationship between participants may lead to unreasonable decision-making in obstacle avoidance behavior and cause traffic accidents. Therefore, it is necessary to explore a safe driving behavior decision-making method from the perspective of knowledge reasoning for existing autonomous vehicles. This paper proposes an ontology-based self-driving car scene modeling and driving environment relationship inference method in an urban environment. By establishing a structured ontology model that considers the semantic relevance of scene elements, the driving scene is modeled. The knowledge-based reasoning decision model imitates the behavioral decision-making process of a human driver by the mapping relationship of "scene feature-driving action". This type of model stores driving knowledge. The driving knowledge here is mainly manifested from rules, cases to driving actions. The mapping relationship is used to establish a knowledge base containing driving knowledge and experience to provide behavioral decision-making suggestions for the future decision-making system of the vehicle.

Over the years, cancer has been attracting great attention worldwide due to its widespread and lethality. People have been trying to predict cancer using the machine in the hope of early diagnosis and treatment. Experts and scientists have been examining the performance of both basic machine learning techniques and some advanced neural networks on cancer prediction. In this work, we used convolutional neural networks (CNN) to make predictions based on images from histopathologic scans, known for their specialty in image recognition. We chose Visual Geometry Group 16 (VGG16) as our model, and we ran two experiments, one using pretrained parameters and one without. For both models, we ran a total of 10 epochs and recorded the accuracy, precision, and recall on both the training data and the testing data. According to the experiment results, the model trained from scratch had a slightly higher accuracy, with the highest accuracy being 89.01%. In addition, we designed a user interface so that people can access it more conveniently. Users can upload their scanned images, and the model will make a prediction and inform the users. In addition, we also designed a Natural Language Processing (NLP) chatbot to guide the users

The NBA (national basketball association) is one of the most well-known sports associations worldwide. Although under the influence of the epidemic, there are still 12.52 million people watching the sixth game of the final, and this number is only for the United States. Also, apart from the fascinating games, what is more interesting is the tremendous amount of data generated by the hundreds of games played per season. Lots of analyses and predictions using the data could be done in various ways. What we did in this essay was to make predictions that may help people win money in betting for MVP per season. More than thirty years of NBA games with fifty columns of features were used for analysis. Then four models for prediction were built. The result turned out to be 67% in winning a bet on MVP, which is quite good.

As the first driving force leading the development, technological innovation has become the strategic support of the modern economic system. Basic research and applied research have become the source of scientific and technological innovation activities, which determine the stamina of later product development and industrial upgrading. Increasing investment in R&D human and capital is an inevitable requirement for building an innovative country. The efficiency of R&D investment has become an important indicator to measure the high-quality economic development driven by scientific and technological innovation activities in innovative countries. The article is based on the perspective of innovation value chain, and then uses the DEA model to measure the efficiency of R&D investment in 31 provinces across the country, and divide the efficiency of R&D investment in various regions of our country into four types. It also gives policy recommendations based on the characteristics of the efficiency of R&D investment in each region.

The formation of voids during processing of glass fiber reinforced polymer may have a significant effect on the mechanical properties. In this paper, CFD simulation has been used for modelling of resin vacuum infusion process of fiberglass reinforced plastic pultruded sheet. The simulation results show that the velocity of resin flow decreases gradually as the infusion process continues, and reaches stability finally. The pressure gradient in the horizontal flow channel is an order of magnitude larger than the normal channel. The quality of the resin infusion process is closely related to the velocity of resin flow. The regions where the infusion defect occurs always have a lowest resin flow rate. The pressure gradient field in resin flows involves the infusion defect which usually appears in the area with a lower pressure gradient. This article research will provide references for the resin infusion process of bulk composite with fiberglass reinforced pultruded sheet.

As one of the core components of drilling and tapping machine, the spindle box unit of drilling and tapping machine is connected with the spindle unit of drilling and tapping machine on the bearing and fixed with the column unit of drilling and tapping machine through linear guide rail. Its stress and deformation have an important impact on the geometric accuracy and service life of the whole drilling and tapping machine. In this paper, the spindle box element of i5M1.4 drilling machine is studied. The maximum stress and strain of the spindle box are obtained from the static finite element analysis, and the first six natural frequencies and vibration modes of the spindle box are obtained from the modal analysis.

Under the condition that the main geometric parameters of pump remained unchanged, the optimization of pump performance was realized by reducing the effective flow area ratio of impeller outlet and inlet. The effective area ratio of impeller outlet and inlet was changed by means of quasi orthogonal method. With super-low specific speed centrifugal pump M23-12.5 as the research object, on the basis of the pump test results, the paper constructed 7 different impellers which were different effective flow area ratios and were loaded into the same complete casing geometric model including the front and back of the pump cavity and the seal rings. With CFD method, this paper analyzed the effect of different effective flow area ratios of impeller outlet and inlet on the performance and the distribution of flow field structure of super-low specific speed centrifugal pump. The results showed that with the impeller effective flow area ratio decreased, head and shaft power had a decline in varying degrees at the same time efficiency increased. Thus on the premise of improving efficiency, area ratio should be reduced appropriately; with the further narrowing of the impeller flow passages the internal axial vortex flow and the hydraulic loss became smaller; because the absolute speed of the impeller exit increasing after the passages narrowing, vortex near the tip clearance of impeller was enhanced and hydraulic loss was increased.

The three Hertz contact stress states of ball and raceway of pitch bearing are analyzed by sub-model technology. and the stress intensity factors of raceway sub-surface crack front are calculated according to the state of maximum Hertz contact stress appearing at the edge of raceway. The stress intensity factors are solved by interaction integration method, and the change of stress intensity factors in short distance pure rolling process is analyzed. At the same time, the influence of crack depth and crack inclination angle on stress intensity factor is analyzed. The results show that the subsurface cracks of raceway are mainly sliding and tearing failure, and the possibility of crack propagation is the largest when the crack is at the edge of Hertz contact pressure. The closer to the crack front of the free surface, the greater the value of stress intensity factor KII, the more obvious the sliding failure. With the increase of crack depth, crack propagation becomes more difficult. When the inclination angle of the cracks is small, the crack propagation is mainly opening mode, and with the increase of the angle, it gradually changes to the composite expansion of the sliding mode and tearing mode.

Using the similarity theory of centrifugal pump and the law of impeller cutting, the hydraulic components are geometrically enlarged, the capacity of the impeller and other hydraulic components of the existing centrifugal pump in the system is expanded, and the flow at the rated point is increased while keeping the head unchanged. At the same time, the installation scale of the pump unit is guaranteed to remain unchanged, so as to achieve the effect of low-cost capacity expansion. Through practice, the feasibility and economy of this method are verified.

This research proposes a method to fabricate gratings cross-scale by relying on the self-rebound properties of the flexible substrate. Compared with the traditional grating manufacturing method, the method has simple and high manufacturing efficiency, and can improve the electron beam lithography (EBL) process. First, an EBL is used to fabricate a grating structure with a period of 1μm on a silicon wafer, and the grating structure is replicated onto a polydimethylsiloxane (PDMS) film. Then, the film with the grating structure is pasted on the stretched PDMS substrate to form a whole. Finally, the flexible substrate is released to obtain a grating structure with a period of less than 1μm. In this way, the grating period can be descended from the micron level to the sub-micron level, and the grating period can be controlled according to the magnitude of the pulling force, and it has the potential to use a grating master with one micron level period to manufacture multiple sub-micron gratings. The experimental results obtained a grating structure with a period of 800nm and a grating area of 5mm2 . The experimental results mainly verify the effectiveness of the proposed use of the self-rebound properties of flexible substrates to manufacture cross-scale gratings

The influence of parameters on load distribution between teeth of HCR involute spur gear was studied. The load distribution was studied and verified theoretically by test. Under the condition of given parameter, the change curve of load distribution and maximum load sharing rate were studied when addendum coefficient, pressure angle, tooth number, modification coefficient and base pitch error respectively changed. The test result show that, under high loads, the maximum load sharing rate are linearly correlation to addendum coefficient, pressure angle, modification coefficient and base pitch error. With the increase of the number of teeth, the change rate of the maximum load sharing rate decreases from large to small, then tends to be flat. The rate of change in load sharing at the point of tooth engagement is small when the top height co-efficient, pressure angle, number of teeth and coefficient of variation change, and relatively large when the base joint deviation changes.

An automatic method based on CATIA secondary development is used to design the fixture of the brick-type skin antenna. The fixture is destructed into typical frames. The Python language is used to read data in EXCEL and make the CATIA module to create parts of fixture of the brick-type skin antenna in batch. The installing holes and curved surfaces of the fixture are consistent with the shaking table and antenna. The first-order modal frequency of the automatically modelled fixture is larger than 1000 Hz. The fixture can successfully be as a connection between the machine and the shaking table with high stiffness. The automatic design of the fixture of the brick-type skin antenna is realized by this method. It also has strong practicality and expansibility for other similar machines.

According to the manufacturing characteristics of diamond rollers, the key technologies of diamond rollers' feature information extraction, feature machining scheme decision-making and NUMERICAL control program post-processing were studied. The latest product model data exchange standard STEP AP 242 was adopted to realize the extraction of diamond rollers' manufacturing feature information. The 3d model of diamond roller is used as the carrier of manufacturing information by integrating product model data exchange standard and product data management. The integrated system of CAD/CAPP/CAM/PDM for diamond rollers is developed on SolidWorks and Aras Innovator platform, and is verified by examples.

Aiming at the problems of poor loading accuracy and slow response speed of the permanent magnet synchronous motor as a load simulation system, a hybrid flux observation control algorithm is proposed to improve the loading accuracy and speed up the response speed. Hybrid flux observation control algorithm calculates stator flux by using voltage current hybrid observation to complete SVPWM direct torque control; Direct torque control takes torque and stator flux as control variables, but it is difficult to measure torque and stator flux directly in practical application, so it is very important to estimate torque and stator flux accurately. The proposed hybrid model flux observation algorithm can estimate the torque and flux. The simulation results of the algorithm are in agreement with the theoretical analysis, which proves the feasibility of the control strategy. In addition, the designed hybrid flux observation load simulation system can realize the simulation of various load types, such as constant torque load, constant power load and fan pump load. Finally, the experimental platform of load simulation system is built and the experimental results show that the system can accurately estimate the torque and flux.

The photoelectric theodolite is an important device for photoelectric tracking measurement, and has the functions of real-time monitoring and automatic tracking. With the continuous improvement of the optical theodolite tracking measurement requirements, various traditional training and testing methods have appeared various deficiencies. This paper takes the photoelectric theodolite servo tracking system as the research object, and based on the momentum moment theorem and Euler's dynamic equations, a non-linear two-axis coupled dynamics model of the tracking frame in the system is derived. The electromechanical dynamics model describing the dynamic characteristics of the controlled object in the tracking servo system was used. The correctness of the theoretical model was verified by comparison test method and computer simulation method. For the follow-up research, the design of high-precision tracking controller and control parameter selection. And optimization provides a theoretical reference and model basis. The experiments in this article show that the optimized photoelectric theodolite car model has fully met the requirements in terms of vibration intensity, and the maximum vibration intensity can be controlled to reach less than 80% of the technical requirements of the photoelectric theodolite (≤0.5g), indicating the addition of a vibration isolation device . The ride comfort of the photoelectric theodolite has been significantly improved.

There are many factors affecting the green development of enterprises by environmental tax law, but traditional economists are limited by the development of technology and can only analyze the relationship between them by establishing an ordinary econometric model. The development of computer technology makes it possible to analyze this problem in depth. Dr. Li Xiangju established a spatial econometric model by using computer, brought more influencing factors into the scope of the model, and reached a widely recognized conclusion through computer big data analysis. By analyzing Dr. Li Xiangju's computer modeling research, this paper deeply discusses the scientificity and theory of computer modeling research. It provides theoretical support for the wide application of computer modeling.

The traditional identification method of watered-out layer is mainly based on logging interpretation, the evaluate model is established according to the change of logging curve of watered-out layer. but the error is usually very large. Based on ensemble learning method, this paper selects random forests algorithm to establish classifiers of different reservoir watered-out levels, and it verifies with actual oilfield data. The results show that the accuracy of this method reaches 96.2%, and it can be effectively used to identify watered-out layers.

This paper summarizes the research results on defect segmentation and recognition by researchers at home and abroad. Through analysis, it is found that deep learning has an absolute advantage in seam defect segmentation based on X-ray. In the field of weld defect identification, the application of intelligent algorithm has become the mainstream of research.

Cancer detection has become much more significant in recent times than it was before since the number of patients suffering from cancer is rising year by year. But the actual case is that not all cancerous symptoms can be detected even by those experienced physicians by looking at their patients’ histopathologic images through their eyes. To improve this situation, we proposed a model based on convolutional neural networks to help classify the histopathologic images into 2 categories: benign and malignant. Our model, in this way, can extract main features from a certain image and learn from these specific features for classification. We then evaluate its performance on an online public dataset consisting of over two hundred thousand different training images and over fifty thousand validating images. Our model returns a relatively high accuracy on the test data which the dataset provides, and it does outperform some of the existing models in making the binary classification. In addition, we also designed a user interface to put our model into practice. With the help of this interface, users can simply upload an image from their local directory for our pre-trained model to process.

Aiming at the real-time 3D reconstruction of large-format components, this paper proposes a method for extracting the centerline of laser stripes against ambient light interference. The Zhang-Suen thinning algorithm is optimized to effectively retain the data information of the laser fringe endpoints. In the process of fitting the centerline of the la-ser fringe with the least square method, a method of extracting the region of interest is proposed. Through the anal-ysis of experimental data, this method effectively improves the anti-interference ability of the algorithm, is suitable for more scanning occasions, and reduces the conditional limitation of using laser three-dimensional scanning in the industrial measurement process.

Aiming at the problem of salient target detection in complex environment, this paper proposes a method of video image personnel detection. In this method, the improved dark channel priority processing strategy is used for image denoising and enhancement. Based on this, the model of antagonistic network is constructed by using edge constraint, which solves the problem of fuzzy edge and different prediction scores in the target region. Experimental results show that the average detection accuracy is higher than 98% and the average recall is higher than 94.12%.

Object detection, a significant and challenging issue in computer vision with various branches, receives great attention continuously. This paper reviews classic methods of both traditional and deep learning based detectors. Based on overall architecture and anchor, deep learning based detectors can be classified more precisely, including recent improvements. Common datasets and evaluation metrics for accuracy and speed are introduced as well. Besides, future directions, research hotspots and applications are discussed. Novel and captivated tasks such as 3D object detection, camouflaged object detection, rotated object detection are included. Anchor-free detectors are likely to be the mainstream of future object detection methods. Researches are focusing more on sophisticated situations such as small objects, occluded targets and dense distribution.

The human body pose estimation mainly locates the position of the key points of the human body in the image. The stacked hourglass network uses top-down and bottom-up feature extraction methods to obtain better results in the task of human pose estimation. However, in the process of feature extraction, the resolution of the image will be lost, and it will have a greater impact on the positioning of the key points of the human body. Therefore, this paper incorporates a polarized self-attention mechanism into the stacked hourglass network. A polarized self-attention (PSA) module is added before the second convolution of the basic residual block and added before the max pool down-sampling and the nearest neighbor up-sampling for each stage of the hourglass module. The space and channel of the PSA are used to maintain a high feature resolution, thereby improving the accuracy of the model's positioning of the key points of the human body posture. Finally, experiments on the human body pose estimation data set (MPII) show that the improved network PCKh@0.5 reaches 92.6%, which is 1.5% higher than the original model, which further illustrates the correctness and effectiveness of the network.

The deep neural network (DNN) as the computing core of the multiplication operation consumes a lot of hardware resources, which is not conducive to the development of DNN on a hardware platform with limited resources. In order to solve the above problems, this paper proposes an acceleration method of basing on a deep neural network (DNN) to accelerate multiplication operations on resource-constrained hardware platforms. This method can support sparse calculations to optimize calculation delays. At the same time, inspired by neurosynaptic plasticity and stochastic computing (SC), an acceleration method using simple logic gates for reasoning tasks is proposed. Experimental results show that the hardware resource consumption of the proposed acceleration method is 1.4 times lower than that of the traditional method solution in DNN.

Target plate method is an important method to detect the scattering characteristics of fragment warhead，In view of the low efficiency and unified judgment standard that manual work detects the number of fragment perforations and pits on the target plate. An image processing method based on hyperboloid threshold segmentation is proposed. Firstly, the geometric distortion correction of the target plate is realized with the homography matrix. The target plate is segmented with straight line fitting using the boundary straightness and the target vertex, which is smoothed by guided filtering. Secondly, according to the illumination change with different positions on the target plate of the flashlight, the image gradient method is used to generate hyperbolic threshold to segment the fragment perforations and pits. At last, the region growth based on gray similarity seed points is used to detect the fragment perforations and pits. The method can quickly and accurately segment the fragment perforations and pits region, and provide the basis for objective and accurate evaluation of warhead power.

Aiming at the visibility prediction problem that is currently concerned in the transportation and aviation fields, an effective detection method based on complex network theory is proposed. Firstly, preprocess the images in the FROSI data set to obtain the gray image after the background difference subtraction, and derive and calculate the transmittance. Secondly, a corresponding mathematical model based on a complex network is proposed. After three steps of gray contour extraction, recognition parameter extraction, and image recognition and classification, a simulation platform is built and canny operator is used to complete image edge feature contour extraction. Finally, the processed image is combined with the formula to detect the visibility change trend in different time periods and compare it with the true value for comparison, the corresponding evaluation results are obtained from the relative error analysis.

Pulmonary lung nodules, whether malignant or benign, often represent early symptoms of lung cancer. However, it is challenging to detect a pulmonary nodule by computed tomography before they become mass for helping people have timely treatment. In this paper, the idea we theorized is to scan the images to find their own pulmonary nodules, diagnose whether they are malignant or not, and report patients’ conditions. More specifically, we created a structure on top of the pre-build C3D structure. It achieves a deep convolution model with 6 layers and trains that by utilizing images from Lung Nodule Analysis 2016, which is the subset of Lung Image Database Consortium image collection (a diagnostic and lung cancer screening thoracic computed tomography scans datasets). These images are segmented into multiple pieces in the direction of the z-axis and pre-process with a more direct connection to the annotations from the big CT scans. After the diagnosis, it will be broadcasted in the form of long sentences and its voice implemented by Natural Language Processing (NLP). Experimental results illustrate that this CNN model has high robustness, which deserves confidence from people with potential pulmonary lung cancer.

Due to the severe damage to the health of a human being caused by breast cancer, it is rather crucial to explore novel approaches to detect images of lymph node sections to replace traditional methods, which are both time-consuming and inaccurate. This paper proposed an efficient cancer detection method by applying various kinds of convolutional neural networks (CNN) with a user interface. To achieve the best accuracy, experiments and comparisons are made between VGG16, ResNet-18, and EfficientNet, which are three popular models in the field of image processing. We also made a comparison to emphasize the importance of pre-trained weights of VGG16. Our models achieved high accuracy with pre-trained weights. In particular, VGG16 achieved 0.8340 in accuracy and 0.8326 in recall on the PatchCamelyon dataset and was used in our application. Our source code is available at https://github.com/bqdqj/Cancer-detection-based-on-tensorflow-and-PyQt5.

Skin cancer has been one of the most common forms of cancer around the world. Due to the efficiency and high accuracy of artificial intelligence, more and more hospitals are using it to assist doctors in finding out cancer quickly. In our work, we trained a convolutional neural network with a dataset called ‘Skin Cancer ISIC’ to detect appearances of 9 different kinds of skin cancers. Firstly, we trained a convolutional neural network model with the original data from the dataset. It contains several convolutional and max pooling layers, and its accuracy achieved 45%. Although the classification of 9 species will undoubtedly decrease the accuracy of the model compared to making fewer species, 45% is not an acceptable value for medical judgment. To improve the accuracy of our model, we used data enhancement and retrained our model. During the training of the dataset, we find out that the pictures of the dataset are not in the same standard. The brightness, contrast, size, and shape of the images are different, which increased the difficulty of learning. By rotating pictures, adjust brightness, equalizing histogram, adding random noise, flip and adding random USM, we reach an accuracy of 60%. Moreover, we employed ResNet-50, a pre-training model, as the convolutional neural network model to further improve our accuracy. The final testing result gets an accuracy over 65 %, which is a huge improvement from the beginning.

After a few cases of COVID-19 had emerged in Wuhan, China, the disease had rapidly spread all over the world. As the World Health Organization confirmed that COVID-19 is an infective and dangerous respiratory virus, it is important to wear masks to reduce the rate of infections. Thus, a face mask detector is essential to encourage people to wear masks and prevent the virus from spreading. In this paper, we proposed a mask detection model based on transfer learning of MobileNet. This face mask detector can be applied to images, videos, or live video streaming to detect whether the people have worn a mask or not. The model has achieved an accuracy of 99.88% on the training set and 99.64% on the testing set. In addition, we used the K-Fold Cross Validation to evaluate our model, and the average accuracy was 97.4%.

Being part of hematopoietic lymphoid tissue tumors, lymphoma has aroused wide concerns in the field of cancer research. Therefore, a lymphatic cancer diagnosis has become a popular question in medical image processing, with numerous works are proposed. Recent works focus on how to get the appropriate image for a better cancer diagnosis. However, the question of how to get a higher accuracy just by learning from the dataset of the images of lymphatic node sections has not been paid attention to. This paper aims to solve this problem by applying a convolutional neural network (CNN), which has nine convolution layers, four dropout layers, three maxpooling layers, two dense layers, and one flatten layer in total. Experiments on the PatchCamelyon dataset show that our method can handle the Lymphoma cancer diagnosis problem with an accuracy of 93.15%, much higher than that of the LeNet and AlexNet, which are 83.17% and 84.49%, respectively. With such high accuracy, our model has the potential of acting as a reliable assistant for doctors when diagnosing lymphoma cancers.

With the advent of the information age, the development of science and technology has reached an unprecedented speed, especially computer science, and machine learning has been a hot topic among scientists and researchers. Stock market prediction lies in this area that helps investors understand what stocks they should purchase or when to purchase. More importantly, they can earn money by successfully predicting the trends of the stock market. In this work, we use relevant theoretical knowledge of machine learning and neural networks and set up neural network models in Python's programming language. At first, we collect stock prices data of Apple and Tesla from 2020 to 2021. Later, we compute them by using the models we build to analyze and predict the trend of stock prices from these two companies. At last, we find out that our models work nicely before day 340. However, the models fail to predict the trends of stock prices of these two companies starting from day 340 to day 360.

Histopathology is the primary tool employed in breast cancer diagnosis. It involves examining metastatic tissues of lymph nodes under a microscope. Histopathologists are responsible for making tissue diagnoses, and the process is challenging and tedious. To diminish their workload and allocate more time to efficiently maintain patients' care, deploying an intelligent system to support the diagnosis is reasonable. Therefore, we introduced a new breast cancer diagnosis application based on deep learning technology in this paper. The application's foremost objectives were to handle the vast dataset of digital pathology scans and train deep residual networks to classify small patches from the sizable whole slide images with higher accuracy. Experimental outcomes indicated that our model could achieve 97.3%. Another noteworthy feature was a FAQ chatbot that we implemented for patient consulting.

Infrared target tracking is a fundamental task and plays an important role in military, weapon guidance and other fields because of the superiority of infrared imaging system. However, it is a challenging task due to lower resolution, smaller size and less texture information of infrared targets. Although varieties of methods demonstrate promising performance on visual target tracking, they cannot directly perform well on infrared images. To address this issue, we propose an infrared target tracker based on siamese network, named InfTrans, consisting of three components: a feature extraction backbone, an encoder-decoder transformer and a prediction head. In the InfTrans, transformer architecture is employed to exploit spatial and temporal information effectively. Firstly, considering the superior ability in capturing long-range dependencies, encoder-decoder transformer is connected after backbone to capture relationship between template and search patch along spatial and temporal dimension. Secondly, considering the parallel computing power of transformer, InfTrans receives search patch sequences containing multiple patches as input of search branch and processes multiple frames in parallel. Finally, a prediction head is employed to output bounding box sequences. InfTrans views infrared target tracking as a parallel bounding box prediction problem. Extensive experiments show that the proposed tracker achieves promising performance on public infrared dataset.

In order to improve the performance of infrared small target detection under complex background and noise interference, a single-stage infrared small target detection algorithm combining deep neural networks and self-attention mechanism is proposed. First, a lightweight fully convolutional neural network based on the encoder-decoder architecture is designed to segment infrared images to achieve background suppression and target enhancement. Then, A self-attentional mechanism is introduced to further suppress false alarms. Finally, an adaptive threshold method is used to separate small targets. As for the network structure adopted here, multiple down-pooling layers are used to reduce. The simulation results show that the algorithm is superior to the typical infrared small target detection algorithm in terms of detection rate, false alarm rate and operation time. It is suitable for infrared small target detection under complex background.

The target detection capability of radar has always been one of the key contents of radar research. The rapid development of radar and other fields in recent years has made it difficult for the FFT algorithm to meet the real-time requirements of data processing. Sparse Fast Fourier Transform (SFFT) is a feature based on sparse signal frequency domain. For some sparse signals, SFFT algorithm has shorter operation time and faster speed than FFT algorithm. This paper uses the constant false alarm algorithm for target detection, and compares and analyzes the time complexity and signal processing efficiency of the SFFT algorithm and the traditional MTD method. The analysis of the detection results shows that the algorithm in this paper has better detection performance and higher computing efficiency.

Coal mills are important equipment of thermal power units work in hostile environment, so the health working condition of coal mills is a key factor for the normal operation of power plants. To monitor the operation of coal mills and make early warning of fault diagnosis, it has to collect the vibration/sound signals of coal mill operation and pre-process the sound samples to extract Mel-frequency cepstrum coefficients (MFCC) as features, and use the set of MFCC to train Gaussian mixture models (GMM). Finally, it uses the trained GMM to test the operation of coal mills. The test results demonstrate that audio anomaly detection based on MFCC and GMM can be used to effectively identify the on/off status of coal mills.

To improve the detection accuracy of the radar security system and reduce the probability of false alarms, a Range-Doppler multi-target detection method based on classical spectrum estimation is proposed to solve the problem that the triangular modulation waveform sent by FMCW radar produces false targets in multi-target detection.In this method, the echo signal is constructed to generate a signal matrix by the difference value filling zero,the range Doppler image is drawn, and the target information is obtained by processing the image, which effectively avoids the interference of false targets.Combined with the security early warning strategy of the security system, the software design of the system is completed.The simulation results show that under the interference of Gaussian white noise, the Range-Doppler multiple target detection method based on classical spectrum estimation has higher accuracy and faster speed under the same conditions, which can improve the reliability of the security system.

Fault detection and diagnosis of Rolling mill have a very significant role in the practical application. Extracting the bearing fault signal from a multicomponent signal mixture using a variety of signal de-noising method is the first problem to diagnose the rolling mill bearing. Rolling mill bearing belongs to the bearing of low speed and heavy loading. Diagnosis is difficult because of the interference signal and the complex fault mechanism. Wavelet packet de-noising method was applied to extract the fault characteristics of vibration signals. The analysis results show that the wavelet packet analysis method can effectively remove mixed in the interference signal of the rolling mill rolling bearing signal. The characteristics of rolling mill rolling bearing fault signal become more obvious, and the fault diagnosis result becomes more accurate.

Geometric structure feature is the most intuitive feature of ship in SAR image, and it’s important basis for ship Automatic Target Recognition in SAR image. Minimum Bounding Rectangle (MBR) extraction is the basis of obtaining ship geometric structure feature in SAR images. Generally, there is an angle between ship target and SAR image azimuth, so ship orientation estimation is the premise of exactly extracting the MBR. In this paper, the Target Pixel Statistical Method is used to extract the ship's principal axis is proposed, the accurate orientation of the ship target is determined by using the optimal correlation coefficient , and the ship slice perpendicular to the azimuth is obtained after image rotation. Then, the target length and width threshold coefficients are set to determine the MBR of the ship. The experimental results show that for higher length commercial ships, this method has high accuracy and good stability.

Aimed at improving the detecting ability of spaceborne microwave radar system on the week target, a window function should be added when processing the LFM signals at the cost of SNR loss and mainlobe spreading. To solve this problem, transmitting NLFM signal in the radar signal generator is a better solution. In this paper, we propose a spaceborne NLFM signal design principle, and give the simulation result. And then the hardware implementation using Xilinx Zynq Ultrascale+RFSoC device is introduced. Finally, the test result is given with the Chebyshev window, which the PSLR result is better than -46dB.

In order to study the corrosion phenomenon of automotive metals in the actual environment and different test methods, Zinc materials is selected as the research object. The corrosion phenomenon of Zinc plate in atmospheric environment, NSS neutral salt spray test and CCT cycle corrosion test are analyzed through the dimensions of corrosion morphology, weight loss and electrochemical corrosion analysis, and the corrosion fitting matrix of zinc material is obtained. The material corrosion characteristics under different corrosion atmosphere conditions are quantitatively analyzed, which provides a reference for exploring automobile intelligent evaluation methods, shortening corrosion test cycle and accelerating corrosion verification progress.

Finite element software is used to check the strength of the rocket sled during recovery. In this paper, the finite element method is used to model the double track sled and the water brake. The water brake is loaded and the constraints, contacts and connections are defined. The water resistance of the water brake is loaded when the double track sled is recovered by using the fluid mechanics simulation. At last, the static analysis is used to analyze the force on the key parts of the double-track sled, and the strength of the whole sled is checked. At the moment of entering the water, the stress concentration occurs at the connection part of the sled exit with the body beam and the connection part of the support seat with the body beam, so the connection strength should be strengthened. This method provides technical support for body design, water brake design and sled recovery application.

In the processing of TC4 titanium alloy profile, the "peeling" defect appeared at the punching position, which led to the product being scrapped. For the reason of positioning quality problem, the failure analysis was carried out on the punching area of titanium alloy profile. Through metallographic observation and microstructure analysis, it was found that the "peeling" phenomenon in the punching area was finally located in the internal delamination defect of titanium alloy profile, which was caused by the folding during the hot deformation of profile and the extension during the further extrusion deformation. When the hole is punched, the defect is deformed by the machining force, forming a macroscopic "peeling" phenomenon.

Non-contact temperature measurement is one of the more effective methods for measuring high-temperature targets. As the front end of the entire temperature measurement system, the optical path system is responsible for receiving infrared light. The ability of the optical system to receive infrared light directly affects the accuracy of the final temperature measurement result. It is necessary to conduct an experimental analysis on the optical path system to improve the measurement accuracy of the smelting furnace's online infrared continuous temperature measurement system, which has a very important role.

Aiming PHS 2GPa material and its B-pillar products at object, through the accelerated corrosion procedure, firstly, weight loss and corrosion data are collected to analyze the corrosion rate of metal materials and the compatible performance of the substrate/layer; secondly, and tensile strength and elongation are collected at break, to analyze the residual mechanical properties of the steel matrix; finally, combined with the weight loss and mechanical properties data, the migration behavior of the corrosion mechanism under different corrosion cycles is analyzed. The results show that pitting corrosion occurs at the defect position of the oxide layer on the metal surface, and pitting corrosion can increase the material difference. When the weight loss is 0.92%, the mechanical properties are reduced by about 10.46%. Uniform corrosion can weaken the effect of pitting or perforation on the body, and the mechanical properties of materials are stable. The elongation at break of most samples is <10%, which meets the performance requirements of high-strength steel for vehicles.

in this paper, the simulation of gas-liquid-solid three-phase flow is carried out by numerical simulation, and the flow velocity, pressure and particle trajectory distribution are obtained. The influence of gas content, sand content and sand shape on multiphase flow is analyzed. The results show that gas content has the greatest influence on multiphase flow. With the increase of the air content in the wellbore, the average velocity and average pressure on the wellhead surface decrease. The simulation results change little with the change of sand content and sand shape.

As an important wet tree species in the construction of wetland ecosystem, Taxodium hybrid ‘zhongshanshan’ has been widely planted in the lakeside of Dianchi Lake. It is an important link in the water governance of Dianchi Lake. This article uses the average biomass method to catch Laoyuhe river wetland zhongshan carbon ability to analyze and estimate the value of carbon sequestration oxygen release zhongshanshan’s carbon capacity and build growth model. There was no significant difference of zhongshan sequoias carbon ability. The carbon sequestration capacity of zhongshanshan decreased. The carbon storage of ‘zhongshanshan’ in Laoyuhe wetland and Dianchi Lake is 2439 t and 122400 t, respectively. The value is 5.83 million yuan and 293 million yuan respectively. In 2017, the contribution rate of carbon sequestration and oxygen release value in the growing season was 19.5%.

To study the energy-saving effects of different enclosure structure, typical single residential buildings in the Wenzhou area selected as the research objects, and simulation tests were carried out in the same environment and the same period and analyzed the simulation results. The results show that in terms of walls, the best energy-saving effect type for the internal surface temperature of the wall is ceramsite concrete composite blocks, and the best energy-saving effect type for indoor temperature is non-clay sintered thermal insulation blocks, and the best energy-saving effect for indoor and outdoor temperature differences is foam concrete block; in terms of roof, the best energy-saving effect type of plant height is plant height 0.1m and plant height 0.5m; the best energy-saving effect type of leaf area index is leaf area index 0.3 and leaf area index 1.0; the best energy-saving effect type for leaf reflectance is leaf reflectance 0.85; the best energy-saving effect type for roughness is smooth and very rough; the best energy-saving effect type for thickness is 0.1m and 0.051m; the best energy-saving effect type of electrical conductivity is soil electrical conductivity 0.7 and soil electrical conductivity 0.4; the best energy-saving effect type of soil specific heat is soil specific heat 501J/(kg·℃) and soil specific heat 1500J/(kg·℃); the best energy-saving effect type of soil density is 2000kg/m3 of soil density

According to the results of clayey water electric conductivity (EC), foraminiferous abundance and grain size analysis in the borehole CZ85 from the west coast plain of Bohai Bay, combined with the AMS14C dates, the sedimentary strata of borehole CZ85 is detailed divided into nine units from bottom to top: Unit I (40.0–33.0 m), Delta plain sediment affected by seawater; Unit II (33.0–29.4 m), lagoon sediment; Unit III (29.4–22.2 m), sediment in the intertidal zone; Unit IV(22.2–19.2 m), Delta frontal sediment; Unit V (19.2–15.8 m), Delta plain sediment; Unit VI (15.8–9.0 m), watercourse-flood plain sediment; Unit VII (9.0–8.0 m), salt marsh sediment; Unit VIII (8.0–3.6 m), tidal creek sediment; Unit IX (3.6–0.5 m), flood plain sediment. This study shows that the location of borehole CZ85 was affected by seawater twice since the Late Pleistocene：(1) The lagoon sediment from Unit II and sediment in the intertidal zone from Unit III. (2) Salt marsh sediment from Unit VII formed during 7.01-6.72 ka cal BP, which was affected by seawater, caused by the maximum period of the Holocene MIS1 transgression in the western plain of the Bohai Bay. The grain size analysis can identify the hydrodynamic events in sediment accurately, combing with clayey water electric conductivity (EC) and abundance of foraminifera is an effective method to differentiate the sedimentary facies, that compensate the lack of credibility for using a single index.

Because disk friction loss of ultra-low specific speed centrifugal pump accounts for a large proportion of shaft power, it is important for accurate calculation of the loss. According to the structure of the pump and the characters of the flow of the fluid in the pump cavity, a new disk friction loss formula is put forward, which considers to add two important factors that include the leakage influence coefficient and geometric size of pump cavity. By means of CFD method, according to the results of numerical calculation prediction and test performance curve, the following conclusions are obtained by comparison of accuracy among the new formula and two existing disk friction loss formulas: the new formula is more suitable for calculation of ultra-low specific speed centrifugal pump disk friction loss compared with existing formulas.