This PDF file contains the front matter associated with SPIE Proceedings Volume 12305 including the Title Page, Copyright information, and Table of Contents.

Aiming at the outstanding problems of long on-site debugging time and high risk of operation and parameter adjustment of industrial control systems in important production and processing bases of oil and gas stations, such as joint stations, transfer stations, and water injection stations, this paper adopts digital twin technology, rapid physical access technology for equipment, PLC and Configuration technology develop the simulation and test platform of the oil and gas station industrial control system, build the mapping relationship between the virtual space simulation model and the measured system entity, and realize the rapid access and reliability verification of the oil and gas station and warehouse PLC industrial control system. The platform significantly improves the operation quality of the industrial control system and reduces the occurrence of production safety accidents.

Short wave communication is a common method in the military field for long distance communication. In order to improve the anti-multipath fading of the short wave sets, a kind of short wave chaotic frequency hopping sets is designed. According to the character of the sets and the practicable of the address generator, a chaotic frequency hopping sequence is selected. The system simulation is carried out on the basis of analyzing anti-jamming ability in the short wave multiple access fading channel. Numerical results show the anti-multipath fading ability is well in the short wave sets based on the chaotic FH-SS technology.

As a new generation of digital technology, digital twin has attracted the attention of academia and industry in recent years, and has been applied in many fields and industries. An intelligent radar prognostic and health management architecture is built based on digital twin. In this paper the fusion application of five-dimension digital twin model and radar use/maintenance is put forward. The design idea of digital twin in radar status monitoring, fault prognostic, health evaluation and management is discussed. An overall architecture and new solutions for the intelligent radar maintenance support are provided in order to promote the further application and development of digital twin in the field of radar.

With the increasing speed of construction and the continuous improvement of its corresponding infrastructure, the expressway plays an increasingly important role in promoting the development of China's economy, tourism, transportation, and cultural industries. However, at present, traffic congestion in the expressway is becoming more and more frequent, which has seriously affected people's travel. In this paper, according to the congestion area of the expressway, the control method of the graded variable speed limit is proposed to realize the dynamic control of vehicle speed on the road, improving the traffic efficiency and safety. Through the analysis of the simulation results, after the implementation of hierarchical variable speed limit control, the speed change of vehicles on the road section is more gentle. The speed fluctuation is smaller, which shows that the research on the hierarchical variable speed limit control in traffic bottleneck areas can provide a theoretical basis for alleviating road congestion and reducing traffic accidents. The research is of great significance for improving traffic efficiency and safety.

Laser cleaning technology as a green and environmentally friendly cleaning method, in recent years in the field of industrial cleaning rapid development. This paper introduces the current status of research on the main parts of the laser cleaning machine control system, including laser cleaning detection system, scanning vibrator control system, laser power control system, control algorithms and other control parts. The characteristics of existing control systems and control methods are reviewed, and it is found that the main problems in this field are the difficulty and high cost of control system development, few links between different control methods and low automation, etc. An outlook on future development is given, including optimization of laser cleaning machine control research directions, enhancement of control technology and establishment of cleaning databases.

Aiming at the problems of frequent transfer, rough transfer mode and large transfer range of high-voltage cable in open-pit coal mine, which lead to low cable allocation efficiency, untimely maintenance and repair, and even cable loss and theft, a low-power high-voltage cable positioning method in open-pit coal mine based on GPS is proposed. In order to reduce the battery loss, the cable positioning terminal adds the motion state detection function. When the cable is moved in a large range and scale, the positioning terminal wakes up and increases the transmission frequency, and continuously sends the GPS positioning information to the control center through the LoRa network; When the cable is stationary or the movement is not obvious, the positioning terminal will sleep for a long time and send positioning information to the control center with a long transmission cycle. The electronic fence function is adopted to realize the anti loss and anti theft function of high-voltage cable. If the high voltage cable appears outside the electronic fence, the system will alert the administrator. The control center analyzes and processes the high-voltage cable position information to realize the purpose of monitoring and management. The practical application results show that the use of this method not only improves the allocation, repair and maintenance efficiency of high-voltage cables in open-pit coal mines, but also effectively reduces the loss and theft of cables.

In order to strengthen the understanding of PLC techniques, improve the ability of automatic production of automobile in China, this paper analyzes the application of PLC techniques in automatic production line of automobile. This paper first describes the basic principle of PLC techniques, and analyzes the key technology and design principles of PLC techniques control system. Finally, it summarizes the specific application of PLC techniques in automobile production line. This article aims to give some inspirations and help the relevant workers, strengthen the understanding of PLC techniques at the same time, realize the rapid production of automobile automation, promote the improvement of automobile production efficiency and quality.

In recent years, with the popularity of deep learning, speech synthesis technology has developed rapidly and achieved many good achievements. Among them, the technology of speech synthesis for personalized voiceprint features has also become a research focus. In the existing work, the model for personalized voiceprint feature speech synthesis based on GANs has achieved certain results. The model successfully synthesized speech with personalized voiceprint features in a non-autoregressive way, but the audio quality of the synthesized speech and efficiency was low, and the model training time was long. In this paper, we improve the model through techniques such as multi-domain signal processing. Specifically, we reduce a lot of training time by optimizing several parameters of the model. In addition, the architecture of the model has been improved to a certain extent, which effectively improves the MOS score of synthesized speech.

With the continuous development of robot research field, the research and application of humanoid biped robot has received widespread attention. Humanoid biped robot integrates many frontier disciplines such as machinery, electronics, computer, control and biology¹. In this paper, according to the theoretical basis of human biology, the mechanical structure of the biped robot body is designed, and the mathematical knowledge needed to establish the coordinate system of each rod of the biped robot is given. The forward and inverse kinematics of the robot are modeled respectively by using the homogeneous coordinate transformation, which lays a theoretical foundation for the subsequent gait planning². A walking model based on 3d inverted pendulum model was generated, and a dynamic walking experiment was carried out to verify the feasibility and effectiveness of the gait planning method. The stable gait path is programmed by simulation, and then its stability is determined.

In the process of operation, spraying UAV will be affected by the coupling effect of pipeline tension, spray reaction force, spraying mechanism load and other factors. These dynamic loads will directly affect the stability of flight attitude of intelligent spraying platform, resulting in the spraying quality can not be guaranteed. Therefore, a spraying stability control method of UAV based on active Disturbance Rejection control (ADRC) algorithm was proposed to solve the spraying stability control problem and improve the spraying quality of UAV. The simulation results show that the proposed multi-load coupling spraying UAV stability control method has high control accuracy and meets the requirements of spraying UAV control stability and robustness.

This paper designs an automatic obstacle avoidance tracing vehicle control system based on single chip computer. Based on STC89C52 single chip computer as control core, this paper detects the preset tracing guide line with infrared photoelectric sensor and feeds back the detection signal to single chip computer. The single chip computer will analyze and judge the acquired information and control L293D chip will drive DC motor for the control of left and right wheel rotation speed and adjustment of vehicle rotation direction so that the vehicle can automatically drive along the preset tracing and achieve the automatic tracing function. The operator uses HC-SR04 ultrasonic sensor to achieve obstacle avoidance and distance measurement function. The display module and alarm module equipped is able to have real time acquisition, display and control of the vehicle operation status and parameters. It designs and debugs the hardware circuit and designs the control software program. The vehicle is able to realize automatic tracing, obstacle avoidance and distance measurement function through physical test. It can run stably and reliably. The system circuit of simple structure and easy operation is applicable to unmanned factory, logistics, scientific exploration and other fields, which is of strong practicability and promotion value.

It has put forward higher requirements for the security management of the warehousing with the rapid development of global economy, sharp increase of warehouse and enlargement of storage scale. This paper designs a warehouse remote intelligent monitoring system. This system takes STC15F2K60S2 as the core of single chip computer, temperature and humidity sensor, smoke density sensor and human infrared pyroelectric sensor as acquisition elements of ambient parameter. The ambient parameter acquired can be displayed in OLED display module and single chip computer has data communication with the mobile phone through WIFI module. The single chip computer will control the action of execution mechanism for cooling and dehumidification when the warehouse ambient temperature is higher or lower than the preset value and give an alarm and send relevant information to the mobile phone in case of any illegal intrusion. User can know the real time ambient environment information, set the ambient parameter threshold and switch manned or unmanned mode through mobile phone. This paper gives the system design thinking, designs the system hardware and software, draw the system circuit diagram and compile the system source program and complete the physical system. It finally debugs the system hardware and software and the debugging result shows that the system can meet the control requirements and realize the remote intelligent monitoring.

During the construction of the Lotus Exclusive Line Project of Shijiazhuang Cigarette Factory, problems may occur at different stages such as equipment commissioning, process testing, product testing, and mass production that restrict the production operation, quality assurance, equipment maintenance, and efficiency improvement of the Exclusive Line. In response to this, modern industrial information, digital, and intelligent core technologies are used, with the goal of creating a smart workshop integrating autonomous control, intelligent scheduling, custody monitoring, autonomous operation and maintenance, and efficient management. To this end, it is necessary to carry out an intelligent stand-alone application through key equipment such as cut strips drying and bulk loosening and conditioning. Moreover, the matching relationship is explored between equipment processing performance and process indicators, and the application scenarios of intelligent instrumentation are fully used to solidify the foundation of intelligent control and big data application. Based on the material-based driving strategy of the cell production line, the utilization rate of the line is improved, and the construction and practice of production management mode for intelligent manufacturing of Lotus Exclusive Line are finally accomplished.

Taking the gas extraction pump system as the research object, an improved SOA algorithm is proposed and applied to the model identification of the system, and then the identification accuracy of the system is verified and analyzed under different working conditions. ISOA algorithm solves the problem that it is easy to fall into local optimal value in the later stage of the algorithm, and will not increase the complexity of the algorithm. Compared with the standard SOA algorithm, the ISOA algorithm has higher convergence accuracy and faster convergence speed; and has stronger all search ability. Under shutdown condition, the results show that the maximum relative error is 7.62%, and the average relative error is 1.48%. It shows that the simulation results of the model are in good agreement with the actual data, and the method of using ISOA algorithm to identify the gas extraction pump model is reliable and effective.

With the increasing demand for human travel, autonomous driving vehicles are gradually becoming a multi-domain, multimodule product of advanced technology. The motion control module, as an extremely important sub-module of the autonomous driving system, is divided into lateral control and longitudinal control. Therefore, in order to precisely control the vehicle's motion trajectory, the vehicle's control algorithm needs to have a more sophisticated design. Based on Matlab\Simulink and Carsim, the paper simulates the vehicle kinematics model and control algorithm, and uses model predictive control algorithm to control the speed and heading Angle of the test vehicle. Through the simulation platform test, the error between the actual trajectory and the expected trajectory is small under the preset conditions, and the lateral control stability is good. The results show that the application of model predictive control algorithm in vehicle lateral control can improve the safety of autonomous driving and have better control effect.

The positioning system based on the time difference of arrival (TDOA) has a systematic error in a specific direction. This paper uses the Multi-agent Deep Deterministic Policy Gradient (MADDPG) method for environmental modeling to improve the error distribution of the positioning system. The element configuration and execution steps of reinforcement learning modeling are analyzed in detail. And then the travel path is planned for the purpose of improving the positioning accuracy. At the same time, the model of the positioned unit under the confrontation condition is established to better get rid of the positioning. In the simulation environment, the camp of unmanned aerial vehicle (UAV) constantly learns and adjusts the strategy to detect the strategy of enemy using the effective area critical line to avoid positioning. The training process is summarized, and suggestions are put forward for the confrontation decision of both camps.

In order to improve the labeling effect and the work efficiency of the airline baggage labeling machine, the neural network is applied into the tension control for the tension fluctuation of the paper tape in the winding section. Firstly, the mechanism of tension generation is analyzed, and the mechanical device and dynamic model of tension control system in winding section are established respectively. Then combined with the coordinated control strategy, the simulation model of the winding section system of the labeling machine is built by Simulink. Finally, through the co-simulation of the winding segment model and the neural network controller, the control effect under the running state of the labeling machine is studied, and the comparison with the traditional PID control is analyzed. Compared with the traditional PID control, the simulation results show that the neural network PID has faster response time and smaller steady-state error , and has better dynamic performance. The stability control of the tension between paper tapes by the neural network algorithm is verified by simulation, which provides a theoretical basis for the continuity and accuracy of the subsequent labeling process and the guarantee of labeling quality.

Belt conveyor is the important coal production and transportation equipment. if it operations in high speed for a long time in case of no coal or little coal, it will cause a lot of power loss and belt damage. An improved yolov5 real-time coal flow detection algorithm is presented. The Swin Transfomer attention mechanism is used to improve the traditional convolutional receptive field limited problem, and the weighted feature fusion splicing method is applied to adaptively select the backbone feature extraction, so that the network structure can obtain the global semantic information of the feature map and effectively improve the detection ability. Compared with YOLOv5 algorithm, the experimental results show that the mAP performance improves by 2.1% and reduces the detection time by 10.6%, which can quickly and accurately detect the conveyor belt coal flow in real time.

Dense 3D reconstruction is crucial in the field of intelligent robot, drone application, and intelligent control. However, most state-of-the-art real-time dense 3D reconstruction methods require depth camera. In this paper, we propose a real-time dense voxel 3D reconstruction method that relies only on binocular cameras without depth sensors. Our method is based on binocular disparity estimation and sparse pose estimation and can be dynamic adjusted while the keyframe pose changes in SLAM(Simultaneous Localization and Mapping) system. Our result can be used for path planning and control of robots and vehicles. Experiments on our virtual dataset and KITTI benchmark show that our method can achieve real-time performance of 5~10Hz and localization accuracy of 0.107(Chamfer Distance).

Miss distance evaluation is one of the important contents of weapon effectiveness analysis. It plays an important role in the development demonstration, system evaluation, operational use and formulation of operational scheme of weapon system. Based on the environmental conditions and miss distance evaluation requirements of naval gun shooting at sea, this paper puts forward the miss distance evaluation method of naval gun shooting based on video image, uses UAV to obtain the video image signal of naval gun shooting from the air, uses digital image processing technology to determine the position of target and water column, calculates the dispersion error and completes the miss distance evaluation.

Traditionally, the voltage and current sampling wiring for the device of low-voltage station area acquisition and monitoring is complex. Furthermore, there are many nodes, and each node is measured by wiring, consequently safety and efficiency have to face great challenges. This paper proposes a new idea of independent voltage and current sampling, using high-precision synchronous time synchronization and high bandwidth wireless transmission technology to promote the miniaturization design of wireless metering devices and smart terminals. It is verified that the synchronous timing accuracy of less than 10us, and the high bandwidth transmission above 1Mbps can be achieved to meet the requirement of 200Kbps. The results show that the ping-pong timing mechanism and broadband wireless technology used in this paper achieve the expected effect. It will improve ubiquitous sensing, data fusion and intelligent application capabilities of distribution equipment at all levels in the future.

The traditional mechanical lightning arrester monitoring device needs to climb the tower manually. The number of records is limited, with the safety risks of lightning data false alarms and inspection personnel falling. The online monitoring scheme of transmission line arresters can automatically measure leakage current and record lightning time and cumulative lightning times. The arrester monitor with microprocessor (MCU) as the core, composed of solar cell and wake up, leakage current collection, lightning strike count pulse, wireless receiving/transmission control, LCD and other circuits. The arrester online monitoring system consists of an arrester monitor, concentrator and network monitoring system. The lightning arrester online monitoring system can be applied to different tower types and voltage levels. The system has been implemented in Fushun, Liaoning Province, with a good operation state and total application value.

This paper designs a intelligent system that can automatically guide car owners to park quickly in order to solve the problems, such as difficult parking, long distance, slow parking, difficult finding of the exit with the improvement of production force and the increase of the vehicle number. This system is able to automatically detect whether there is vehicle in the parking place by using infrared sensor and display the total number of the parking places and remaining parking place information in the garage on the display screen at the garage entrance. There is corresponding display screen that can display the total number of remaining parking places in the regions of the parking lots through the analysis and calculation of the main control module of the one chip computer. In addition, it uses a dot matrix to visually display the specific position of the spare parking places in different regions, conveys to the vehicle owners whether the parking place has vehicle information through indicator and guide the vehicle owner to rapidly find the parking place with voice announcement. This system is of accurate positioning, low cost, easy installation and anti-interference, which can achieve the purpose of humanized, high efficient and intelligent parking.

In this paper, it is concluded that the falling accidents about buildings is the highest through data analysis. In order to promote the steady and healthy development of the construction industry, a comprehensive information model based on BIM technology combined with REVIT, RFID, VR and so on. Effective control of high fall safety management and provides a new way to prevent safety accidents from falling from high buildings.

In recent years, the rise of driverless technology makes the video transmission bandwidth of vehicle platform larger and larger. With the increase of the number of cameras used in cars, a large number of cameras will inevitably produce a large amount of data, which greatly increases the data bandwidth of the on-board platform. "Bandwidth" has become one of the biggest challenges faced by automobile manufacturers. This paper mainly studies the video acquisition and storage technology of multi-channel camera based on GMSL (Gigabit multimedia serial links), which can well solve the problem of data transmission bandwidth. Firstly, the principle of GMSL technology is studied; Secondly, taking FPGA processor as the core control chip, a multi-channel video acquisition and storage system based on GMSL is designed, the overall design scheme and hardware architecture are given, and the hardware circuit design of the system is completed; Finally, the technical feasibility of the system is verified, and the experimental effect of the system is good.

Aiming at the problem of insufficient disturbance suppression ability of photoelectric stabilized platform system under traditional ADRC, an Active Disturbance Ｒejection Controller based on variable gain differentiator(VGD-ADRC) is proposed. In order to alleviate the contradiction between observation accuracy and noise sensitivity, a variable gain differentiator is used to replace the linear extended state observer. At the same time, the tracking error feedforward and disturbance feedback are separated to reduce platform overshoot. By comparing the simulation of VGD-ADRC and traditional LADRC, the results show that the VGD-ADRC can eliminate the overshoot of the system, improve the noise suppression ability and enhance tracking accuracy of the photoelectric stabilization platform.

Instrument sound extraction and beat marking in music signal processing are important applications for analyzing the emotions expressed in music. this paper focuses on the extraction and localization of snare drum and bucket drum sounds in the musical loop sequence of a drum set. Firstly, the author uses artificially extracted snare drum and bucket drum sounds from a drum music loop database to construct two average model to represent the general sound model. Then the pre-processed signals are filtered by matching the snare drum average models of the snare drum with the input drum music loop signal through a cascaded matched filter. after that, the pre-processed signal is sketched by a local maximum sampling operation to useless features. Then the overall signal is binarized to 0 and 1 by an adaptive threshold. Afterwards the point which value is 1, is considered as the onset for potential snare drum sound signal. with comparing the difference in amplitude and the difference in variance distribution for all possible snare drum sound signal, the position of final target signal in input drum music loop signal is confirmed. Finally, I suppress snare drum signal derived from the first layer in the original sound and repeat the above steps for the detection and tracking of the barrel drum. To demonstrate the robustness of the method, I use soft-sound sources databases and expanded the number of samples to 210 by random slicing, and by comparing the original data, I found that the method achieved an overall segmentation accuracy of 80.5% and a correct detection rate of 94.4%% in a cyclic sequence of drum kit sounds without consecutive hits. This reflects that the algorithm also gives acceptable results compared to other methods.

In this article, the efficiency of two methods of speech enhancement algorithm, both of which used Deep Neural Network, are compared. The first method is based on mapping and the second one based on masking. For the method based on mapping, the input of the network is the power spectrum of noisy signal and figure out the power spectrum of suppressed signal. For the algorithm based on masking, the input is the Time-Frequency spectral of noisy signal, and then figure out the Ideal Ratio Masking (IRM) of the signal. And by multiply the element of IRM matrix with the corresponding element of spectral, the signal of pure speech is expected to be got. And the compare of the PESQ scores of the ideal signal and the result figured by the different methods shows, that in the method based on the masking performs better.

The uncertainty of the model parameters of the quadrotor brings great challenges to the controller design, especially the uncertainty of the arm length will affect the virtual control input, resulting in the deterioration of the control performance. For the quadrotor system with variable arm length, a control strategy based on integral backstepping method is proposed in this paper. By dividing the quadrotor system into attitude and position subsystems, the arm length is regarded as an unknown parameter in the attitude subsystem. Then, the adaptive method is used to estimate the unknown parameters, and the integral backstepping method is used to design the attitude controller. The position controller is designed based on the integral backstepping method. The stability and tracking performance of the closed-loop system are proved by using Lyapunov theorem. Finally, the numerical simulation results show the effectiveness of the designed control law on a quadrotor model with uncertain arm length.

Nowadays, the terminal detection of electric distribution cabinet is mainly completed manually. Professional technicians find the pair of devices and binding posts according to the circuit wiring diagram, and this process can be cumbersome, time-consuming and error-prone. This paper designs and implements an intelligent wiring detection assistant system, which manages the basic process data of wiring table and panel layout diagram and make the work process-oriented and visualized. It guides the worker to deal with the pairs of connecting wires in an optimized order and locate the detection points quickly according to the hot points of panel layout diagram. It also provides a credible detection result analysis. The proposed system improves efficiency and reliability from the traditional wiring connection quality detection of electric distribution cabinets.

With the rapid development of artificial intelligence, the research on perceptual intelligence is becoming more and more mature. For the next stage of cognitive intelligence research, knowledge graph (KG) is one of the key directions. Knowledge reasoning is an important part of knowledge graph and has a wide range of application requirements. To solve the problems of current large-scale KG, including poor interpretability, low accuracy and efficiency of knowledge reasoning, this paper proposes a method RLPTransE which combines knowledge representation, relational path with deep reinforcement learning. This method generates a path vector by semantically combining the vectors of all relations on the path. Then, a reinforcement learning environment is established in the vector space, and through training a multi-step inference policy network based on diverse reward functions, the reinforcement learning agent can efficiently complete inference in the process of interacting with the environment. This paper conducted link prediction and fact prediction experiments with the RLPTransE method on NELL-995 and FB15K-237 datasets, and compared with knowledge representation-based, relational path-based and fusion-based methods, the experimental results show that RLPTransE method achieves better performance on large-scale dataset inference tasks.

After the emergence of e-commerce platforms and the rapid growth of users, the growth of domestic e-commerce traffic has gradually peaked. It is known that improving flow conversion is an important and urgent issue. This paper mainly constructs an accurate user purchase prediction algorithm in two aspects: Feature Engineering and model construction Xgboost is introduced to construct an iterative lifting tree for the original features, measure the importance of the features, and select the features to reduce the complexity of the model solution and improve the prediction accuracy; In terms of model building, the multi-model fusion method based on stacking is used to design a two-layer fusion model. The first layer uses xgboost and lightgbm as basic learners to predict users' purchases from different angles. The second layer uses logical regression to fuse the prediction results of the basic model with meta learners to obtain more accurate purchase predictions. The experimental results show that the accuracy and stability of the model are better than the benchmark model. Accurate purchase prediction helps improve the user experience and marketing promotion effect to improve the traffic conversion rate.

When dealing with the problem of less variables, there are usually many simple and practical methods to solve this kind of problem. However with the scale of the problem continues to grow, the effect of these methods will decline linearly or even exponentially. At the same time, many real problems have the large amount of variables which leads to a significant increase in the dimension of the problem. Usually, we call this kind of problem as large-scale optimization problem. At present, the common idea for large-scale problems is to reduce the dimension of the problem, that is, the decomposition of problem variables. At present, coevolution framework is a very effective method to solve large-scale problems. On this basis, considering the problem of resource allocation, a contribution based CC framework is produced. However, the accuracy and efficiency of problem decomposition are difficult to meet at the same time. In this paper, based on the contribution knowledge of variables, a new decomposition method combining differential grouping(DG) and matrix filling is proposed to realize the fast dynamic decomposition of large-scale problems. Firstly, in the initial stage, DG is used to calculate the correlation between some high contribution variables. These variables are obtained by calculating their contribution ranking to the problem. The correlation between other low contribution variables is finally decomposed by matrix filling. The experimental results on CEC2010 testing benchmark problems confirm the effectiveness of the method proposed in this paper.

Typhoon is a common natural disaster, easy to harm the lives and property safety of people in coastal areas of China, so it is of great significance to accurately predict the path of typhoon. In this paper, the LSTM neural network algorithm is used to dynamically predict the typhoon path considering the longitude and latitude, central pressure, wind speed and typhoon intensity. The historical typhoon data of China National Meteorological Network are used for training and forecasting, and the error points in training and forecasting process and improvement measures are introduced in detail. The LSTM network model used in this paper can use the 24-hour typhoon data to predict the typhoon path at the next moment. Compared with THE RNN model, the accuracy has been further improved.

With the development of intelligent transportation, traffic sign detection has become a very important task. Traffic sign detection requires the positioning and classification of traffic signs in the road environment. Due to the complexity and diversity of road environments, traffic signs can only take up a small proportion of videos or pictures, and common algorithms still show high false detection rate and high time and calculation overhead in practical application. So far, traffic sign detection is still a difficult task. At the same time, most traffic sign data sets used by algorithms at the present stage have unreasonable data structures or incomplete database types. Blind application of unstructured data sets easily lead to the difficulty of obtaining good results in model training. Therefore, this paper decided to carry out the traffic sign detection task based on the lightweight YOLOv5 (You Only Look Once) neural network model. In order to achieve excellent target detection effect, this paper adopts optimized Tsinghua-Tencent-100K (TT100K) data set to train the model. The experimental results show that the trained models mAP@0.5 and mAP@0.5:0.95 reach 59% and 40% respectively, which basically meet the requirements of application.

Transformer has good feature extraction ability and has achieved good performance on various NLP tasks such as sentence classification, machine translation and reading comprehension, but it does not perform well in named entity recognition tasks. According to recent researches, the Long Short-Term Memory (LSTM) usually performs better than Transformer in NER task. LSTM is a variant of Recurrent Neural Network (RNN), because of its natural chain structure, it can learn the front and back dependencies between words well, which is very suitable for processing text sequences. In this paper, the BiLSTM network structure is embedded into the Transformer Encoder, and a new network structure BiLSTM-IN-TRANS is proposed, which combines the sequential feature extraction capability of BiLSTM and the powerful global feature extraction capability of Transformer Encoder. The experiments can reflect that applying the model based on BiLSTM-INTRANS could work better than applying one of LSTM or Transformer alone in the NER task.

With the increasing application in real life, drones have received more and more attention in recent years. In particular, drones show great potential when it comes to solving the last-mile delivery problems. Large-scale vehicle routing problems with drones (LSVRPDs) are challenging because most real-world problems typically involve thousands of delivery nodes. In this article, we proposed a two-layer heuristic optimization algorithm called THOA to address them. In the upper layer, three local search operators are used to generate different routes, each of which consists of several nodes. In the lower layer, a Branch-and-Bound method with special constraint is suggested to optimize the routes obtained from the upper layer. The proposed THOA is compared with a simulated annealing algorithm in six scenarios with different number of nodes. The experimental results demonstrate the effectiveness and efficiency of THOA, especially in VRPDs with a large number of nodes.

The recent transformers shows competitive performance on computer vision tasks, such as classification, detection, and segmentation. Inspire by its success, in this paper, we explore its application at some more notoriously difficult vision tasks such conditional generative adversarial networks and propose a transformer based conditional generative adversarial networks for image generation. Our model employs the transformer architecture to develop a generator discriminator and residual network as discriminator. To improve the performance, a spectral normalization technique is used to normalize the weights of discriminator and the hinge loss are determined for model optimization. The experimental results on four public datasets shows that our approach is capable of producing the high quality images with good consistence and diversity and outperforms existing works.

This study was designed to assay the weight loss rate, hardness, color and V_C of winter jujube with a texture analyzer under different storage temperature conditions. The zero order and first order kinetic model obtained from the physical and chemical indexes of winter jujube through the origin 8.0. The changes of winter jujube’s quality were followed firstorder model according to experimental data in this study. The shelf life model of winter jujube can quickly and accurately predict the changes of winter jujube’s quality in order to indicate the remaining shelf life of winter jujube during cold storage. The shelf life model of winter jujube can lead the safety consumption of fresh winter jujube for consumer.

The grounding grid is a critical facility to ensure the safety of personnel and equipment. After many years of operation, its cross-sectional area is often reduced due to corrosion, resulting in a decline in electrical safety performance. In engineering, the random excavation method is often used for fault diagnosis of grounding grid, which has a large workload and poor diagnosis effect. In this paper, based on the electrical network theory, an optimized particle swarm algorithm is constructed to diagnose grounding grid faults. MATLAB is used to optimize the measured data iteratively. It has high precision, solves the problem of blind excavation in engineering, and realizes the location of the fault area of the grounding grid.

In the current SMAC paper, almost no one has analyzed the reason of the difference of success rate% between PyMARL algorithms used in SMAC. The purpose of this article is to study the code differences and the idea differences between VDN algorithm and COMA algorithm in PyMARL framework used in SMAC environment, and the influence of these two algorithms on the result of SMAC experiment. The conclusion of this paper is that VDN algorithm is superior to COMA algorithm in SMAC. Because in almost all scenarios of SMAC, the algorithm based on value function (VDN) can solve simple cases very quickly and effectively, while the algorithm based on strategy gradient (COMA) converges to local minimum easily, and it is extremely difficult for COMA to come up with a perfect strategy.

The grains and oils supply chain has many characteristics, such as long life cycle, complex links, multiple sources of information, fragmented data, and many risk factors, etc. Based on the industrial Internet identification resolution system, the method and technology of information exchange in the whole supply chain of grains and oils are studied. Through the analysis of my country’s industrial Internet identification resolution system, design identification and resolution connection middleware architecture. The identification resolution system is the nerve hub that supports the interconnection of all things in industry. By opening up the data of the whole supply chain of grains and oils, it will realize the extensive interconnection of upstream and downstream industries and cross-fields. It breaks the "information island" and ensures the quality and safety of the whole life cycle of grains and oils. On this basis, a feasible information exchange framework for the whole supply chain of grains and oils is constructed. This has practical implications.

During the image capture process, it is difficult to maintain relative stillness between the camera and the subject, resulting in blurred images. In order to solve this problem, based on the generative adversarial network, this paper introduces a residual module to solve the gradient dispersion problem generated during the training process. The multi-scale theory is introduced to process the image to better extract the detailed information of the image. Finally, the experiment shows that the scheme has achieved good results.

With the proposal of "Made in China 2025" and the implementation of the strategy of manufacturing power, China's manufacturing industry is struggling in the traditional extensive, labor-intensive road, the development of intelligent manufacturing has become the only way for the world's developed countries to develop the manufacturing industry. This study constructed a theoretical model of "the new generation of information technology-integrated interconnection-intelligent manufacturing capability", and introduced policy support as a moderating variable to discuss the triggering effect and improvement path of the new generation of information technology on intelligent manufacturing capability. The results show that :(1) the new generation of information technology has a positive impact on enterprise intelligent manufacturing capability. (2) Integrated interconnection plays a partial intermediary role between the new generation of information technology and enterprise intelligent manufacturing capability. (3) Policy support positively moderates the relationship between integrated interconnection and enterprise intelligent manufacturing capability.

With the rapid development of artificial intelligence, the research on perceptual intelligence is becoming more and more mature. For the next stage of cognitive intelligence research, knowledge graph (KG) is one of the key directions. Knowledge reasoning is an important part of KG and has a wide range of application requirements. At present, knowledge reasoning methods in large-scale KG still have problems of poor interpretability, low reasoning accuracy and efficiency. Knowledge reasoning methods based on deep reinforcement learning have better interpretability and stronger reasoning ability. This paper introduces the research progress of knowledge reasoning based on KG, and makes an analysis of the current knowledge reasoning methods based on knowledge representation, relational path and deep reinforcement learning. At the same time, this paper conducts link prediction and fact prediction experiments on related knowledge reasoning methods on NELL-995 and FB15K-237 datasets, and summarizes the datasets, evaluation methods and indicators involved in these experiments. In conclusion, the proposed future knowledge reasoning model is based on the fusion knowledge representation, relational path with deep reinforcement learning methods.

Due to the emergence of highly intelligent new cybercriminals in the Internet, traditional case investigation methods cannot analyze the massive amount of data efficiently and accurately. Therefore, we can use relevant technology of big data to conduct correlation analysis on the network identities of the people involved, describe the relationship between the characters, and study the overlapping communities based on the separation of multiple roles. Then, according to the features that the ontology of the users involved and the associated targets contains multiple roles, the high-value targets hidden in the case can be mined. By mining, analyzing, and modeling the big data involved in the case, and constructing a portrait of the relationship between personal identities, it is possible to give an early warning and judgment on potential criminal tendencies, and to carry out demonstration applications in the public security system to combat complex new cybercrimes.

At present, the commonly used text classification methods are based on the classification function provided by the supervised learning algorithm. Faced with massive data, it has the problems of slow classification speed, low accuracy and single classification function. In response to this urgent problem, this paper designs a text classification system based on AI technology. The Visual C++ programming language is used in the design of the classification system, and the modular design is used for the system function modules, which makes the system stable and efficient under this development environment. In order to compare the applicability of manual classification methods and AI automatic classification methods, this paper conducts classification efficiency experiments and classification accuracy experiments. In the classification efficiency experiments, it is found that the larger the amount of text, the faster the AI automatic classification method is compared to the manual classification method. The faster it is, in the classification accuracy experiment, comparing the classification accuracy of the two methods for five types of texts, it is concluded that the average classification accuracy of the AI automatic classification method is 85.38%, which is 6.32% higher than that of the manual classification method. In general, the introduction of AI automatic classification methods into the classification system can achieve efficient and accurate classification.

Nowadays, China’s highway charges are manual, semi-automatic (MTC), and electronic toll collection (ETC), and traffic efficiency is relatively low. To improve the speed of highway toll stations, free-flow charging based on the Global Navigation Satellite System (GNSS) is a charging method with higher traffic efficiency. This paper analyzes and designs a sub-meter high-precision lane-level navigation and positioning system based on BeiDou’s ground-based augmentation system. Based on the technical principle of highway free-flow charging, a new mode and method of BeiDou lane-level highway vehicle wide-area free flow trusted charging is designed by adding mobile communication base station verification.

In this work, this paper discussed the maneuver of autonomous vehicles, the application of deep learning technology in autonomous driving and improvements in the recognition and training of autonomous driving. First, we discussed how the pure pursuit algorithm calculates the route to the destination. Then, for the maneuver of autonomous vehicles, we listed the use of fuzzy logic and actuator. For driving safety, we introduced eight factors that it depends on. To apply deep learning technology in autonomous driving, we introduced how deep learning is applied to autonomous driving by explaining the four main parts of the pipeline of autonomous driving: perception, localization, prediction and decision making. We also noted the KITTI dataset, popularly used in the deep learning academia. Finally, for improvements in the recognition and training of autonomous driving, we presented three ways: using neural networks, using more effective datasets, and virtual environments that simulate the real world.

In this paper, aiming at the uniqueness check of standard address import in mobile resource management system, a standard address intelligent import method based on Geocoding and TF-IDF is proposed. Address names are standardized through Geocoding to uniform the naming rules of addresses at all levels and remove the interference of non-mandatory level names on results. The address names recorded in the system are used as the address database, and the similarity of address names is calculated through TF-IDF. The address names with high similarity are filtered out, and then the address names are judged manually. Experiments show that this method can effectively improve the accuracy of uniqueness verification and greatly reduce the manual workload.

According to the multi-dimensional and multi feature characteristics of building energy consumption data, considering the fluctuation of long-time series prediction, partial attention mechanism, full convolution neural network and full-time attention mechanism are introduced to capture the potential correlation between features and targets and the weight between features, and an improved GeoMAN (Geo-sensory Multi-level Attention Networks) building energy consumption prediction model is proposed, Realize the prediction of building energy consumption data. The improved GeoMAN model is used to compare the prediction indexes with four groups of common models, and other data sets are input to verify the generalization ability of the model. The experimental results show that the model performs better in RMSE, Mae and MRE, reflecting better predictability and robustness.

Aircraft cables are widely used in the field of aeronautics and astronautics to provide power or contact information between various systems. Different types of defects inevitably occur in the cable insulation layer during actual use and correspondingly relevant measures should be taken. Therefore, it is essential to judge the defect type in the cable insulation layer. In this paper, an aircraft cable insulation layer defect detection system based on ultrasonic guided waves (UGWs) was built to collect defect reflected signals. The wavelet packet decomposition was used to extract the normalized energy of a reflected signal as its eigenvectors which could effectively distinguish defect types. The identification accuracy of aircraft cable defect types could reach 92.86% by using BP neural network.

Aiming at the problems of difficult layout of monitoring positions, low degree of automation, low accuracy of direction finding and positioning, and many blind areas in the implementation of electromagnetic environment monitoring and direction finding and positioning of important radiation sources, the research on electromagnetic radiation source monitoring and positioning based on UAV platform is carried out. The system architecture is divided into four functional modules: monitoring and positioning, ground presentation, air platform and data transmission, and the module capability requirements are analyzed based on the application scenario. The final system consists of multiple airborne terminals that communicate with the ground through 4G mobile network, and uses TDOA positioning method to locate the ground radiation source. The system has good monitoring and positioning effect, and can provide means support for the automatic and accurate generation of electromagnetic spectrum situation in important areas under some special circumstances.

It is always assumed that the listening room is a free-field conditions in previous acoustic experiments. However, a completely free-field conditions can impossibly be achieved due to reverberation of sound waves by the walls, which can adversely affect listening experience. So, room compensation is used to solve this problem. It can effectively reduce the influence of reverberation. Conventional adaptive filters fail to overcome the complexity of multi-channels. So, the first approach is introduced—eigenspace adaptive filtering which is based on decoupling of multiple-input-multiple-output system with generalized singular decomposition. But the varying feature of listening room alters related matrixes constantly, increasing the computational complexity of compensation processing. Then, another approach called wave domain adaptive filtering, mentioned here in the context of RLS algorithm, can be a feasible alternative to avoid computational difficulty. This paper also evaluates the performance of using circular harmonics as wave field representation.

In the context of carbon neutrality, distributed energy has ushered in new development. The distributed energy system in the industrial park also needs a reasonable and reliable data management system to manage the energy data in the park. In order to meet the requirements of distributed energy in the industrial park for data source modularization, clock synchronization, and sampling rate, a data management system based on distributed data acquisition, MDSPlus database, and timing synchronization system is designed. The system realizes efficient and reliable data collection, storage, and management for distributed energy.

In order to solve the problem that some high-performance deep learning neural networks are not ideal for application in embedded devices due to the defects of high complexity and large computation. From the perspective of easy hardware implementation, the hardware implementation algorithm of OFDM channel compensation based on deep learning is researched and designed, and the channel compensation scheme of conventional pre-equalization + DNN fine compensation is designed based on the idea of joint efficient implementation of FPGA + GPU.

A large number of electronic devices equipped by modern vehicles may become a means for hackers to invade vehicle network system. In order to prevent attacks, it is necessary to introduce the security mechanism of CAN-FD vehicle bus. But CAN-FD uses broadcast message transmission mechanism, lack of security protection, vulnerable to network attacks. Under security constraints, CAN-FD message packaging with low bandwidth occupancy(utilization) is a prerequisite for intelligent application operation in automotive information physics system. On this basis, a signal packaging security constraint that can effectively reduce the hardware cost of Electronic Control Unit(ECU) security design is proposed. In view of the shortcomings of the basic genetic algorithm that it is easy to fall into the local optimal solution prematurely and has poor local ability in the later stage, a hybrid genetic algorithm based on Newton method is proposed, and the local search technology is added to the genetic algorithm. This local search technology is to set up a selection mechanism and selectively use the Newton method to determine whether the algorithm is convergent. The performance of the hybrid genetic algorithm is much better than that of the basic genetic algorithm. The former can overcome the shortcomings of the latter that it is easy to fall into the local optimal solution prematurely and its later local ability is poor. The experimental results show that the algorithm could obtain lower bandwidth utilization, and the influence of the security design on the bandwidth utilization is analyzed by experiments.

In recent years, deep neural networks have achieved high accuracy in many classification tasks, including speech recognition, object detection, and image classification. Although the deep neural network is robust to random noise, when some special disturbances that cannot be detected by the human eye are added to the neural network input, these special disturbances will still cause the deep neural network model to output wrong predictions. For the defense method against adversarial samples, we propose an adversarial training method based on the combination of feature distillation and metric learning. This method is to pretrain a fixed teacher network training method and use clean sample training. The student network uses adversarial samples for adversarial training. During the training process, the clean samples are used in the middle layer features of the teacher network to guide the adversarial samples in the middle layer features of the student network, and the middle layer features of the adversarial samples are repaired in the student network to achieve good results. At the same time, considering the relationship between adversarial samples and clean samples in the student network, a metric learning loss is introduced in¹ the middle layer features of the student network, so that the distance between the adversarial samples and the clean samples is closer than that between the adversarial samples and the confused samples. This makes the deep neural network model more robust. Finally, we perform gray-box, white-box and black-box attacks to verify the effectiveness of our method. Our algorithm significantly outperforms state-of-the-art adversarial training algorithms.

Aiming at the problem of complex target data allocation of anti-radiation UAV, with the support of anti-radiation UAV target data, a combat target allocation algorithm is proposed to solve the optimal task allocation scheme of anti-radiation UAV. Firstly, the model of combat target allocation is established, and the problem of target allocation strategy is transformed into the problem of maximizing the mathematical expectation of cluster combat effectiveness. Secondly, the combat target allocation algorithm is proposed and the global optimality of the proposed algorithm is deduced. Finally, the simulation results show that the algorithm in this paper is more efficient and can obtain the global optimal solution compared with the classical traversing method and genetic algorithm. The dynamic combat target allocation algorithm proposed in this paper can greatly improve the combat effectiveness of anti-radiation UAV.

Logistics UAV delivery has been well developed in the fight against COVID-19 pneumonia, and attracts more and more scholars to research. Ant Colony Optimization (ACO) is one of the effective solutions to solve the UAV task assignment problem. The algorithm adopts the principle of positive feedback to speed up the evolution process. However, the algorithm has some defects, such as long search time, easy to fall into local optimum and so on. Aiming at the defects of ACO, we put forward two improvements in this paper: On the one hand, differential distribution of initial pheromone is proposed to avoid blind search in the initial stage and improve the convergence speed. On the other hand, we will reduce the number of candidate nodes in the dynamic strategy, and ants choose the next node in the dynamic candidate list to reduce the calculation of local exploitation. Simulation results show that the improved ACO can significantly improve the convergence speed and has a good effect on solving the task assignment problem of logistics UAV.