This PDF file contains the front matter associated with SPIE Proceedings Volume 12171, including the Title Page, Copyright information, Table of Contents, Author and Conference Committee lists.

In hard disk drive (HDD) manufacturing processes, there are unrecovered serial number images about 0.01% from the standard optical character recognition (OCR) reading and deep learning approach. We found several failures from two main causes, i.e. manufacturing process and image capture process during standard OCR reading. We proposed classification model used for recognizing the serial number reading failures based on object detection You-Only-Look- Once (YOLO) algorithm and EfficientNet-B0 classification network as well as histogram analysis. The 1000 images captured by digital camera were used for training (600 images) and validation (400 images) the ROI detection model. The other 2100 captured images were used for training and testing classification OCR failure from manufacturing process model. The model testing was performed in 900 images contained 9 causes (classes) of failures. The proposed model reaches F1 score = 0.94.

The previous two-stage deep learning model for detecting and classifying misidentified serial numbers on the defect hard disk drive (HDD) slider was proposed by authors. We found that the threshold level adjusted during preprocessing process could limit the robustness of the two-stage model in large-scale manufacturing. Thus, we proposed a three-stage deep learning model comprised of 1) region of interest (ROI) detection and cropping, 2) character detection and cropping, and 3) character classification. Object detection algorithm and classification network used in this model are based on YOLO v.4 and EfficientNet-B0. The 1000 images captured by the digital camera were used for training (600 images) and validation (400 images) of the ROI detection model. The other 1000 captured images were used for testing the performance of the proposed three-stage model, then we compared them with those obtained from the previous two-stage model. The proposed three-stage model reaches F1 score = 0.997 and recovery rate up to 95.9%, while the two-stage model yields only 0.948 and 73%, respectively.

Influenza results in massive casualties and property loss each year due to the high infectivity and multiple variation of its virus. Thus, early diagnosis of influenza viruses plays a crucial role, however rapid an on-site detection can hardly be achieved currently. Organic electrochemical transistor (OECT) has the characteristics of high sensitivity, fast response speed and good biocompatibility, and has been widely used in the detection of DNA and protein. In this work, we developed an OECT-based influenza virus sensor. Through the detection of HA peptide antigens, it shows that the sensor has a detection limit of less than 10^-9M.

Traditional adaptive beamforming algorithms require high accuracy for steering vectors(SV), array models and desired signal(DS). However, the performance of the beamformer will seriously degrade when the DS is present in training snapshots. For the purpose of improving output performance of adaptive beamformer, a novel adaptive beamforming algorithm is proposed. This approach estimates the desired signal SV and reconstructs the sampling covariance matrix (CM) based on integrating over a undesired signal region. Furthermore, only a little prior knowledge is required, such as the approximate incident angle of the DS. The proposed algorithm remove not only the influence of the DS in the sampling covariance matrix, but also the effect of background noise perturbation, which is significantly improved compared with other methods. The results of data simulation experiments confirms that the beamformer has a excellent performance in output performance.

In this paper, we propose an optimal respiratory waveform selection algorithm based on range-multiple beams using a 77GHz frequency modulated continuous wave (FMCW) multiple-input-multiple-output (MIMO) radar. Generally speaking, the human chest is a multi-scattering target and the optimal monitoring position changes in a small range during breathing movement. According to this motion feature, we roughly locate the target in the range-angle candidate box based on range fast Fourier transform (FFT) and Capon direction of arrival (DOA) algorithm, respectively. Additionally, the fixed beamforming is utilized to algin the detected target site which can reduce the interference of clutter and enhance the signal-to-noise ratio (SNR). Then, the extended differential and cross-multiply (DACM) algorithm is further applied for phase unwrapping and the optimal respiratory waveform is extracted based on the features of respiratory periodicity. Ultimately, the respiratory rate is estimated by the frequency-time phase regression (FTPR) algorithm. Experiments with and without interference are conducted and the results show that the proposed algorithm can obtain accurate respiratory rate with mean square errors (MSE) 0.6862 breath²/min compared with the reference vital signs data.

With the increasing use of digital array radars, radar signal processing systems have higher performance requirements. This article introduces a signal processing hardware design for the two-dimensional phased array digital multi-beam system. Because of its digital multi-beam characteristics, it is very demanding on the radar signal processing system's computing power, processing speed, and data throughput. This paper proposes a design of signal processing hardware based on Xilinx FPGA Virtex-7 and two multi-core digital signal processors (DSP) to meet the requirements of two-dimensional phased array digital multi-beam system. Subsequent experiments and engineering practices show that this design scheme can fully meet the requirements of the system.

The low-light image enhancement plays a crucial role in computer vision and multimedia applications. However, it is still a challenging task, as the degraded images reduce the visual naturalness and visibility. To address this problem, we build a novel variational Retinex model to accurately estimate the illumination and reflectance components. The illumination and reflectance are jointly updated by alternating optimization algorithm. Experimental results on several public datasets demonstrate that the proposed method outperforms the state-of-the-art methods in Retinex decomposition and illumination adjustment.

Due to the excellent performance of radar sensors in terrible environments, the potential application of radar in intelligent transportation has become a research hotspot. This paper proposes an approach to estimate the number of vehicles in bidirectional and multi-lane using MIMO radar. The radar is installed at the top of the overpass to track the vehicles. The core idea of the proposed approach is using the vehicles' trajectories to realize vehicle counting and adjusting the overtaking and lane changing situations. In the proposed approach, we first propose a method to avoid missing target detection. To improve the accuracy of vehicle tracking, we exploit the doppler information in the trajectory association. Finally, in order to eliminate the false alarm, we use the continuity and length features of trajectory in the vehicle counting process. Experiment results show that the proposed approach can achieve an average vehicle counting more than 95%.

Unscented Kalman filter (UKF) based on correntropy criterion shows robustness when power system measurement suffers from non-Gaussian noise. To improve the performance of traditional algorithms, this paper proposed a generalized mixture correntropy unscented Kalman filter (GMC-UKF) for power system dynamic state estimation. Specifically, we construct the mixture correntropy by two generalized Gaussian kernels. After introducing the weighted state error and measurement error into the mixture correntropy cost function, we adopt fixed-point iteration to obtain optimal estimation. Finally, the robustness and accuracy of the proposed algorithm for power system state estimation are verified on IEEE-30bus.

In the context of the strategic goal of building energy Internet, State Grid Corporation of China has established a crossdisciplinary joint construction and common Internet of Things management platform, which carries out standardized access and unified online control for equipment in transmission, power transformation, distribution and other fields. In order to improve the access debugging efficiency of the IoT management platform accessed by terminal devices, this paper designed a full-scene simulation detection system for the intelligent IoT system oriented to the energy Internet. By using the existing construction ideas of intelligent IoT system and following the complete architecture of "cloud, tube, edge, terminal ", the construction method of the simulation and coordination platform of intelligent IoT system of "information-physical coupling" is realized. Advanced technologies such as micro-service and distribution are adopted to build a lightweight simulation test system. A small amount of server resources can be used to simulate and test whether the communication protocol of the side device supports the communication protocol requirements of the intelligent IoT management platform. The functions of reporting equipment data collection and issuing control commands are tested, so as to complete the simulation, verification and testing functions of the whole electric power industry chain intelligent IoT system covering both internal and external businesses, and provide technical support for the construction and promotion of the IoT management platform of State Grid Corporation of China. Combined with the whole scene simulation and detection system of the intelligent Internet of Things system, the operating state sensing monitoring scene of the power distribution area is built to verify the functions of data reporting of the fusion terminal and command issuing of the simulation and detection system.

With the increasing use of digital array radars, radar signal processing systems are faced with the challenge of real time processing for a mass of data. A kind of field programmable gate array (FPGA) realization of signal processing is proposed for two-dimensional phased array digital multi-beam radar in this paper. This scheme can realize the real-time signal processing of 6 beams at the same time. Each beam processes data from 8400 range gates. The details of implementation including digital multi-beamforming (DBF), pulse compression and moving target detection (MTD) are described. The results show that the design is correct and feasible.

Using a natural conversation paradigm, this study investigated the acoustic characteristics of Mandarin utterances in drug addicts. Twenty-one native speakers of Mandarin, including four heroin addicts, two 3,4-methylamphetamine (MDMA, also known as ecstasy) addicts, and 15 healthy controls without any history of drug abuse, were recruited for the speech production experiment. In comparison with the healthy controls, heroin addicts exhibited a higher mean F0, a lower mean intensity, a higher variability in both F0 and intensity, and a lower H1-H2, while MDMA addicts exhibited a higher variability in both F0 and intensity. Discriminant analysis based on these acoustic parameters further showed a good accuracy of differentiating the three groups of speakers. These findings provide the basis for future research into identifying drug addicts on the basis of speech signals.

It is not easy for a single passive platform to realize target location by just bearings-only measurements, unless it carries a time-consuming laborious maneuvering. By contrast, with two platforms we could simply achieve accurate target location through triangulation approach without the need of platform maneuvering. However, this two-platform approach introduces an inter-platform association problem in multi-target scenarios, that is, one should recognize and exclude the false targets generated from wrongly paired bearings. To solve this, in this paper, a target motion analysis (TMA) based approach is proposed. Through a pre-TMA of all possible targets, the method calculates the confidence coefficients of all possible targets for recognition (real or false), which are then used to associate inter-platform bearings and select real targets. Simulation verification is performed through using Monte Carlo technique. The result demonstrates that the proposed method can achieve accurate inter-platform multi-target data association and multi-target location.

Unstructured road segmentation is a key task in self-driving technology and it’s still a challenging problem. Mostly available point cloud datasets focus on data collected from urban areas, and approaches are evaluated for structured roads or urban areas, which has considerable limitations in rural areas such as fails at night, road without boundary lines, and no markings. In this regard, we present a new large-scale aerial LiDAR dataset of rural roads with hand-labeled points spanning 500 km² of road and nine object categories. Our dataset is the most extensive dataset contains a critical number of expert-verified hand-labeled points for analyzing 3D deep learning algorithms, allowing existing algorithms to shift their focus to unstructured road data. The nature of our data, the annotation methodology, and the performance of existing state-of-the-art algorithms on our dataset are all described in detail. Furthermore, challenges and applications of rural area road semantic segmentation are discussed in detail.

Cardiac electrophysiology and drug study on the cardiomyocyte culture make great progress. In the present study microelectrode array (MEA) was used to detect the mechanically beatings of the whole heart and the cardiac tissue slice in vitro with multi-sites. The experiment results show that the cardiac tissues in vitro can keep good status and favorable beating. Additionally, the cardiac tissue slices can maintain the structure and characteristics of the whole heart, which is accordant to the excitation-contraction coupling mechanism. Using this sensing technology based on the myocardial tissues, we synchronously obtained the electrical signals of different positions of myocardial tissue, which is helpful to judge the direction of myocardial electrical signal propagation and conduction.

With the development of the Internet of Things technology and the continuous improvement of people’s living standards, smart devices based on the Internet of Things are emerging one after another and intelligent mattresses are one of the important research directions. Many scholars at home and abroad have conducted a lot of researches on the vital signs monitoring of intelligent mattresses to improve the efficiency of elderly care. This article introduces the monitoring technology of intelligent mattresses, expounds the mainstream data collection methods and data analysis methods, comparative analyzes the advantages and disadvantages of different monitoring technologies and discusses the application of intelligent mattresses in vital sign monitoring, sleep monitoring and elderly care, finally summarizes and prospects the development of intelligent mattresses.

Power Internet of Things (IoT) devices are increasingly being used in power systems. Due to the variety and complexity of devices, there is no mature solution for security detection. In order to solve this situation, this paper proposes a power Internet of Things device security detection system. Based on key technologies such as device fingerprint identification, network protocol identification and firmware security analysis, comprehensive security analysis is conducted on power Internet of Things devices from the perspectives of vulnerability, configuration security and firmware security. The automatic system design greatly improves the efficiency of users to carry out the detection work. Through the design and implementation of power Internet of Things equipment security detection system, users can quickly and efficiently carry out network access security detection of power Internet of Things equipment. Based on this, it is decided whether to allow the Internet of things devices or systems to access the power system network, avoiding security risks and hidden dangers caused by the Internet of things devices entering the network.

In order to address problem about the channel phase error, a channel phase calibration method based on deep learning is proposed. Using data mining to replace the traditional method can not only improve the flexibility and stability of the method, but also achieve better results. Firstly, we use the frequency response function to model the channel characteristics, and the channel mismatch model is established to simulate the errors of the channel. Secondly, the error generated by the channel is introduced into the signal to generate the analog data set. Through the training and fitting, we achieved the all-phase calibration. At the same time, a variety of different channel parameters are simulated, and the generalization ability of different channel parameters get verified. Finally, the model network is evaluated in the form of test standard deviation. According to the results, the standard deviation can be controlled within 3°, which proves the effectiveness of this method. In this paper, Octave was used to generate the simulated data set for preprocessing, PyCharm platform was used to build the neural network, and the model was trained based on TensorFlow.

Aiming at the inaccuracy of Non-Local Means (NLM) algorithm for measuring the similarity of neighborhood blocks, an improved Non-Local Means denoising algorithm based on Difference Hash (dHash) algorithm and Hamming distance is proposed. The traditional algorithm measures the similarity between neighborhood blocks by Euclidean distance, so the ability to preserve edges and details is weak, which leads to the blurred and distorted images after filtering. To this end, the Difference Hash algorithm containing the gradient information is introduced, the difference hash images are generated from neighborhood blocks, and the Hamming distance of the difference hash images is calculated to measure the similarity of the neighborhood blocks. Finally, the Euclidean distance is improved. Experiment results show that the proposed method can preserve edges and details while denoising the low-noise images. Compared with other improved algorithms, the running speed of the proposed algorithm is also greatly improved, which has a certain application value.

Skin cancer, the primary type of cancer that can be identified by visual recognition, requires an automatic identification system that can accurately classify different types of lesions. This paper presents GoogLe-Dense Network (GDN), which is an image-classification model to identify two types of skin cancer, Basal Cell Carcinoma, and Melanoma. GDN uses stacking of different networks to enhance the model performance. Specifically, GDN consists of two sequential levels in its structure. The first level performs basic classification tasks accomplished by GoogLeNet and DenseNet, which are trained in parallel to enhance efficiency. To avoid low accuracy and long training time, the second level takes the output of the GoogLeNet and DenseNet as the input for a logistic regression model. We compare our method with four baseline networks including ResNet, VGGNet, DenseNet, and GoogLeNet on the dataset, in which GoogLeNet and DenseNet significantly outperform ResNet and VGGNet. In the second level, different stacking methods such as perceptron, logistic regression, SVM, decision trees and K-neighbor are studied in which Logistic Regression shows the best prediction result among all. The results proves that GDN, compared to a single network structure, has higher accuracy in optimizing skin cancer detection.

This paper illustrates the improved signal processing system architecture and processing flow for a two-dimensional (2- D) phased array digital multi-beam radar. 2D phased array digital multi-beam radar is required to achieve electronically scanning in both azimuth and elevation, which demands higher processing performance of the system. To improve the real-time performance of the system, we modify the signal processing system architecture. In this paper, we will focus on DSP to demonstrate the utility of the improved system architecture and processing flow. And a new detection strategy which combines dual-threshold variable index constant false alarm (DT-VI-CFAR) and 2-D Cell Average CFAR (CACFAR) is proposed. Practical validation shows that this detection strategy significantly reduces the complexity compared to traditional 2-D CA-CFAR, while improving the detection probability compared to traditional DT-VI-CFAR.

The defocus deblurring raised from the finite aperture size and exposure time is an essential problem in the shooting process, which seriously affects the quality of the images. However, studies based on defocus deblurring in monocular images yielded good results, while those on binocular images are rare. The current methods directly merge the left and right views regardless of their unique features. Objects within the camera’s DoF will not have a difference in phase, while light rays from outside the DoF will have a relative shift that is directly correlated with the amount of defocus blur. In this paper, we firstly proposed an enhanced multi-stage network for defocus deblurring using dual-pixel Images. Taking into account the parallax between the left and right views, the first two stages learn the information of them, respectively, and correct the deviation of the images under the supervision of the ground truth. The third stage consists of EERG and ERGS. It merges with the feature map of the previous stage, so that the left and right views are mutually enhanced, and a good restored image is obtained. ERGS uses the residual block as the basic unit to restore the details of the blurred area while maintaining the clear. Experimental results show that our proposed network can achieve better accuracy than state-of-the-art approaches on the public DPD dataset.

Task allocation of multiple unmanned aerial vehicles (multi-UAVs) is a typical NP-hard problem. In this paper, according to practical battlefield needs, mathematical model is constructed based on complex constrains of task allocation, and objective function is constructed based on multi-UAVs’ global voyage and task time. An improved strategy of particle position based on basic Particle Swarm Optimization (PSO) algorithm is applied to the problem, and reasonable allocation schemes are obtained. The allocation schemes meet the complex constrains including task sequence, time window, UAVs’ capacities and flight path, and can be chosen and adjusted flexibly by the decision maker according to the practical battlefield needs. A large number of simulation experiments show that improved PSO algorithm is effective and provides a reference for multi-UAVs’ task allocation problem with complex constrains and multi-objectives.

With the development of UAV technology, because of its advantages of low cost, easy operation and high flexibility, UAV has been applied to many fields such as transportation, patrol inspection, live broadcasting and so on. However, due to the problems of low rate, high delay and poor interaction, the task execution efficiency of UAV is not very satisfactory. In recent years, the rapid rise of 5g technology has brought another opportunity to the field of UAV. This paper proposes the application of UAV Based on 5g communication technology, which overcomes the current bottleneck of UAV. It provides a solution for the field application of UAV, and promotes the development of UAV.

Low-light images are generally produced by shooting in a low light environment or a tricky shooting angle, which not only affect people's perception, but also leads to the bad performance of some artificial intelligence algorithms, such as object detection, super-resolution, and so on. There are two difficulties in the low-light enhancement algorithm: in the first place, applying image processing algorithms independently to each low-light image often leads to the color distortion; the second is the need to restore the texture of the extremely low-light area. To address these issues, we present two novel and general approaches: firstly, we propose a new loss function to constrain the ratio between the corresponding RGB pixel values on the low-light image and the high-light image; secondly, we propose a new framework named GLNet, which uses the dense residual connection block to obtain the deep features of the low-light images, and design a gray scale channel network branch to guide the texture restoration on the RGB channels by enhancing the grayscale image. The ablation experiments have demonstrated the effectiveness of the proposed module in this paper. Extensive quantitative and perceptual experiments show that our approach obtains state-of-the-art performance on the public dataset.

The performance of facial action unit (AU) detection is often limited owing to the lack of annotated AU data and data imbalance. The data scarcity problem can be partially mitigated by abundant expression data, as AU detection and facial expression recognition (FER) are closely related. Accordingly, in this study, FER and AU detection are trained jointly in a multi-task learning framework, in which FER serves as an auxiliary task for AU detection by providing supplementary information. Meanwhile, we propose to use an attention gate unit between the two tasks to flexibly select valuable information from each other. To address the model bias issue caused by AU data imbalance, a smooth class-weighted loss is adopted to alleviate the dominance of negative AU classes. The best average F1-score obtained using our approach on the BP4D dataset is 63.5%, which is very close to the state-of-the-art performance and exceeds the single task baseline by 3.2%.

Voice Disability is one of the common disabilities experienced by children. Speech being the major mode of communication, it is important to rectify the voice-related problems at an early stage in life. Painful endoscopic techniques like laryngoscopy are used by doctors to identify the voice disability. In this work, an algorithm is devised to measure the severity of voice disability in children using signal processing techniques. Spectrogram and curve fitting techniques are used to detect voice disability. The normal and pathological curve fitted functions are passed through an adaptive signal processing system. Correlation between the normal function and tuned pathological function is obtained which is used to determine the severity of the disability. The reported work on this topic is language-dependent and uses machine learning algorithms that need large databases. In this work, adaptive signal processing techniques and the use of voice acoustic parameters are explored. Sound samples used are vowel sounds that are independent of the language and a range has been assigned to quantify the severity of the disability.

Considering Pulse Code Modulation-Frequency Modulation (PCM-FM) signal detect in high dynamic and low Signalto- Noise Ratio(SNR) situation，a low SNR spectrum-correlated detect method that is not sensitive to high dynamic situation is proposed. Firstly, this method uses the power spectrum of the known PCM-FM signal to match the power spectrum of the receiving signal, searches for the maximum filter output within the doppler frequency range, and constructs the detect statistics. Then, this method uses the probability of false alarm to compute detect threshold, and compares with the detect statistics to determine whether PCM-FM signal exists. The simulation result shows that the proposed method can detect PCM-FM signal quickly with high probability in high dynamic and low SNR situation.

The actual radar exposure area contains different types of landforms, resulting in non-uniform radar clutter, which significantly reduces the target detection performance and makes it difficult to maintain a constant false alarm probability. This paper proposes an enhanced minimum description length CFAR(EMDL-CFAR) based on median absolute deviation. The algorithm has good target detection performance in the clutter edge environment, in addition, also guarantees the detection performance in the multi-target environment. Using the insensitivity of the median absolute deviation to interference, select different reference samples after the clutter edge detection determines the clutter edge position, and then use the median absolute deviation(MAD) hypothesis test to eliminate the interference from the samples. The performance of the algorithm under different clutter backgrounds is evaluated through simulation, and the superiority of the algorithm is explained.

As low-resolution radar is still the main radar in service in China, the ground target classification and recognition technology of low-resolution radar has a wide application prospect in modern military and civil fields. This paper mainly studies and compares two main types of automatic target recognition and classification method for low-resolution ground radar: conventional recognition based on feature extraction and neural networks, and the conclusion is that the latter has better performance and needs less time to train.

The former model in this paper fuses the time domain and frequency domain features of ground target echo, then simulates, compares and analyzes the performance of different classifiers. The classifiers studied include: naive bayes classifier (NBC), decision tree classifier (DT), linear discriminant analysis (LDA) classifier, k nearest neighbors (KNN) classifier and support vector machine (SVM) classifier. Five-fold cross validation is adopted in the experiment to effectively avoid the impact of arbitrariness on the results caused by the random division of the sample set into training sample set and test sample set. Besides, based on conventional convolutional neural networks, a new neural network structure named multi-scale residual neural network (Multi-scale ResNet) is proposed for one-dimensional feature target recognition, which effectively reduces the data dimension through auto-encoder and solves the problem of performance degradation caused by the difficulty in training too many levels of traditional convolutional neural network. The bayesian hyper-parameter optimization method is utilized to optimize the hyper-parameters of different classifiersl. Finally, compared the accuracy of the two types of target recognition, the best performance of the pattern recognition is the support vector machine, which recognition rate is 91.2%, while multi-scale residual neural network recognition rate is up to 99.6%.

This paper proposes a novel 5-dimensional hyperchaotic coupling synchronization system for physical layer security in OFDM satellite systems. The proposed approach is designed based on the Lyapunov stability theory. The proposed chaotic-based satellite data encryption/decryption system is validated using a numerical simulation study. Additionally, to demonstrate the efficiency of the proposed chaotic encryption structure, we analyzed its key space. The proposed chaotic encryption structure is very sensitive to the initial key, and a tiny discrepancy as small as 10−19 would lead to a completely different sequence. The key space of the proposed scheme is up to 10³⁴⁰.

Cognitive communication countermeasure system utilizes artificial intelligence technology to quickly realize electromagnetic dynamic perception and electronic jamming strategy generation. In the complex electromagnetic environment of the modern battlefield, continuous phase modulation (CPM) signals are getting more and more attention due to high spectral efficiency and power efficiency. CPM signal denoising processing helps to improve electromagnetic dynamic perception performance. In this paper, a novel model, namely attentional denoising autoencoder (ADE), is proposed with enhanced signal denoising by introducing self-attentional mechanism into the autoencoder. The proposed method divides the one-dimensional communication signal sequence into fixed-size signal patches satisfying the same modulation law, and then utilizes the parallel computing of the self-attention mechanism to model the dependencies between the signal patches, and finally average pooling is used to synthesize the information of each signal patch to reconstruct the signal. The simulation results demonstrate that the proposed model is superior to other methods in terms of the denoising effect, and has a high degree of waveform recovery, which is helpful for the subsequent perception and processing of CPM signals.

The concentration prediction of mixed gases is crucial to the pattern recognition research of electronic nose (E-nose) systems. The response experiments of the E-nose towards the ethanol and propanol mixture with different concentrations are carried out. Five kinds of machine learning algorithms, including linear regression, support vector machine, K-nearest neighbor, random forest, and decision tree, are used for training the multiple output regressors to predict the content of each component simultaneously. The R² score, root mean squared error, and mean absolute error are used to evaluate the performance of these models. The relationship between prediction accuracy and concentration distribution has also been studied. The results show that the model based on the random forest has superior performance for forecasting the concentration of ethanol and propanol, with the R² score more than 0.98 in the 5-fold cross-validation. This study provides a significant inspiration for designing a multi-output regression model to realize the quantitative prediction of mixed gases by the E-nose.

Images taken in low-light conditions often have the problem of poor visibility. Besides inadequate lightings, different types of image quality degradation, such as a large amount of noise and color loss due to the limited quality of cameras and camera ISO setting, cause low quality of the captured image. However, directly amplifying the darkness of the lowlight image will inescapably bring into the pollution of the image. Therefore, the task of low-light image enhancement needs to kindle the dark regions and remove image degradation. To achieve this task, our work builds a Retinex theorybased neural network, which decomposes the input images into an illumination map and a reflectance map. Illumination map, representing the light information, is used for brightness adjustment, while reflectance map, representing the color information, is responsible for reconstructing low-light image into enhanced image with adjusted illumination map. However, there are few studies that notice the derivative of the image is used to solve the noise problem in Retinex decomposition and use spatial attention-based residual structures to increase the effect of light enhancement. For Decomposition sub-Network (Decom-Net), we purpose derivative features to alleviate the occurrence of noise in the reflectance map in the process of low-light image decomposition. For Illumination Enhancement sub-Network (Relight- Net), we use the Gaussian blur for reducing the problem of brightness enhancement degradation and build the Residual Spatial Attention Block (RSAB) to enlarge the volume and increase the capability of pixel-to-pixel mapping. Experiments are implemented to shows the effectiveness of our network, which improves the performance of previous methods on a large scale.

The method of intention recognition based on HMM for aerial target is provided in the paper. The three HMMs are constructed in order to recognize the attack intention, evasion intention and escape intention respectively. The “left-right” structure is chosen to establish the HMMs. The target sensor data: target speed、approach angle and distance from enemy target to our side , is utilized as the observation sequence of the HMMs. Here, the discretization results of target sensor data are the input to the HMMs. The accuracy of target intention recognition is over 80.0% in simulation experiments, which show that the method is available and effective.

Ensuring information security is of great significance to a satellite communication system. In this paper, a chaotic encryption scheme is proposed for a satellite communication system. Firstly, five chaotic sequences are generated by a five-dimensional chaotic system. And then, XOR, scrambling and interpolation are performed on the data in turn. The simulation results show that the bit error rate of this method approaches 0 when the SNR is greater than 18, and the bit error rate of the illegal receiver is as high as 35.3% when the correct key is unknown. Therefore, secure communication can be achieved.

Aiming at the feature extraction of motor imagery electroencephalogram (EEG) signals of four types, this paper proposes a new method combining discrete wavelet transformation (DWT) and common spatial patterns (CSP). First, DWT method is used to select the appropriate frequency band according to the frequency features of signals, and the energy mean of the selected frequency band signal is used as a time-frequency feature. Second, CSP method is proposed to solving double classification problem to solving recognition of four types signals problem and extract spatial features. Finally, fusion features are fed into the support vector machine (SVM) classifier and the classification accuracy reached 72.92%. The result is 6.95% better than using only the CSP method and 12.16% better than using only the DWT method, which verify the effectiveness of the proposed method.

This paper aims to repair missing regions which are corrupted along arbitrary directions. It presents a mixed image inpainting method based on Markov random field. By using Belief Propagation scheme in low gray-levels and alternately suggesting a directional Gaussian Graphical model (DGGM) for multiple references in a high-level range, it gains a balance between the model accuracy and the computation complexity in realization. On the basis of an existing method in [1], it improves the method in high level inpainting for the task under small train sets and large corrupted regions, by introducing these multi-head reference clues. Experimental results are given, the inpainting quality of different kinds of images with different sizes and contents under different parameter settings are compared in metrics of the peak signal noise ratio and the structural similarity index. The significance of parameter settings and the efficient computational cost could demonstrate the feasibility of this method.

Gaussian inverse Wishart probability hypothesis density (GIW-PHD) filter has been considered a promising algorithm for tracking an unknown number of multiple extended targets (MET) with ellipsoidal shapes. However, when the MET are close to one another with irregularly varying shapes, the tracking accuracy will degrade seriously due to the incorrect measurement partition. To address the problem, we propose a new multiple extended target tracking and classification algorithm based on the shape driven strategy under the framework of PHD. First, the B-spline curve technique is employed to estimate the irregular MET shapes, and then the shape features are extracted for improving the measurement partition and state update for the closely spaced MET. Finally, the MET are classified according to the estimated shape information and the Gaussian mixture implementation of the proposed algorithm is derived and presented in this work. Experimental results show that the proposed technique has a better tracking performance than the existing GIW-PHD for the closely spaced MET with irregular shapes.

According to the requirements of high resolution real time imaging of curved trajectory Bistatic Synthetic Aperture Radar (BiSAR), an improved RD algorithm is proposed in this paper. Firstly, the slant range of curved BiSAR is established by Chebyshev polynomials. Secondly, linear range cell migration (LRCM) and Doppler linear phase are compensated, and the high order approximate two-dimensional spectrum of echo signal is obtained by the method of series inversion (MSR) and Chebyshev decomposition. Finally, the focused image is obtained by matched filtering and phase compensation. By using Chebyshev polynomials to approximate slant range and spectral phase, the focus quality of BiSAR data is improved. Experimental results show that the algorithm can effectively compensate the motion error caused by three-dimensional acceleration and improve the imaging quality of the long distance edge point targets.

This paper studies and presents the problem of scale effect of multistatic sonar. To study the detection performance of the multistatic sonobuoy system, the track-initiation probability is used as the detection probability of the target. The monostatic distributed and multistatic distributed detection modes are taken as examples. The influence of sonobuoy arrangements on sonar detection area is analyzed. The results of experiments show the method is available and compared with monostatic sonar system, multistatic system has scale effect. Moreover, the scale gain increases when the arrangement mode changes from triangular and square to hexagonal layout. Taking into account the flexible deployment of the multistatic sonobuoy system and the easy node expansion characteristics, we can make full use of the scale effect of multistatic detection, so as to effectively improve the capability for detecting submarine.

Recently, deep learning has been widely used in image denoising. However, most of the existing deep learning-based methods are not adequate in blind denoising for additive white Gaussian noise (AWGN) images and real-world noisy images, which are still noisy or blurred. The difficulty is how to handle different noise levels and different types of noise with only one pre-trained model. In this paper, we propose a multi-scale adaptive feature enhancement network (MFENet) to improve the performance on blind image denoising. The MFENet is based on residual learning and batch normalization to speed up the network convergence. In the MFENet, dilated convolution and deformable convolution can expand the receptive field to obtain rich information from different scales. The deformable convolution is also able to adjust the sampling position to fit different shapes of objects. Spatial attention is used to enhance important features in the large amount of information. The experimental results show that the proposed method for blind denoising outperforms the state-of-the-art methods on both synthetic and real-world noisy images.

Advancement of digital signal processing and networking has raised many security and copyright concerns, thus it is very important to protect the authentication of digital data. In this work, an audio watermarking algorithm has been proposed which can be efficiently used for tamper detection and is also robust against reasonable attacks. Also, the watermarks are inaudible. The proposed algorithm can easily detect tampering as the watermarks are embedded at each frame without causing any audio degradation. In the proposed technique, first the audio signal is compressed using Graph Based Transform (GBT), for which watermarks are embedded into Line Spectral coefficients (LSFs) that are derived from linear prediction (LP) analysis with dither modulation-quantization index modulation (DM-QIM). Watermarks thus embedded in all frames are not only inaudible to the Human auditory system but also potentially provide robustness against meaningful attacks. This work also focuses on Blind tamper detection which is made effortless due to the proposed embedding algorithm. To measure the robustness of the algorithm, general processing of watermarked signals was done along with fragility testing. Quality of the audio was measured using Perceptual Evaluation of Speech Quality (PESQ) and Short-time objective intelligibility (STOI). The maximum PESQ score and STOI score of 2.8781 and 0.8150 respectively was observed without any attack on the audio signal. Tamper detection and quality measurement are the major contributions of this work. Detailed metric evaluation for attacks such as Scaling, Resampling, Filtering, Compression and Addition of White Gaussian noise (AWGN) has been computed and compared. The proposed technique makes tamper identification easier and gives framewise security.

Continuous Phase Modulation (CPM) signal has been widely used in modern satellite, mobile communication and military communication system due to its good spectrum and power utilization and constant envelope characteristics. The premise of demodulation or jamming of CPM signal intercepted in military communication confrontation is the accurate estimation of signal parameters. Aiming at the difficulty and complexity of multi-h CPM signal modulation index estimation, this paper puts forward a kind of blind estimation algorithm of the modulation index based on first order cyclic moment and the second-order cyclic cumulants, using the first and second order cyclic properties in frequency domain on the spectral properties to signal parameter estimation. The estimation algorithm is suitable for both single h CPM signals and multi-h CPM signals, where the multi-h CPM requires the synchronization of the guidance sequence. Simulation results show that the algorithm has better performance under low signal-to-noise ratio with less symbols required. When the number of symbols is 128 and the signal-to-noise ratio is 15 dB, the accurate recognition rate of the multi-index CPM signal can reach 98% and the recognition rate can reach 99% with the allowable error of 1/32.

With the rapid development of genome sequencing technology, a large amount of genome data has been generated, it also brings the storage problem of this massive data. Therefore, the compression of genome data has become a research hotspot. We propose a new genome data compression algorithm called LCMRGC (low memory consumption referential genome compressor) for FASTA format sequences. The algorithm uses the suffix array data structure to support the search of matching strings, and uses the binary search method to accelerate accurate matching, so as to obtain better compression ratio. Experiment results on standard genome data show that the proposed algorithm significantly reduces the memory requirement for program operation, and is competitive in compression ratio and compression time.

As a remote sensing instrument of rain drop size distribution, Parsivel disdrometer is widely deployed to meteorological observation stations. The microphysical characteristics of a severe precipitation affected by cold vortex are analyzed in this paper with the help of Parsivel disdrometer. After quality control of raw datasets, the precipitation process is classified as stratiform and convective precipitation. It is showed that convective precipitation has much larger values of liquid water content, rain rate and radar reflectivity factor than stratiform precipitation.

Gait is one of important features to assess elderly physical condition which is directly relative to health status. The changes or abnormalities in gait may reflect risks in health. Different kinds of gait analysis methods have been proposed, such as pressure sensors based method and wearable equipment based method. Recently, with the development of computer vision technologies, more automatic, effective and non-intrusive ways are demanded to measure gait at either nursing facility or at-home in daily environment. In this paper, we propose an automatic approach for non-cooperative persons for gait analysis using 3D camera. This approach applies a tracking based recognition method to identify the targets on captured videos. Then 3D skeleton based behavior analysis is performed to select skeleton series of walking from daily behaviors. Finally gait characteristics are defined and calculated on selected skeleton series of each identified person for gait ability evaluation. Evaluations have been performed on a real-world environment where people do not stop in front of the camera. The result shows that the accuracy of recognition and behavior analysis method reaches above 90% for multiple persons, which is better efficiency and comparable accuracy than the previous methods and our approach is suitable for gait analysis in daily environment.

To fully understand the energy consumption characteristics of 5G base-station, a DBSCAN-based energy consumption pattern clustering identification method is proposed for 5G base-station. Firstly, this paper analyzes the daily-curve characteristics of power consumption behavior in typical application scenarios of 5G base-station for further pattern clustering identification. Then, the proposed pattern clustering identification method is depicted based on DBSCAN (Density-Based Spatial Clustering of Applications with Noise) clustering decision, which is composed of the feature extraction for power consumption daily-curve of 5G base-station. Finally, the experiment is implemented using actual operation data of 5G base-station as data source. The experiment results illustrate that the proposed method can effectively identify the clustering characteristics of the energy consumption behavior for 5G base-station.

This paper applies the elliptic function to design graph filters, which can obtain arbitrary precision for step graph spectral responses. This method takes the mathematical form of the traditional elliptic analog filter, and its zero-pole points are recalculated. The approach obtains the coefficients of graph filters at a low computation cost by the polynomial multiplication rather than solving a nonlinear problem. Elliptic graph filters can control the ripples in the pass- and stop-band and width of transition band. Numerical experiments show the proposed approach outperformances the compared methods in designing the desired graph frequency responses.