Communications, Signal Processing, and Systems

Multi-object tracking task aims to identify and track all targets in the video. It has important applications in intelligent monitoring and other fields. Two problems can affect the accuracy of the multi-object tracking task. First, occlusion between targets will lead to interruption of tracking trajectory and switch of tracking target. Second, quality of the object detection results will directly affect the tracking accuracy. In this paper, we adopt a single-object tracking algorithm based on deep learning is introduced to solve the first problem and develop a discriminant network scoring the accuracy of detection and prediction bounding boxes to solve the second problem. The experimental results show that the proposed tracker performs better than other competing methods.

The power grid as the national manufacture steps toward combination with advanced information technology. The industrial control system of the power grid is exposed to the wide-opening internet with the industrial internet rapidly developing. Traffic classification is the significant step for security situation awareness platform which supervises the ICS operating status and suffers from the defect of low classification accuracy by conventional port-based or DPI methods. Therefore, the two-dimensional CNN model for the power grid industrial control traffic classification is proposed in this paper, which extracts the raw data’s features to train a two-dimensional CNN to fit the distribution of the features. The experiment’s result shows that this model can recognize and classify the ICS traffic accurately with an accuracy of 94%. Through the cross-validation, the result shows that this model also has outstanding generalization ability with an accuracy of 93%.

Cross-domain person re-identification (re-ID) is a challenging task in computer vision research. We intent to learn a discriminative model when given labeled source and unlabeled target datasets, in order to obtain good results on the target test datasets. In this paper, we add structure information for pedestrian images. To this end, we evaluate the influence of structure information for cross-domain re-ID in the deep networks and a series of experiments with different settings on Market-1501 and DukeMTMC-reID are conducted. The influence of structure information on the effectiveness in the optimized deep models is shown in the experimental results.

In this paper, we propose a deep model of course quality evaluation using deep neural network, which is simple and efficient to accurately evaluate the course quality. Specifically, we utilize multi-layer fully-connected (FC) layers with different activation functions to model the network. We perform a series of simulation experiments to evaluate the effect of FC layer number and activation functions. Simulation experiment results verify that the proposed model has excellent performance.

With the development and progress of society, air quality issues have a greater impact on mankind. However, the number of indoor environment monitoring instruments currently on the market is small and the price is high, and among them, the remote monitoring function is almost missing. This makes it difficult for people to obtain the air pollution index conveniently and in real-time. Faced with such a situation, a remote indoor environment monitoring system was designed to promote the realization of the dream of the interconnection of everything in contemporary society. Compared with mainstream air monitors on the market, this remote indoor environment monitoring system not only has an LCD, but the biggest highlight is the realization of WIFI wireless transmission. People can remotely control the instrument through mobile phone software to realize monitoring and feedback the monitoring results immediately. The design can also realize interaction with the cloud, upload height information to the cloud for data analysis and storage.

Cloud classification is an important and challenging task in cloud observation technology. For better classification, we present a method based on multi-task learning using multi-modal information. We utilize different loss functions to conduct multi-task learning. We implement a series of experiments on multi-modal ground-based cloud datasets for different tasks. Experimental results show that multi-task learning is effective for cloud image classification using multi-modal information, and it can improve the results of cloud image classification.

Heart disease has been present younger trend due to the surge of life and work pressure. Therefore more and more people begin to pay close attention to their heart condition, temperature, and motion information, in order to monitor their health. In medicine, ADS129R is used to collect, amplify and filter ECG signals, and LMT70 is used to measure body surface temperature. Based on STM32, 51 MCU, this paper displays the temperature information, the number of steps, the distance information, the ECG waveform and the heart rate processed and calculated by the algorithm on the LCD screen in real time. This paper establishes a complete set of wireless motion sensors, and uploads the data to the server through WIFI.

With the development of automatic parking system, path planning and path tracking have been the focus of attention. For parallel parking and vertical parking, this paper mainly studies the implementation of the strategy of the parking path planning and tracking algorithm. The fitting method based on a polynomial function is used to plan the optimal parking path reasonably, and the pure pursuit algorithm is used to track the path after planning. The feasibility of path planning and the effectiveness of path tracking are verified by simulation experiments. This paper is also of great significance to the field of intelligent driving.

Ordinary household medicine boxes store various medicines for emergencies, but their function is limited to medicine storage. Most elderly people need to take medicines for a long time due to physical reasons, but they often miss or forget to take medicines due to memory decline, decreased vision, and inconvenience, which causes great harm to their health. Based on this, this work has designed a “household intelligent medicine box based on the Internet of Things”. The work has the functions of medication time monitoring, medication quantification, medication classification, medication reminder, etc. Based on the results of the SolidWorks mechanical structure modeling, the Arduino UNO controller is used to control the precise rotation of the MG945 steering gear to control the output of the medicine. The user presses the button and the device automatically delivers the medicine. The work provides health testing services such as heartbeat detection and uses the voice module to remind you to take medicine at the time set by the timer. Also, according to the function of the medicine box, a matching client APP has been developed. By using the HC-05 Bluetooth module for intelligent control of the medicine box, the user can open the APP with a mobile phone, input corresponding instructions to control and insert medicines information. The APP can judge whether the elderly take medicine on time and set the time, type, and quantity of medicine by checking the feedback data of the medicine box. APP records the remaining amount of medicine and the time of taking medicine for the elderly by connecting to the remote database.

At present, the visual parking assistance system in intelligent driving generally has the problems of unclear parking image quality and high hardware cost. In order to reduce the difficulty of parking and improve the ability to adapt to the environment, this paper proposes a vehicle assistance system based on parking image enhancement. Firstly, Retinex algorithm is used to balance the image illumination information and enhance the color saturation, so that it can adapt to more complex environmental conditions; secondly, Ackerman steering theorem is used to draw the dynamic parking aid line, and the coordinate transformation technology is used to output it to the vehicle screen. The adaptability and effectiveness of the developed system are verified by the relevant experimental research.

GPS and star sensor integrated navigation is a new direction of joint positioning and attitude determination. Based on the integrated navigation method of Kalman filter, an attitude accuracy evaluation method of GPS/star sensor based on hybrid $$\chi^2$$ χ 2 detection is proposed in this paper. Aiming at the possible failure problem of the combined filter of GPS and star sensor in the actual scene, an attitude filtering accuracy evaluation method based on the hybrid detection of innovation $$\chi^2$$ χ 2 detection and state $$\chi^2$$ χ 2 detection is proposed. By constructing equivalent weight factors based on the hybrid $$\chi^2$$ χ 2 detection in Kalman filter, the attitude accuracy of GPS and star sensors can be evaluated, so as to realize error compensation in the case of failure and improve the anti-deviation performance of filtering. Experimental results show that the proposed method can effectively evaluate the accuracy and improve the anti-deviation performance, and has certain advantages compared with the standard Kalman filtering method in the case of failure.

At present, due to the improvement of people’s material living standards, the number of cars is growing rapidly. Urban road congestion and parking space tension, bring more and more safety problems. Intelligent parking has become a hot issue in assistant driving research. Therefore, this paper develops the open web interface of Amap to realize the function of automatically finding parking space and obtaining parking route. Then the edge detection operator is used to detect the parking space, and finally the Amap smart positioning is used to guide the driver to park. Users can query the real-time information of the use of parking space in the destination parking lot anytime and anywhere, and improve the success rate and accuracy of parking. It not only saves the parking time, but also improves the utilization rate of the parking lot. The real-time and effectiveness of the developed system are verified by the simulation experimental research.

With the acceleration of 5G commercial and new infrastructure construction process, Amplitude Probability Distribution (APD) measurement will be widely used in EMC testing of digital communication systems. The International Special Committee on Radio Interference (CISPR) has made requirements for APD measurement receivers in the new international standard for electromagnetic compatibility testing. By analyzing the mathematical model of APD, this paper completes the overall design of digital high-resolution APD receiver, and realizes the long-term statistical measurement of random interference with low probability. The research work in this paper lays a foundation for the development of high performance APD measuring receiver.

In the environment of Internet of vehicles ranging, the image sensor works for a long time, and the brightness is not enough or uneven when shooting, which will produce noise, resulting in poor image quality. In this paper, based on the LIF (Leaky Integrate-And-Fire) biological neuron model with noise immunity, NST (Noisy Softplus and Tanh) activation function is proposed and applied to PSM-Net to effectively overcome the influence of noise on image quality. The NST function is used to fit the response mechanism of LIF model, and the unknown $$\mathrm{\alpha }$$ α are added to make the NST function curve change with the change of input noise, so it has anti-noise performance. On this basis, the non monotonic property is added to the negative half axis to improve the processing accuracy of the function in the deep network. Experiments show that NST function is stable in the presence of noise input, and the error value is 13.504 lower than ReLU function in the presence of large noise input with variance value of 0.2, which effectively improves the processing accuracy of binocular stereo disparity prediction network.

In this paper, multi-spectral satellite images, Light Detection and Ranging (LiDAR) point clouds and thematic data are used for recognition and extraction of interference areas, and based on high resolution digital orthophoto map, use hue-saturation-value (HSV) or red-green-blue (RGB) color model to identify individual red-attack pine trees. Then compare the recognition results and accuracy between the different color models according to visual interpretation and field verification. Research shows that the recognition ac-curacy based on HSV color model is much higher than that based on RGB color model. At the same time, the collaborative utilization of multi-source data can greatly reduce the interference of dead grasslands, shadows, bare land, and roads, and reduce the error rate from 60% to 20%. This paper provides technical ideas and methods for accurate recognition of red-attack pine trees based on unmanned aerial vehicle (UAV) high resolution images.

Imbalance can affect the classification. In this study, a new algorithm to classify the pathology by considering the imbalance between classes has been built. This algorithm used principal component analysis (PCA), synthetic minority oversampling technique (SMOTE), and cluster membership degrees. Both PCA and Expectation-Maximization algorithms are used to give new features combined with the existing features. This proposed method is associated with Support Vector Machine (SVM) and Naive Bayes (NB) for severity classification of UA speech and TORGO speech databases. Another point in this work revealed the importance of articulatory of the TORGO database. The evaluation of this method on the two databases showed the significant results where the increase for TORGO database articulatory features, auditory features, and their combination was respectively 8.03%, 16.68%, and 17.79% compared to SVM performance and 3.04%, 23.67%, and 13.5% compared to the NB performance. The increase for UA speech was 5.13% and 24.63% compared to SVM and NB performance respectively. Additionally, the proposed method has outperformed four well-known imbalanced classification algorithms.

This study aims to design a simple and portable electrocardiograph (ECG) monitoring system. The key technologies are to develop the STM32 microcontroller and design the digital filtering of human ECG wave. At present, there is no simple sensor for direct amplification and filtering processing. Here, AD1292R preamplifier and STM32F1 series single chip microcomputer (SCM) are used in the design. And 24 bit ADC with high precision is adopted to collect ECG signals, and the filtering scheme is selected, so as to dynamically display the ECG and heart rate. Compared with other ECGs, this kind of ECG monitoring system has some advantages such as small volume and low power consumption.

With the increasing amount of remotely sensed data and the large-scale deployment of remote-sensing constellations. The problem of poor timeliness of remotely sensed data is becoming more pronounced. This paper constructs a double-layer satellite model for download quickly of remote-sensing constellation in the context of Tianlian satellites and space-ground integrated satellite systems. A joint scheduling algorithm based on relay satellites and inter-satellite links (ISL) of remote-sensing satellites (JSA-RISL) is proposed by integrating the QoS demand of remotely sensed data, energy consumption of satellites, and topology of satellite-ground links and other characteristics. Behavioral gaming and greedy transmission of emergency data through remote-sensing satellites. Improve the throughput of remote-sensing constellations. Download quickly of emergency data from remote-sensing satellites is achieved. The simulation shows that the JSA-RISL algorithm is suitable for the scenario of Tianlian satellite assisted download of remotely sensed data, which can effectively to improve the timeliness of emergency data.

The edge of the feature image contains rich data information, which is an important feature information of the image. The real-time display of the image is required in the actual system. In this paper, a real-time image filtering and edge detection system is designed by using Gaussian filtering and Sobel edge detection algorithm. The system is implemented on FPGA. According to the processing flow of image acquisition, image processing and image display, the function of real-time image display on the screen is realized. The image display effect under different threshold settings is compared, and the appropriate threshold settings are proposed.

Zero-Shot Action Recognition (ZSAR) is achieved by learning the mapping relationship between visual space and semantic space. Existing methods usually utilize the SOTA backbone network to construct the visual space. These methods have two major limitations. First, the human motion information that crucial for action recognition is easy to be confused in the background. Second, the key information which can reflect the correlation between actions may fall into oblivion, due to the redundancy in video sequences. In this paper, we propose an Attention-based Video Disentangling Matching Network (AVDMN) to solve the above problems. Specifically, we decompose segment-wise video into background stream and human motion stream by proposing a video disentangling mechanism. Furthermore, to further highlight the correlation between actions, we design an attention module to extract the key component of the above information. Finally, a relationship learning module is introduced to learn and measure the distance or similarity between video representation and action labels. Experiments on three realistic action benchmark Olympic Sports, HMDB51, and UCF101 show that the proposed architecture achieves favorable performance among ZSAR methods.

In radar sensor network (RSN), the waveform optimization of radar members is significant to the performance of target tracking. In this paper, we propose an intelligent waveform optimization (IWO) algorithm for RSN, which optimizes the waveforms of all radar members jointly through a global state error covariance matrix. The effectiveness of the proposed IWO algorithm is demonstrated by comparing with the distributed waveform optimization (DWO) algorithm that the radar members optimize waveforms independently. Experiment results illustrate that our algorithm achieves better tracking performance than that of DWO in RSN, in terms of tracking root-mean-square error and entropic-state.

In recent years, due to the increase in the number of ships, the channel environment has become increasingly complex, and virtual AtoN have been widely used as new high-tech beacons. In view of the setting of virtual AtoN in the harbour area. This paper presents an experimental analysis of the channel capacity and system blockage rate of the AIS system. Experimental results show that too many virtual AtoN can take up AIS communication resources, reducing channel capacity and increasing the likelihood of communication conflicts. Taking Shanghai port as an example, this paper analyses the maximum number of virtual AtoN that can be set under the conditions of the number of vessels in the region, which can be used as a reference for setting virtual AtoN in the harbour area. The specific number of beacons to be set should be adjusted according to the communication conditions and the waterway situation. The reasonable setting of virtual AtoN can reduce the impact on AIS communications and better utilise their navigational advantages.

Multimodal wireless sensor networks (MWSNs) consist of variable types of sensor nodes that can do many important tasks, such as target tracking, health care, and environment monitoring. Besides, as a typical application of flocking algorithm in mobile sensor networks, the coupling target and tracking (CTT) provides information on how to effectively utilize the mobility of sensor networks. In this paper, we construct a MWSNs model, using a fuzzy logic based flocking control scheme that run on a distributed Extended Kalman filter algorithm (EKF) for collaborative tracking of a target. The scheme based on fuzzy logic considers both the tracking performance and detection performance of each sensor comprehensively to obtain an accurate estimation of the target position, which is used to lead the tracking process of target. The simulation results show that under the proposed scheme, the sensor can form a stable tracking formation with low tracking error on the target in a short time, ensure support high-level applications.

Accurate damage assessment for orbital targets is of great significance for combat mission planning. Compared with infrared and other detectors, radar has the advantages of longer detection distance and less influence by weather. The damage assessment using radar detection information has certain application value. This paper uses theoretical modeling methods to extract damage features from the target’s motion, background radar plot, radar cross-sectional (RCS) and high resolution range profile (HRRP). A Bayesian network is used to fuse different types of damage features, and the performance of the proposed damage features is verified in the numerical experiment. The result shows that the target RCS and HRRP changes are the most effective features for damage assessment of orbital targets.

In this paper, according to international standards, referring to CASIA Chinese phonetic library, recording the phonetic library based on Henan dialect. And use computer software to reduce noise. Use Matlab to preprocess the speech signal such as framing and windowing, and extract the short-term energy, formant frequency, pitch frequency, Mel cepstrum coefficient, and other characteristic parameters contained in the speech signal. Use KNN, BP neural network, PNN neural network, and LVQ neural network to recognize and process recorded dialect. It can be seen from the recognition rate of the LVQ neural network is the highest. For the five emotions, the recognition results of various algorithms are different.

As a primary tool in reinforcement learning, learning automaton (LA) has been widely applied in the area of electronic engineerings such as signal processing and wireless communications. The scenario that an LA operates may have a large number of actions. The existing LA schemes, which are designed in a fixed manner, are inefficient in massive-action environments. In this paper, we propose an efficient LA (ELA) scheme based on a novel learning framework. The well-designed framework includes a brand-new action set adjustment strategy, a convergence judgment strategy, and an exploration-exploitation strategy. Experimental results in various massive-action environments indicate that the proposed ELA scheme achieves significant improvements in both convergence rate and convergence time, especially in environments with an extremely large number of actions.

This paper presents a fault estimation and compensation problem for T-S fuzzy model in term of a descriptor system approach. Set the fault as a state, then a new system named augmented descriptor system is established. An observer is given to show state and fault’s estimates, simultaneously. Fault compensation is also obtained according to the estimated sensor fault information above, which make this system can normally work when faults occur. A numerical example is used to prove that this method has effectiveness.

Since 2019, the sudden outbreak of COVID-19 has made huge impacts on various aspects of society, especially the financial industries that are closely related to the national economy and people’s livelihood. Finance is a data-intensive field and its traditional research models include supervised and unsupervised models, state-based models, econometric models, and stochastic models. However, the above models are prone to lose their effectiveness in the situation of an extremely complex financial ecosystem with a large number of nonlinear unpredictable effects, such as those caused by COVID-19. To address this issue, we comprehensively explore and fuse Stochastic Block Model (SBM) and Cox Proportional Hazards Model (COX) for a reliable and accurate financial risk prediction. Specifically, SBM, which is popular in social network analysis, is employed to capture the impact factors on the financial industry in public emergencies, and COX is then leveraged to determine the duration of the impact factors. An extensive experimental evaluation validates the effectiveness of our framework in predicting financial risk.

The worldwide spread of COVID-19 has greatly hit global economy by now. The world’s major economies including both developed and developing countries have felt the resulting impact on their financial markets. Accordingly, learning residents’ consumption structure is significant for boosting consumption demand and recovering financial market. In this paper, the Extend Linear Expenditure System (ELES) model is explored to learn both urban and rural residents’ consumption structures of China during COVID-19. In specific, the indices of marginal propensity to consume, income elasticity of demand, and price elasticity can be yielded via the ELES model based on the disposable income and the consumption data. Furthermore, the consumption structures before and during the corona virus epidemic can be quantitatively compared. Extensive experimental results demonstrate that the epidemic has made profound impacts on the consumption structure of residents. Among them, the marginal propensities on food and medical services have greatly increased, while the proportions of other expenditures have been decreased.

In this paper, the problem of fault estimation (FE) for discrete T-S fuzzy system is considered. An observer-based fault estimator is designed by applying the extended state method and the error dynamic system is asymptotically stable and the disturbances’ effect meets the $$H_\infty $$ H ∞ performance. Sufficient conditions and observer gains for the FE problem are shown. An example shows the feasibility of the proposed FE scheme.

Marine Scotland is tasked with reporting on the environmental status of Scottish marine waters, an enormous area of water extending from the shoreline to deep oceanic waters. As one of the most important variables, chlorophyll concentration (Chl) plays an important role in the seawater quality monitoring. Currently, the Chl observation is mostly done by expensive ship-based surveys that have very limited spatio-temporal coverage. Satellite based ocean colour remote sensing has the potential to significantly enhance monitoring capabilities but this opportunity has not been widely adopted by statutory reporting bodies across Europe due to concerns over satellite data quality. To break through this bottleneck, in this paper, we explore to implement advanced machine learning techniques to automatically estimate the Chl via the historic time series of ocean colour remote sensing data during from July 2002 to September 2019.

With the acceleration of the aging process of our country's population and the rapid development of the urbanization process, chronic diseases have gradually become the main factor affecting the health of residents [1]. The emergence of wireless body area networks (WBANs) provides a simple and low-cost method for the health monitoring of patients with chronic diseases [2–4]. The efficient utilization of node energy resources is the key to the practical application of WBANs. In this paper, we propose a collaborative routing algorithm based on data priority. The algorithm allocates the maximum number of hops for this data transmission according to the priority of the data transmission, and then finds transmission path which consumes minimum energy resource. So as to achieve efficient utilization of energy resources. Finally, through the simulation of the maximum number of hops obtained by the priority of the transmitted data, it is concluded that a fixed node with multiple maximum number of hops has a longer life time.

Icing on a large area of high-voltage lines brings damage to the power grid, and many countries in the world suffer from it, and China is one of them. However, the current de-icing methods of high-voltage lines mostly have problems of low efficiency and low safety rate. In the analysis of the advantages and disadvantages of the high voltage line inspection work, the status quo and development trend of the high voltage line patrol robot, and on the basis of adopting other methods to clean the ice, this paper designed a high voltage line de-icing robot based on STC89C52 microcontroller. The robot can complete the functions of walking on the high voltage line, de-icing and real-time data monitoring through Bluetooth communication, SMS communication and self-detection. Through laboratory test and adjustment, the expected function has been realized well.

FlexRay bus is widely used in applications due to its efficient network utilization and system flexibility characteristics, but variations in physical transmission characteristics in complex usage environments can lead to unstable communication quality. In this paper, the test methods for FlexRay physical layer testing are analysed, test evaluation items such as insertion loss, return loss, crosstalk, characteristic impedance and eye diagram are proposed, a test equipment is designed and the metrics are verified.

In order to achieve the accurate measurement of pipeline’s end position, a method based on centerline vector with the help of multi-vision is proposed. This method uses pictures taken by eight industrial cameras to extract the pipeline’s centerline. And location of the center of end plane is determined with the aid of two data, the centerline vector and the pipeline’s outer diameter. The deep learning method is used for obtaining the camera calibration matrix which is used to reconstruct the space coordinates of endpoints. The relative position of two endpoints is calculated. The measurement experiment shows that the maximum relative deviation between the measured value and the standard value of this method is 0.6%, and the method is simple and measurement efficiency is high.

The performance of deep learning in the field of computer vision is better than traditional machine learning technology, and the image classification problem is one of the most prominent research topics. The methods of computer vision are used in industry production. We proposed an image classification model which segments the image at different scales on the basis of Deep-ViT to obtain image information of different scales. When this model is applied to the classification of tube head shapes, the accuracy rate can reach more than 90%, and towards the classification of tube head materials, the accuracy is about 98%.

Condenser tube is a common component with a wide range of applications. During the production process, there are unqualified samples with defects on the surface. It is of great significance to realize the automatic and efficient detection of the condenser tube’s defects. In view of the thread characteristics of the metal condenser tube’s surface and the characteristics of high-speed movement on the production line, we build an image acquisition device including three cameras and LED tubes. Then, we apply the classic neural network Resnet50 to extract low-, middle-, and high-level features of the condenser tube image. The trained model finally realizes the binary classification of the condenser tube image (images with defects vs. normal images). The result of the experiment shows that the precision value reaches 81% and the recall rate exceeds 94%. The detection speed is about 0.011 s/frame. In addition, the image acquisition device is used to shoot the same defect multiple times when the tube is moving. So there can be multiple consecutive detection results for the same defect in time series, which greatly improves the detection accuracy.

Fault research is systematic engineering that mainly contains fault detection, identification, and fault handling phase. A novel approach that combines the application of machine learning in anomaly detection with Fault Tree Analysis (FTA) organically to determine the characteristics of anomaly has been proposed previously. The presented solution of this paper that focuses on the prediction and processing of fault events is the subsequent work of the existing framework. We leverage the modular decomposition method to segment the fault tree, then take advantage of the Binary Decision Diagram (BDD) and Markov switching process to calculate the occurrence probability of the top event in the fault tree. Finally, we apply it to the safety design of the Integrated Modular Avionics (IMA) core processing system to verify the effectiveness of our application approach, and provide appropriate redundant IMA core processing system architecture for different safety levels of functions and ensure the system design meets the aircraft safety requirements.

Deep Learning algorithms based on convolutional neural network (CNN) have achieved huge success in semantic segmentation. However, in these networks, sub-sampling will cause a large loss of image detailed information. In this work, we design a novel method for recovering some of the lost pixels. We use two-dimensional discrete Wavelet Transform (DWT) to extract image boundary detailed information and combine the segmentation result of the convolutional network to recover some of the lost details. We analyze the influence of the algorithm parameters and wavelet function on the final prediction. In our experiments, our algorithm has accuracy improvement compared to the deep network on the PASCAL VOC 2012 dataset.

This paper focuses on synchronization for complex-valued shunting inhibitory cellular neural networks (SICNNs) with distributed delays and designs a novel feedback controller to ensure module-phase synchronization. For the discussion of module-phase synchronization, a lemma is given to show the existence of the bounded solution of the drive system. By constructing a Lyapunov functional and employing the inequality technique, sufficient conditions for module-phase synchronization of complex-valued SICNNs are derived. Finally, the validity of obtained results is demonstrated by a numerical example.

In order to evaluate the influence of hydraulic oil and hydraulic transmission oil on the low-temperature cold start performance of the winch, a hydraulic winch low-temperature cold start performance evaluation test bench was developed. The overall structure and working principle of the test rig are described, and the key subsystems of the test rig are designed, including the ultra-low temperature freezing chamber subsystem design, and the motion control and data acquisition subsystem. The cold start performance and working parameters of the hydraulic winch were analyzed and measured on the developed test rig, and the overall performance of the test rig was tested and evaluated. The results show that the test bench can simulate the actual operating conditions of the hydraulic winch, and the bench has good stability, and can evaluate the low-temperature cold start performance of the hydraulic oil.

Due to the urgent needs of military electronic products, especially space technology and new defense equipment for primary batteries with high mass and energy density. In this paper, the electrical properties of metal-air battery and lithium primary battery were tested. The results show that in terms of output stability, the voltage of lithium primary battery is very stable, and the voltage fluctuation of metal-air battery which needs electrolyte is generally fierce. In terms of mass power density, 1 # and 2 # lithium-carbon fluoride batteries have obvious advantages over metal-air batteries.

Since the use location of man-portable photovoltaic power supply, field mobile photovoltaic system and other equipment will change at any time, the impact of light incidence angle on its power generation capacity is extremely significant. This paper mainly studies the volt-ampere characteristics of solar cells of two material systems, thin silicon and copper-indium-gallium-selenide, under different incidence angle conditions, and the results show that: with the increase of light incidence angle, the open-circuit voltage of the various types of solar cells tested decreases slightly, but the change is not obvious; the short-circuit current, maximum working power and photoelectric conversion efficiency decrease more and more rapidly, and when the light incidence angle is in the range of 0°–30°, the changes are not significant.

Due to the urgent needs of primary batteries with high mass and energy density the search for a resourceful and environmentally friendly green energy source is a pressing issue. In this paper, the electrical properties of metal-air battery were tested. The results show that in terms of output stability, the voltage fluctuation of metal-air battery which needs electrolyte is generally fierce. In terms of mass power density, 1 # and 2 # zinc-air batteries have obvious advantages over other metal-air batteries.

Communication is a high-speed communication method that uses optical fiber as the transmission medium and optical wave as the carrier wave. Compared with other communication technologies, communication has many outstanding advantages: large transmission capacity, long transmission distance, high security, strong anti-interference ability, etc. However, due to factors such as material and temperature, it can lead to the drift of the bias point of its transmitter modulator, thus causing the instability of communication. Therefore, by compiling and synthesizing the bias point stability control methods of MZ modulator, this paper hopes to play a positive role in further research on the bias point stability control in the future.

The detection range of the ground-based unmovable single radar is limited, which is not enough to cover the weather system above the meso-β scale, and can not track the convective system moving from one radar detection area to another. When observing with a single radar, many problems will occur due to its geometric reasons, such as static cone region, beam broadening, beam height, and beam blocking. In the regional early warning and forecast, the role of a single weather radar has some limitations. In this paper, the data of multiple radars are unified into a unified coordinate system, and the vertical linear interpolation method (NVI) is used to grid the data of multiple radars. Finally, the distance-weighted method is used to fuse the spliced grid data to realize the high-precision weather radar network mosaic. The networking experiment with eight Doppler weather radars in Sichuan shows that the networking mosaic algorithm based on vertical linear interpolation method and distance weight fusion method can achieve high precision and reliable mosaic effect, which is beneficial for the better observation of the mesoscale weather system.

In order to study the effect of low temperature on the discharge characteristics of high energy density ternary lithium batteries for UAV, two types of ternary lithium batteries, energy type and power type, were used to conduct discharge tests at 25 ℃, 0 ℃ and −25 ℃ with energy recovery type charge/discharge test equipment. The test results show that the energy density of energy type ternary lithium battery can reach 266.11 Wh/kg at room temperature, 237.38 Wh/kg at 0 ℃ and 180.20 Wh/kg at low temperature; the energy density of power type ternary lithium battery can reach 244.72 Wh/kg at room temperature, 221.13 Wh/kg at 0 ℃ and 181.20 Wh/kg at low temperature. The energy density at low temperature is 181.57 Wh/kg.

With the increase of age, the elderly often fall, which seriously threatens their lives. Investigations and studies have shown that timely assistance after a fall can reduce the risk of death. So this paper proposes a fall recognition method based on the human skeleton in the video. First, the video data set is passed through the human pose estimation algorithm to obtain a new video data set containing human skeleton information and annotated. Then build a two-class model (falls and non-falls) based on the three-dimensional convolutional neural network to train and test the fall behavior model. The experimental results show that the method has high accuracy and recall rate and has good application value.

Regarding the location of multi-type barcodes in high-resolution images, it is difficult for the existing barcode positioning algorithms to meet the actual needs of efficiency. In this work, a reliable localization framework is proposed to locate multiple types of barcodes in high-resolution images, which covers two steps: extracting multiple types of barcode features through a joint edge detection algorithm and marking the target barcode region with a bidirectional contour labeling method. The result of the target location is represented by the smallest rectangular fit of the barcode region. According to the experimental results, the proposed method can locate multi-type barcodes in high-resolution images with an accuracy of 99.06% and a speed of 2,906 ms.

Autonomous obstacle detection technology is the premise to realize autonomous navigation in the Marine environment. The fusion sensing of multi-modal sensors can improve the accuracy and stability of obstacle detection. In this paper, on the premise of analyzing and comparing the performance of sensors, two sensors with different modes are selected to be used, specifically, LIDAR sensor and vision sensor. The system selects the unmanned vehicle open source framework MOOS as the development framework, and realizes the real-time reading of webcam video stream through RTSP and multi-threading. At the same time, the real-time data acquisition of LIDAR is realized through UDP communication and multi-threading architecture. Finally, a set of multi-modal data real-time acquisition system is developed. It has the characteristics of high localization rate, low cost, easy expansion and stable performance, which lays a foundation for the subsequent research on multi-modal data fusion technology.

Considering the rich spectral and spatial information of multispectral image, a learned multispectral image compression method named spectral-spatial feature completely separated extraction with tensor CNN is proposed. In encoder, two parallel modules, spectral module and spatial module, are applied to extract spectral and spatial features respectively. Then in order to reduce the loss in the downsampling process, two tensor layers that integrate tucker decomposition and neural network are utilized to decompose the multiway features. The decomposed feature tensors are concatenated along the channel dimension, saving the spectral and spatial features separately. To recover the features, the concatenated tensor is split in symmetric decoder after quantizer and entropy codec. Finally, post-processing module is utilized to remove the compression artifacts of reconstructed image fused by recovered feature tensors. Besides, rate-distortion optimization is embedded to balance the trade-off between the rate loss and the distortion. Experiments on eight-band dataset from the WorldView-3 satellite demonstrate our proposed method outperforms JPEG2000 and 3D-SPIHT in PSNR.

Multispectral image is one of the information carriers, which contains rich spatial and spectral information, and uncompressed multispectral images pose huge challenges for the storage and transmission of information. With the appearance and development of traditional image compression methods such as JPEG, JPEG2000 and 3D-SPIHT, image compression has exerted its performance to the extreme but brings block artifact and algorithm complexity. With the rapid development of deep learning, a deep inter-spectrum prediction network is proposed in this paper using the strong spatial and spectral correlation of multispectral image, which can enhance the reconstruction image quality. In proposed framework, the decoder ‘D’ uses spatial information extracted by the conditioning network ‘Cond’ and the spectral information extracted by the inter-spectrum encoder ‘E’ to predict original image. Our proposed approach is compared with JPEG2000 and 3D-SPIHT in objective evaluation index PSNR. And the experimental results show that our proposed framework have the significant quality improvements.

Considering that hyperspectral images have lavish spectral information, they also have spatial correlation information, a spectral-spatial unmixing method based on two-dimensional convolution autoencoder is proposed in this article. The hyperspectral image first passes through the convolutional layers to obtain the spatial features, and then concatenated to itself to retain the spectral information, at this time, a spectral-spatial tensor is acquired, finally it passes through the fully connected layers to realize the linear unmixing of the hyperspectral image. For the loss function, introduce reconstruction error constraints and abundance sparsity constraints to ensure that the abundance result has physical meaning. Experimental results demonstrate that the proposed method is superior to other sparse algorithms in terms of SRE results and visual effects.

The multispectral imaging which are used for remote sensing imaging has a large amount of data, so this paper proposed a deep learning method which is based on sliced convolutional LSTM for multispectral image compression. Compared with other algorithms, the proposed algorithm further compresses the multispectral images by considering the similarity between the spectra and removing the inter-spectral redundancy. The proposed algorithm is based on end to end framework which is consist of encoder, decoder, entropy coding and quantizer. In experiments, the PSNR of proposed model is compared with that of JPEG2000 to evaluate the performance of our algorithm at several different bit rates.

Altarica3.0 is a high-level modeling language for security analysis, and its basic mathematical form is Guardian Transformation System (GTS). At present, GTS (or Altarica3.0 model) can be compiled into a fault tree or critical event sequence. There are several reasons for turning GTS model into a fault tree: automatically generating a fault tree from a high-level model is easier and less time consuming than creating one from scratch; Second, the high level model greatly improves the model design, sharing and maintenance; Finally, the evaluation tools of the Boolean model are more effective than the state/transition model. In general, however, the cost is the loss of sequence between events: the sequence of events is compiled into the connection of events, so there may be potential omissions throughout the system reliability analysis process and analysis results. To remedy this problem, in this paper we outline an analysis approach that converts failure behavioural models (GTS) to temporal fault trees (TFTs), which can then be analysed using Pandora a recent technique for introducing temporal logic to fault trees. The approach is compositional and potentially more scalable, as it relies on the synthesis of large system TFTs from smaller component TFTs. We verify, by using a redundant system, the effectiveness of the proposed method.

Model simulation is a research method that can discover and improve system design problems through various experiments without the participation of the actual system. In the field of security critical systems and aeroelectronics, a series of security problems are often caused by large and complex system requirements. In this paper, a simulation framework for VRM formal requirement model is proposed to find and solve the problems that occur during the design phase of system requirements. The work includes: comprehensively analyzing each component of VRM model, relying on SysML model to build simulation sequence, executing simulation event propagation system constraints, and displaying system model status in real time. The framework solves the ambiguity of natural language requirements by using VRM formal requirement model, and designs some SysML modeling methods to help emulators build and refine simulation behaviors through case diagrams.

Due to the scale and complexity of the embedded airborne software in the aviation field, the modules of the airborne software will influence each other in ways unexpected by the designer under certain specific circumstances. Therefore, software coupling analysis has become an indispensable part of the airborne software design process. At present, most of the coupling analysis in the aviation field is carried out on R-BT (Requirements-Based Testing). The VRM model is a formal requirement model, which can be well modeled at the system requirement level. According to the definition of coupling and the requirements of coupling analysis in DO-178C (Airworthiness Standards for Civil Aviation Airborne Software), this article provides a set of coupling definition and coupling degree measurement methods based on the VRM (Variable Relation Model). At the same time, the coupling analysis of the flight guidance system in the field of avionics under this set of methods is given to verify the feasibility of this set of methods. According to the coupling definition and coupling degree measurement method proposed in this paper, a set of prototype tools supporting coupling analysis has also been developed.

AltaRica 3.0 is a high-level modeling language for safety critical systems, which can be used for system modeling and safety analysis. At present, there are some tools for AltaRica 3.0, such as OpenAltaRica, which can realize system modeling and step simulaiton. However, they can’t support graphical modeling and display. Therefore, this paper aims to design and implement a modeling and simulation tool with graphical modeling function based on AltaRica 3.0. The main work includes the following aspects: Firstly, a text editor for modeling is constructed. Fault model compilation, fault tree generation and fault tree analysis are implemented based on OpenAltaRica’s engines and Arbre Analyste tool. Secondly, a graph editor is constructed, and the functions of loading, modifying and saving graph models are realized. Then, a simulation method based on step simulation engine is designed and implemented. Finally, the correctness and effectiveness of the modeling and simulation method are verified through the analysis of an example system.

As the scale and complexity of the system tend to be increasing, model-based safety assessment (MBSA) has gradually become a research hotspot in system modeling and analysis. This paper independently designs and implements a model-based system safety analysis tool, and carries out a case study with landing gear system (LGS) as an example, including system hierarchy modeling, component interaction modeling and fault behavior modeling for LGS using AltaRica 3.0, as well as system simulation analysis and fault tree analysis for LGS on the platform. Finally, it is proved that the effectiveness of the tool and the accuracy of modeling and system safety analysis using AltaRica 3.0.

By grouping pixels with visual coherence, superpixel algorithms provide an alternative representation of regular pixel grid for precise and efficient image segmentation. In this paper, a multi-stage model is used for sea ice segmentation from the high-resolution optical imagery, including the pre-processing to enhance the image contrast and suppress the noise, superpixel generation and classification, and post-processing to refine the segmented results. Four superpixel algorithms are evaluated within the framework, where the high-resolution imagery of the Chukchi sea is used for validation. Quantitative evaluation in terms of the segmentation quality and floe size distribution, and visual comparison for several selected regions of interest are presented. Overall, the model with TS-SLIC yields the best results, with a segmentation accuracy of 98.19% on average and adhering to the ice edges well.

A new method for automatic liver tumour segmentation from computed tomography (CT) scans based on deep neural network is presented. Two cascaded deep convolutional neural networks are used to segment the CT image of the abdominal cavity. The first U-net is used for coarse segmentation to obtain the approximate position of the liver and tumour. Using this as a prediction the original image is cropped to reduce its size in order to increase the segmentation accuracy. The second modified U-net is employed for accurate segmentation of the actual liver tumours. Residual modules and dense connections are added to U-net to help the network train faster while producing more accurate results. In addition, multi-dimensional information fusion is introduced to make the network more comprehensive in restoring information. The Liver Tumour Segmentation (LiTs) dataset is used to evaluate the relative segmentation performance obtaining an average dice score of 0.665 based our method.

Based on the self-heating reforming hydrogen production technology, a hydrogen production prototype is trial-produced in this paper through the functional and integrated design of the regenerator, combustion chamber, catalytic bed, purifier, control execution module. The test results show that the maximum hydrogen production capacity of the hydrogen generator is close to 80 slpm, and the hydrogen purity is over 99.99%, of which the CO impurity content is 0.07 ppm, which meets the relevant national standards for hydrogen used in fuel cells. The research results have important reference value significance for promoting the development of methanol reforming hydrogen production technology.

There is no complicated valve and timing mechanism in the micro gas turbine with a total power density of about 0.8–1 kW/kg and more than 31% power generation efficiency. In actual use, the comprehensive power generation efficiency is able to increase to more than 80% if the waste gas can be reused through cogeneration technology, which can increase the. In addition, due to the characteristics of compact mechanical structure and durable work, the micro gas turbine has strong application prospects in civil and military fields. This paper mainly states the development status of micro gas turbine generation technology at home and abroad.

This paper studies an anisotropic interpolation model that can fill in-depth data in a largely empty region of a depth map. We suppose the image is endowed with an anisotropic metric $$g_{ij}$$ g ij considering spatial and photometric information. The interpolation model is the biased Infinity Laplacian or biased Absolutely Minimizing Lipschitz Extension (bAMLE). Given an implementation based on the “eikonal” operator, we investigate different metrics to improve the performance of bAMLE in the depth completion task. To perform this completion, we suppose we have a reference color image and a depth map at our disposal. We used first a standard metric that considers spatial and photometric differences, second, the same metric but using different exponents in each term, and third a new metric that considers a balance term between intensities and gradients of the image. Using this new metric, we outperform our version of the model and other contemporary models in KITTI dataset.

In order to study the rheological properties of lubricating grease, poly - ɑ olefins (PAO) synthetic oil as base oil and 12 - hydroxy stearic acid lithium soap as the thickener were used to prepare lithium lubricating grease. The effects of temperatures on the thixotropic ring areas, storage modulus, loss modulus, strain amplitudes and apparent viscosities of lithium grease were investigated, and the mechanism was discussed accordingly. The results showed that lithium-based grease was a yield pseudoplastic fluid, and its apparent viscosity was dependent on shear rate and time, which exhibited a typical viscoelasticity. With the increase of temperature, the structure of lithium grease became unstable. When the temperature was higher than 100 ℃, the structure of lithium grease will be destroyed seriously, and its performance will change significantly. At 75 ℃, the viscoelasticity of lithium grease was remarkable; At 0 ℃, the lithium grease showed an obvious anti-thixotropy.

We propose a new structure for multilevel pulse amplitude modulation (PAM-4) signal generation for on-chip optical interconnects and data center networks. Our proposed architecture use only one 3 × 3 multimode interference (MMI) Based ring resonator with two segmented phase shifters. Two segmented phase shifters applied the plasma dispersion effect in silicon waveguide are used in the feedback ring resonator waveguide. The structure provides a very steep slope compared with the conventional structure based on Mach Zehnder Interferometer (MZM). As a result, an extreme reduction of power consumption is achieved. Based on this structure, an extreme high bandwidth and compact footprint are also achieved. The device is designed and analyzed using the Beam Propagation Method (BPM) integrated with the Finite Difference-Time Difference (FDTD) simulations.

Compared with the inversion of large-scale continuous wind field, the inversion of small and medium-scale wind field is more difficult. Through the inversion and analysis of small and medium-scale wind field, many disastrous weather can be predicted. It is of great significance to study the formation mechanism of small and medium-scale weather system and the complex structure of atmospheric boundary layer. The GUI interface of MATLAB is used to set the parameters of wind field simulation. Then the radial velocity fields of various small and medium-scale wind fields are simulated according to Rankine mode. Finally, use NIVAP (integration method in natural coordinate system) and EVAPTC (EVAP method for tropical cyclone inversion), wind field inversion methods to invert the simulated wind field, and compare the similarity coefficients of the two inversion results. Through comparison and analysis, we can draw the following conclusions: both these two inversion methods can get better inversion results, but EVAPTC method is slightly better than NIVAP method.

Non-invasive load monitoring technology is to decompose the total load electricity information, get the information of individual load, and then extract the characteristics of these information respectively. After getting some unique features, you can identify what the load is. This technique is a great improvement over traditional intrusive monitoring techniques. This article mainly adopted factor hidden Markov model for load decomposition, using light REDD data set, refrigerator, washing machine, dishwasher and microwave oven, the five electrical data, on the median filtering methods for data preprocessing, after the total load modeling, obtained FHMM model, then for each load clustering, after that, Viterbi algorithm is used to determine the state of each load at each time, then a new coding method is used to code the state of the corresponding load. Finally, the accuracy of the results is calculated, and the conclusion is drawn that the accuracy of the on-off load is the highest, the accuracy of the finite state load is the second, and the accuracy of the continuous variable state load is the lowest.

This paper introduces the AIS system of HY-2C satellite, especially focusing on the electromagnetic compatibility design, Baseband Design and capture algorithm design of the AIS system. The application results of the AIS system in-orbit show that it is reasonable design and easy to operate. Also the AIS system can effectively operate under the complex environment of EMC, which can provide references for subsequent satellite-borne AIS.

A new type of small potato combine harvesting machine, which is suitable for remote mountain hills and big flaw working, is designed in this paper. It can reduce potato breakage rate caused by digging shovel and improve the entry efficiency, achieving the maximum utilization of power. The design of control circuit of potato harvester is mainly to use GPS satellite positioning to control the route, hydraulic steering control, to complete the auxiliary driving of agricultural machinery, Solve the problem of fatigue driving, improve the personnel utilization rate and potato harvest efficiency. Result shows that it is tiny, agricultural machinery and strong adaptability. Thus, potato production efficiency will be improved and predominantly human labor will be reduced.

With the development of precision and efficiency, UAV is greatly developed in the field of plant protection to saving water and protecting, and is widely used in the future. However, drug liquid losing and farmland environmental pollution are existing in the plant protection UAV spraying operation. Drug mist drops drifting, working in complex field operation environment, is vulnerable affected by temperature, humidity, UAV equipment, and natural wind. Thus it lead to utilization reducing, economic losing and health problems. The characteristics, influencing factors, detection methods and detection technology are discussed in this paper. Recently results shows that the influence of fog droplets should be consider in plant protection UAV, and the new fog droplet drift detection technology is needed to provide guarantee, improving the efficiency of quality assurance operation efficiency, saving agricultural resources and protecting the crop growth environment.

Aiming at the problem of too few identification methods for phased array radar and active decoys and the low recognition rate, this paper proposes a polarization identification algorithm combining polarization descriptor and Multivariate Long Short Term Memory Fully Convolutional Network (MLSTM-FCN) neural network. Firstly, this paper realizes the accurate modeling of fully polarized phased array radar and active decoys in detail, and analyzes the polarization characteristics in spatial domain. The polarization characteristics of full polarization phased array radar and active decoy are distributed differently in the same spatial domain. The sample datasets of polarization descriptors of polarized phased array radar and active decoys in spatial domain is obtained. Finally, we train the recognition model with MLSTM-FCN neural network. The simulation results show that the recognition rate is above 92% at a low signal-to-noise ratio (SNR) of 7 dB.

To study the effects of unmanned aerial vehicle (UAV) fuselage scattering and posture variation, a novel non-stationary geometry-based stochastic model (GBSM) is proposed for multiple-input multiple-output (MIMO) UAV channels. It consists of line-of-sight (LoS) and non-line-of-sight (NLoS) components. The factor of fuselage scattering effect (FSE) and posture are considered by introducing an FSE vector and a time-variant posture matrix into the channel model, respectively. Besides, temporal autocorrelation function (ACF) is derived and investigated in detail. Simulation results show that both FSE and posture have significant effects on the UAV channel characteristic. The agreements between theoretical and simulated results verify the correctness of the proposed model.

An accurate and effective method is extremely important for antenna pattern measurement. This is because antenna pattern, which directly shows the ability of antenna radiation, can affect antenna working and the whole system performance. However, it is difficult to measure the radiation pattern of large outdoor antennas. This paper presents a new precise three-dimensional (3D) antenna radiation pattern measurement method based on the unmanned aerial vehicle (UAV) technique. Cylindrical cone and horn antennas are tested to verify the effectiveness of the method. Comparison results show that the 3D antenna radiation pattern can be quickly obtained by using the proposed method with high accuracy.

In order to improve the frequency measurement accuracy and speed of frequency measurement receiver, an improved rife frequency measurement algorithm is proposed. The implementation steps of the algorithm are described in detail based on the analysis of traditional rife frequency measurement algorithm, The simulation results show that the frequency measurement error of the algorithm is less than that of rife algorithm under the same signal-to-noise ratio. In order to be easy to implement in FPGA and improve the running speed of the algorithm, the division operation in the original algorithm is optimized, the division is realized by subtraction, and the logarithmic operation is realized by look-up table, so that the frequency measurement speed is as high as 240 MHz.

In TDOA location algorithm, positioning accuracy is closely related to the layout of observation stations. In order to improve the positioning accuracy, a particle swarm optimization (PSO) algorithm is proposed to optimize the trajectory of the observation platform under the condition that the observation platform can be maneuverably adjusted. The error geometric dilution (GDOP) value is introduced as the cost function, and the trajectory optimization problem is transformed into a non-linear optimization problem, which is solved in real time by particle swarm optimization (PSO). The position of the observation platform is taken as the optimized position of the observation platform, and the target position is solved by Newton iteration method, Experiments show that the method can significantly improve the positioning accuracy of the target.

Foliage environment target detection has been an extremely difficult problem to solve. In this paper, we propose a machine learning approach for sense through target detection. Detection of target can be achieved with an accuracy of 93.7% with our XGBoost based technology on single received Ultra-Wide Band (UWB) radar waveform. This excellent result is achieved with very less computational resource making it a lucrative application in the target field.