For the problems of sparse planar array optimization with side-lobe concave nulls constraints and premature algorithm, a Hybrid Trigonometric Mutation Differential Evolution (HTMDE) algorithm is proposed based on the idea of parameter adaptation. By introducing side-lobe concave nulls constraints matrix, adaptive penalty function is constructed. Time-varying weight combination mutation strategy and crossover strategy improve the initial global search ability and late convergence ability of the algorithm. The constrained optimization of the planar array with peak side lobe level and side-lobe concave nulls is finally realized. The simulation results show that, compared with the algorithm before the hybrid trigonometric mutation strategy, the algorithm not only optimizes the peak side-lobe level of sparse array, but also designs concave nulls in specified side-lobe area to reduce the influence of active interference.

In order to reduce the serious mutual coupling effect across the elements of the existing collocated vector sensor array and further improve the parameter estimation accuracy, a Sparsely Stretched L-shaped Polarization Sensitive Array (SSL-PSA) is proposed, and a novel method for estimating the azimuth-elevation angles as well as polarization parameters is presented accordingly. Firstly, the signal model of SSL-PSA is established. Then, the SSL-PSA is divided into 6 subarrays, thus the ESPRIT algorithm can be utilized to estimate the Rotational Invariant Factors (RIFs). On this basis, a set of fine but ambiguous estimates and four sets of unambiguous coarse estimates of direction cosine are obtained by a series of mathematical operations. Then, four corresponding steering vectors can be reconstructed and the correct coarse direction-cosine estimation can be determined according to the orthogonality of the steering vector and the noise subspace. Finally, the estimates of Direction-Of-Arrival (DOA) and polarization parameter can be achieved by the existing disambiguate method. Compared to the existing polarization sensitive array consists of collocated vector sensor, the proposed one has no collocated configuration, which can reduce the mutual coupling effect. Additionally, the proposed method can also extend the spatial aperture and refine the direction-finding accuracy without adding any redundant antennas. Simulations are carried out to verify the effectiveness of the proposed method.

The traditional range-extended target detection is usually completed under the condition of scattering point density or scattering point number priori. The detection performance is greatly reduced when the scattering point information of the target is completely unknown. To solve this problem, a Range Spread Target Detection method based on Online Estimation of Strong Scattering(OESS-RSTD) points is proposed. Firstly, the unsupervised clustering algorithm in machine learning is used to estimate the number of strong scattering points and the first detection threshold adaptively. Then, the second detection threshold is determined according to false alarm rate. Finally, the existence of the target is determined through two detection thresholds. The simulation data and the measured data are used to verify and compare with other algorithms. By comparing the Signal-to-Noise Ratio (SNR) -detection probability curves of various methods with a given false alarm probability, it is verified that the proposed method has higher robustness than the traditional algorithm, and the method does not need any priori information of target scattering points.

For the problem that mobile robot can not avoid large concave obstacles during navigation, this paper proposes a multi-state integrated navigation algorithm. The algorithm classifies the running state of mobile robot into running state, switching state and obstacle avoidance state according to different moving environment, and defines the state double switching conditions based on the running speed and running time of the mobile robot. The Artificial Potential Field Method (APFM) is used to navigate and observe the geometric configuration of adjacent obstacles in real time. When encountering an obstacle, the switching state is used to determine whether the state switching condition is satisfied, and the obstacle avoidance algorithm is executed to enter the obstacle avoidance state and enter the obstacle avoidance state to implement the obstacle avoidance algorithm. After the obstacle avoidance is completed, the state automatically switches back to the running state to continue the navigation task. The proposal of multi-state can solve the problem of local oscillation of traditional artificial potential field method in the process of avoiding large concave obstacles. Furthermore, the double-switching condition determination algorithm based on running speed and running time  can realize smooth switching between states and optimize the path. The experimental results show that the algorithm can not only solve the local oscillation problem, but also reduce the obstacle avoidance time and improve the efficiency of the navigation algorithm.

In order to improve the ability of noise elimination and channel equalization of strong non-linear signals, a Multi-scale Kernels learning Affine Projection filtering Algorithm based on Surprise Criterion (SC-MKAPA) is proposed on the basis of kernel learning adaptive filtering method. Based on the kernel affine projection filtering algorithm, the structure of the kernel combination function is improved, and the bandwidths of several different Gaussian kernels are taken as variable parameters to participate in the update of the filter together with the weighted coefficients.The calculation results are sparsed by using the surprise criterion, and the surprise measure is improved according to the constraints of the affine projection algorithm, which simplifies the variance term and reduces the calculation complexity. The algorithm is applied to noise cancellation, channel equalization, and Mackey Glass (MG) time series prediction. The simulation results are compared with the traditional adaptive filtering algorithm and the kernel learning adaptive filtering algorithm, it proves the superiority of the proposed algorithm.

A novel design of nonlinear transformation function for the signal detection in impulsive noise is proposed. The proposed method takes the advantage of adjustable fading factors of the exponential function, it can be effective for different models of impulsive noise. By introducing the efficacy as the objective function, nonlinear design is converted into the problem of optimizing the threshold and bottom parameters to maximize the efficacy. Since the efficacy is continuous, derivative, and unimodal, the optimization problem can be easily solved by the  traditional optimization methods, such as the Nelder-Mead simplex method. Analysis shows that the proposed design can obtain the optimal performance in the widely-used models of impulsive noise, including the symmetric α-stable model, the Class A model, and the Gaussian mixture model. Simulation on real atmospheric noise demonstrates that the proposed design is obviously better than the traditional clipper and blanker. Thus, this paper proposes an optimal and uniform solution for suppressing impulsive noise of various models.

The radio frequency fingerprinting of the emitter is complex, and the performance of Specific Emitter Identification (SEI) is subjected to the present expertise. To remedy this shortcoming, this paper presents a novel SEI algorithm based on signal trajectory image, which realizes joint extraction of multiple complex fingerprints using deep learning architecture. First, this paper analyses the visual characteristics of multiple emitter imperfections in the signal trajectory image. Thereafter, signal trajectory grayscale image is used as the signal representation. Finally, a deep residual network is constructed to learn the visual characteristics reflected in the images. The proposed method overcomes the limitations of existing knowledge, and combines high information integrity with low computational complexity. Simulation results demonstrate that, compared with the existing algorithms, the proposed one can remarkably improve the SEI performance with a gain of about 30%.

This paper analyzes the security communication performance of secondary user communication pairs in Cognitive Radio Non-Orthogonal Multiple Access (CR-NOMA) networks, where interference sources and eavesdropping nodes are randomly distributed. The stochastic geometry theory is used to model the eavesdropping nodes and the interfering nodes as a homogeneous Poisson Point Processes (PPP). Firstly, to ensure the reliability of the primary user communication pairs, the power allocation coefficient set of the sender is obtained, and the closed expressions of the connection outage probability and the secrecy outage probability of the secondary user are further obtained. Then, the variation of the power distribution coefficient with the constraint of the primary user’s reliability is analyzed. Finally, the relationship between outage probability of secondary user communication pairs and the density of the eavesdropping nodes and the transmission power is studied. The research shows that the enhancement of interfering signal reduces the reliability of the system, but brings about a significant improvement of security performance. The simulation results verify the correctness of the theoretical analysis.

To solve interference management and efficiency improvement of multi-cell multi-user heterogeneous wireless networks, the downlink Base Station (BS)-user matching and power allocation problem are studied to maximize the interference efficiency of femtocells. Firstly, consideration of quality of service of macro cell users and femtocell users, the problem is formulated as a multivariate mixed integer nonlinear programming problem. Secondly, the problem is decomposed into two subproblems. The BS selection problem is solved by convex optimization technique. The power allocation problem is firstly converted into a convex one by using quadratic transformation method and Dinkelbach approach, then the problem is resolved by using Lagrange dual methods and subgradient methods. Simulations results show the effectiveness of the proposed algorithm by comparing with the existing algorithms in terms of interference efficiency and interference management.

Considering the security risks and privacy leaks in the process of data and reward in the Mobile CrowdSensing (MCS), a distributed security delivery model based on Tangle network is proposed. Firstly, in the data perception stage, the local outlier factor detection algorithm is used to eliminate the anomaly data, cluster the perception data and determine the trusted participant. Then, in the transaction writing stage, Markov Monte Carlo algorithm is used to select the transaction and verify its legitimacy. The anonymous identity data is uploaded by registering with the authentication center, and the transaction is synchronously written to the distributed account book. Finally, combined with Tangle network cumulative weight consensus mechanism, when the security of transaction reaches its threshold, task publishers can safely deliver data and rewards. The simulation results show that the model not only protects user privacy, but also enhances the ability of secure delivery of data and reward. Compared with the existing sensing platform, the model reduces the time complexity and task publishing cost.

In view of the fact that worm viruses can only infect specific operating systems, the virus propagation rule and security performance optimization strategy in multi-operating system heterogeneous network are studied in this paper. First, considering that most viruses can only spread in link between the same operation system, the parameters of heterogeneous edges ratio are introduced into the Susceptible Infected Remove Susceptible (SIRS) virus transmission model, and the influence of heterogeneous edges and network security performance on the single system virus transmission is studied through system equilibrium solution and basic regeneration number analysis. Secondly, according to the moving target defense thought and technology, the network security optimization strategies is designed for non-isomeric random interrupt, non-isomeric random reconnecting and single operating system random node migration, and the variation of the same ratio and the basic number of regenerated numbers in the three strategies and the impact on the safety of the network are anaylrzed. Finally, the correctness of the virus propagation model is verified by simulation, and the network security performance optimization effects of the three strategies are analyzed.

A novel image encryption algorithm is proposed based on a chaos set which consists of discrete chaotic systems and continuous chaotic systems. The chosen combination of chaotic system is dependent on the encryption intensity. The pixel mean value and pixel coordinate value of images are exploited to control the generation of key, thus enhancing the relationship between chaotic key and plain text. In addition, the octet of cipher text pixel is divided into three parts, and then hided into a processed public image, which can promote the external characteristics of cipher text. The image histogram analysis, correlation analysis, and information entropy analysis methods are adopted to identify the security performance, which indicates the effectiveness of the proposed image encryption algorithm and potential application in the image security transmission.

In this paper, an optical image encryption algorithm based on spiral phase transform and new generalized fibonacci chaotic system is proposed to solve the problems of the Fresnel domain double random phase coding system is insensitive to the first diffraction distance, uneven distribution of chaotic sequences and weak resistance to choice plaintext attack. The plaintext image is encoded as phase information and spiral phase transformed to overcame the insensitivity of the first random phase template and diffraction distance of the Fresnel diffraction transform-double random phase encoding system. The sensitivity of the optical keys is improved. The weighted interference between secure image and plaintext image is added to further increase the sensitivity of the optical keys and dimension of key . A generalized Fibonacci chaotic system, which could generate uniform sequences, is constructed to generate phase templates to overcame uneven distribution of logistic chaos and improve the efficiency of key transmission and the sensitivity of the keys. The chaotic initial value and parameters of spiral phase transform are related to SHA-256. It makes the keys change with the plaintext and achieved the effect of “one encryption at a time”, and enhanced the sensitivity of the plaintext and the ability of the resistance to choice plaintext attack and avalanche effect.Experimental comparison shows that this method can effectively increase the plaintext sensitivity and key sensitivity. This method’ robustness and the key space are sufficiently secure. It is a high security optical image encryption method.

RGB-D saliency detection identifies the most visually attentive target areas in a pair of RGB and Depth images. Existing two-stream networks, which treat RGB and Depth data equally, are almost identical in feature extraction. As the lower layers Depth features with a lot of noise, it causes image features not be well characterized. Therefore, a multi-modal feature-fused supervision of RGB-D saliency detection network is proposed, RGB and Depth data are studied independently through two-stream , double-side supervision module is used respectively to obtain saliency maps of each layer, and then the multi-modal feature-fused module is used to later three layers of the fused RGB and Depth of higher dimensional information to generate saliency predicted results. Finally, the information of lower layers is fused to generate the ultimate saliency maps. Experiments on three open data sets show that the proposed network has better performance and stronger robustness than the current RGB-D saliency detection models.

In order to overcome shortcomings of the traditional fuzzy clustering algorithm for image segmentation, such as that are easily affected by noise, sensitive to the initial value of clustering center, easily falling into local optimum, and inadequate ability of fuzzy information processing, an intuitionistic fuzzy clustering image segmentation algorithm is proposed based on flower pollination optimization with nearest neighbor searching. Firstly, a novel extraction strategy of image spatial information is proposed, and then an intuitionistic fuzzy clustering objective function with image spatial information is constructed to improve the algorithm’s robustness against noise and enhance the ability of the algorithm to process the image fuzzy information. In order to overcome the defects of sensitivity to clustering centers and easily falling into local optimum, a flower pollination algorithm based on nearest neighbor learning search mechanism is proposed. Experimental results show that the proposed method can get satisfactory segmentation results on a variety of noisy images.

When imaging a highly reflective object, the light intensity reflected easily exceeds the maximum quantized value of the light intensity received by the sensor, which causes image distortion of the captured image in the saturated region of light intensity and seriously affects the quality of information transmission. In order to improve the data loss in the high-reflection imaging saturation region, a compression-sensing of high-reflection imaging method based on the new sampling theory of compressed sensing is proposed. A specific measurement matrix is used to conduct linear sampling of the target image, and the single light intensity sampling value of the CCD image sensor is combined with the distribution data in the measurement matrix, and the integrated data is restored and reconstructed with the algorithm to achieve the imaging of the measured target in the high-light environment. The peak signal to noise ratio and gray histogram are used as objective evaluation criteria. Experiments show that this imaging method is robust and feasible, with the proportion of saturated pixels in histogram detection 0% and the peak signal to noise ratio 58.37 dB, realizing the imaging without saturated light in the high-light environment, providing a new direction for the application of compressed sensing in imaging.

Sign language recognition involves computer vision, pattern recognition, human-computer interaction, etc. It has important research significance and application value. The flourishing of deep learning technology brings new opportunities for more accurate and real-time sign language recognition. This paper reviews the sign language recognition technology based on deep learning in recent years, formulates and analyzes the algorithms from two branches - isolated words and continuous sentences. The isolated-word recognition technology is divided into three structures: Convolutional Neural Network (CNN), Three-Dimensional Convolutional Neural Network (3D-CNN) and Recurrent Neural Network (RNN) based method. The model used for continuous sentence recognition has higher complexity and is usually assisted with certain kind of long-term temporal sequence modeling algorithm. According to the major structure, there are three categories: the bidirectional LSTM, the 3D convolutional network model and the hybrid model. Common sign language datasets at home and abroad are summarized. Finally, the research challenges and development trends of sign language recognition technology are discussed, concluding that the robustness and practicality on the premise of high-precision still requires to be promoted.

In order to solve the problem of inaccurate location in insulator target detection, this paper proposes an insulator orientation recognition algorithm based on deep learning. By adding angle information to the axis alignment detection frame, it can effectively solve the problem that conventional deep learning algorithm can not accurately locate the target. First, the angular rotation parameters are introduced into the axially aligned rectangular detection frame to form a directional detection frame. Then the parameter offset is added as the fifth parameter to the loss function for iterative regression. At the same time, in order to improve the detection accuracy, Adam algorithm is used to replace Stochastic Gradient Descent (SGD) to optimize the loss function. Finally, the insulator directional detection model can be obtained. The experimental results show that the orientation detection frame with rotation angle can effectively locate the insulator target accurately.

For the non-linear and non-stationary characteristics of motor imagery ElectroEncephaloGram (EEG) signals, an EEG signal recognition method based on Conditional Empirical Mode Decomposition (CEMD) and Serial Parallel Convolutional Neural Network (SPCNN) is proposed. In the CEMD process, the correlation coefficient between the Intrinsic Mode Functions (IMFs) and the original signal is used as the first condition to select IMFs. Based on this, the relative energy occupancy rates between the IMFs are proposed as the second condition to select IMFs. Further, to consider the characteristics between the EEG signal channels and highlight the features in each EEG signal channel, a SPCNN model is proposed to classify the processed EEG signals. The experimental results show that the average recognition rate reaches 94.58% on the dataset collected by ourselves. And the average recognition rate reaches 82.13% on the BCI competition IV 2b dataset, which is 3.85% higher than the average recognition rate of convolutional neural network. Finally, the online control experiments are carried out on the designed intelligent wheelchair platform, which proves the effectiveness of the proposed algorithm for EEG signals recognition.