An RHC-QPSO path planning algorithm is proposed for the dynamic path planning of UAVs in complex environments based on the quantum particle swarm optimization algorithm.The algorithm uses quadtree to establish the real-time environment model.The UAV path optimization is made based on QPSO algorithm, and Kalman filter is used to estimate the trajectory of dynamic threats in space.Combined with the RHC method, an active avoidance strategy is adopted against the dynamic threats.The minimized Φ, ψ, θ and a are selected as process performance indicators, and used as the optimization indicators for each rolling optimization window.The simulation experiments show that, the algorithm can not only effectively realize the dynamic path planning of UAVs with certain prior knowledge of the map in real time, but also prevent the UAV from making large maneuvering to avoid dynamic threats during the planning process, and the smoothness of the path is improved to some extent, which can improve the safety of the UAV.

The traditional methods for Missile Attack Envelop (MAE) calculation have poor pertinence and real-time performance, which are difficult to provide a basis for decision-making in rapidly changing situations.A new method of air combat tactical decision-making is proposed combining offline computing with online computing of MAE.The offline computing analyzes the influence of the enemy aircraft and our aircraft to the MAE, based on which the geometry model is established.The online computing assesses the combat situation quickly and quantitatively by considering the relative attitude information of both sides and the offline MAE model, and generates air combat tactics.The synthesized experimental results show that, the generated tactics significantly improves the operational efficiency, and outperforms the traditional tactics based on the front missile launching principle against targets.

Aiming at the problem of heading control of water-jet propulsion type Unmanned Surface Vehicle (USV) in the horizontal plane, an improved terminal sliding mode control method is proposed.Firstly, the Norrbin model of USV is established.Based on this model, a heading controller based on terminal sliding mode control is designed.At the same time, in order to alleviate the chattering problem brought by sliding mode control, the Sigmoid function is introduced for improving the traditional exponential approaching rate, and the stability of the control system is proved by the Lyapunov method.After that, the correctness of the control algorithm is verified by Matlab simulation experiment.The “Haiyao 01” independently developed by our research group has been used for carrying out a real ship experiment on the lake for the designed heading tracking control algorithm.The experimental results verify the feasibility of the algorithm in practical engineering.

In aerial foggy weather, the clarity and fidelity of the acquired images are easily degraded, and the target is difficult to be detected.To solve the problem, this paper proposes an image defogging method based on Gaussian weighted fusion dark channel images with different near and far views and boundary constraints.On the basis of the dark channel prior, a Gaussian weighting function is constructed with different weights for different distances.The Gaussian function can assign different weights to the positions in the dark channel map with different depths of field.Then the boundary constraints are used for rough estimation of scene transmittance, and context regularization is used for refining the scene transmittance and a more accurate transmission rate is obtained.Finally, the restored haze-free image is obtained.The experimental results show that the algorithm can effectively recover the contrast and real color of foggy images, and has strong robustness.

In order to realize the correct track correlation between radar and infrared sensor, a method for the track correlation based on the optimal track membership is proposed.Firstly, the angle information is taken as the measurement data to calculate the statistical distance between the tracks, and then the membership matrix of azimuth angle and pitch angle, and the resolution membership vector are constructed.The resolution membership vector is taken as the decision threshold and the maximum membership principle is adopted to execute the judgment on whether the tracks are correlated or not.The simulation results show that, the performance of the method based on the optimal track membership is obviously better than that of the nearest neighbor method and the weighted track association method in the application of the track association of radar and infrared sensor.Under the conditions with high target density and large sensor measurement error, it is slightly better than the fuzzy comprehensive function method, but the performance is more stable, with a correct correlation probability above 75%.

In order to improve the detection rate of dim targets in a single frame infrared image under low SNR conditions, and considering the characteristic that the high-order cumulant of Gaussian noise is zero and the statistical properties of the Gaussian process, a target detection algorithm based on third-order cumulant is proposed.Firstly, bilateral filtering is used to suppress the background, and then the gray value of the local neighborhood is used to construct the third-order cumulant, which is taken as the test statistic to divide the target point.Since the gray distribution of the target has the feature of rotation invariance, two rotation transformations are performed when constructing the third-order cumulant, which can further eliminate the influence of background clutter while enhancing the target energy, and reduce the false alarm rate of single frame detection.Experiments show that the third-order cumulant based on rotational transformation can effectively suppress edge clutter and detect dim targets with signal-to-noise ratio greater than 0.7.

The Canonical Correlation Analysis (CCA) is employed to fuse the multi-resolution representations of Synthetic Aperture Radar (SAR) images, which is used for target recognition.The multi-resolution representations can provide hierarchical descriptions for the target characteristics, thus providing richer information for the following classification.In order to keep the correlations between multiple resolutions while reducing the redundancy, the Multiset Canonical Correlation Analysis (MCCA) is adopted to fuse them as a unified feature vector.The fused feature vector inherits the discriminability of different resolutions, which is beneficial to improving the effectiveness and efficiency of target recognition.The Sparse Representation-based Classification (SRC) is employed as the basic classifier to make decisions on the target labels.The performance evaluation of the proposed method is conducted on the public MSTAR dataset and the results confirm its validity.

Aiming at the problems of low accuracy and long running time of CNN in SAR target recognition, a new method of transfering VGG16 is proposed based on the combination of transfer learning with convolution neural network VGG16 structure.Firstly, the target features are extracted by fine-tuning the pre-training model of transfering VGG16 for the SAR target in the MSTAR dataset.Then the feature is classified and recognized by Softmax classifier.Experiments show that:The SAR target recognition rate of the transfering VGG16 is increased to 94.4%, compared with 86.2% and 90.8% of the existing VGG16 algorithm and the transfering LENET method.

Aiming at the problem of target detection and parameter estimation of centralized MIMO radar, an optimization method is proposed to improve the performance of MIMO radar parameter estimation and detection.By using the skew-symmetry of covariance of the received data, a linear programming method is proposed for target detection under the generalized likelihood ratio criterion.The simulation results show that, this method can improve the detection probability and parameter estimation accuracy of radar target, implement sidelobe suppression well, and has shorter running time.

In order to meet the requirements for hitting a target at a specified attack angle while intercepting high-speed maneuvering targets, a three-dimensional multivariable super-twisting fixed-time sliding mode interception guidance law with constraint of attack angle is proposed.Based on the three-dimensional relative motion model of the missile and the target, the interception guidance problem is transformed into a second-order system control problem about line-of-sight angle.Based on the terminal sliding mode control theory and the multivariable super-twisting algorithm, the multivariable terminal sliding mode surface and the interception guidance law with constraint of attack angle are designed respectively.Lyapunov function is used to prove the fixed-time stability of closed-loop guidance system.Simulation comparisons with the non-singular terminal sliding mode interception guidance law show that the proposed interception guidance law has good interception accuracy and robustness.

Due to orthogonalization processing and its own structural constraints, the Prolate Spheroidal Wave Function (PSWF) pulse group scheme has high system complexity and high Peak-to-Average Power Ratio (PAPR) of the modulation signal.Aiming at the problems, and considering the odd-even feature of PSWF signal, a sub-band grouping pulse group scheme based on PSWF parity characteristics is proposed.The communication frequency band is divided into several sub-bands with the same and adjacent bandwidth, and the overlapped wavelet bands are divided within the sub-band to construct the pulse group.The natural orthogonality between PSWF signals with different odd-even characteristics in the wavelet band is utilized to completely avoid the orthogonalization processing.Theoretical and numerical analysis show that:1) Compared with the orthogonalized PSWF pulse group scheme, our scheme can effectively reduce the system complexity;and 2) When CCDF is 10-4, the PAPR of modulation signal can be reduced by about 1.49 dB.

When the desired signal is present in training snapshots, the adaptive beamforming is very sensitive to the mismatch represented by the steering vector mismatch, and its performance degrades rapidly.This situation will be more serious when there is jammer in motion at the same time.To solve the problem, a robust adaptive beamforming algorithm based on covariance matrix taper and steering vector estimation is proposed.The algorithm weights the sample covariance matrix to enhance it, and then estimates the real steering vector by using the enhanced covariance matrix.Finally, the enhanced covariance matrix and the estimated steering vector are used for beamforming.The simulation results show that the proposed algorithm can overcome model mismatch while broadening the null notch, which improves the robustness of the beamformer to jammer in motion and model mismatch.

A radar emitter recognition method based on improved Fireworks Algorithm (FWA) is proposed.The improved fireworks algorithm is used to optimize the parameters and identity the type of radar emitter.Based on the analysis to traditional fireworks algorithm, the explosion radius of fireworks is adjusted by nonlinear decreasing to enhance the ability of optimization of the algorithm.The simulation results show that the improved fireworks algorithm can improve the accuracy of radar emitter recognition.

The conventional dynamic surface of Backstepping control is designed as first order filter.Its drawbacks include slow converging speed of tracking/controlling error when closing to equilibrium point, infinite convergence time and vulnerability of the controller gain.In view of these problems, this paper proposes a novel fixed-time convergent Backstepping control scheme based on fixed-time convergent dynamic surface for higher-order multi-input-multi-output nonlinear systems with uncertainties and external disturbance.Firstly, a new fixed-time convergent Lyapunov theorem is proposed and proved.Based on this new tool, the virtual controller and filter of the subsystem of each level are designed with the fixed-time convergent structure.Compared with conventional schemes, the new scheme has the following advantages:1) Speeding up the convergence of tracking error at two stages of far from and closing to the equilibrium point, i.e., the convergence speed is improved within the whole universe;2) Overcoming the parameter vulnerability problem;3) Both the tracking error and controlling error are fixed-time convergent, which means that the convergent time is limited and a fixed upper-bound is existed no matter how big the initial error is;and 4) Both the virtual controller and the controller are nonsingular.

Hyperspectral remote sensing technology has the unique advantages of reflecting the spatial and spectral characteristics of remote sensing objects at the same time, but these advantages also bring such problems as too many bands and their strong correlation, high data redundancy, and difficult further processing and utilization.Dimensionality reduction can reduce redundant information in the data and improve the processing efficiency.As an important method of dimensionality reduction, feature extraction has the advantage of fast dimensionality reduction.Therefore, feature extraction is of great significance to the application of hyperspectral images.This paper introduces the basic principle of hyperspectral image dimensionality reduction, classifies and summarizes the advantages and disadvantages of different hyperspectral image feature extraction methods, points out the problems in the research of hyperspectral image feature extraction methods, and puts forward the development trend of hyperspectral image feature extraction technology.

In recent years, object detection has become a hot topic in the field of information technology.Whether it is a research institute or a large-scale technology-based enterprise, it is pursuing the best object detection algorithm.RetinaNet is an important discovery.It can be seen that the Facebook AI team, the proponent of RetinaNet, has referenced a large number of object detection cases, making it a hot method in the field of object detection.It is better than YOLOv3 and is not so complex as Fastest R-CNN.In this paper, the application of RetinaNet in military vehicle detection is studied, and some suggestions are put forward for improvement.

Aiming at the problems of poor real-time performance and low efficiency for airborne equipment fault diagnosis, a DWT-SVM fault diagnosis implementation method based on DSP+FPGA architecture is proposed.Firstly, the time-frequency domain characteristics of the sensor data are extracted, and the fault feature vector is constructed.Then the support vector machine is used to realize the fault diagnosis of the airborne equipment.The PHM platform based on DSP+FPGA heterogeneous computing architecture is designed.The embedded computing based on DWT-SVM fault diagnosis algorithm is implemented, which provides a solution for the accelerated calculation of airborne equipment fault diagnosis algorithm.The experimental results show that the efficiency of the on-board equipment fault diagnosis acceleration calculation is 5.63 times of that of the PC platform, while the power consumption is only 17.7% of the PC platform.

Rapid development of scale and functional complexity of avionics system brings higher demands on fault tolerance and reliability of Time-Triggered Ethernet (TTE). As a significant measure of fault management, reconfiguration is widely used in avionics system. Aiming at application failure because of end system module failure in TTE, a reconfiguration management strategy based on software defined networking is proposed. Then, by introducing the constraints of end system margin and resource-supply ability, a message scheduling algorithm based on Satisfiability Modulo Theory (SMT) for fault recovery is presented. The uniformity of link load and worst delay of rate-constraint messages are taken as the indexes for evaluation of the preemptive reconfiguration scheme, and an optimization algorithm is designed based on genetic theory for selection of the preemptive reconfiguration scheme. Simulation experiment verifies the validity of the message scheduling algorithm, and shows that the preemptive reconfiguration scheme after optimization has better performance.

In view of the uncertainty and time-varying feature of internal parameters in the control of hydraulic servo system of large-caliber weapon, the traditional PID control is modified by taking the advantages of high robustness and fine fault-tolerant ability of the fuzzy control, while using the RBF neural network to solve the problem of poor accuracy of the fuzzy control.In addition, the ant colony clustering is used to initialize the initial parameters of the RBF neural network, and the conjugate gradient method is used for optimized training of the neural network.The simulation results show that, the control strategy can better suppress the time-varying and nonlinear problems of large-caliber weapon systems while ensuring the speed and accuracy of the system.

The load-end power output of the 2-DOF cable driven stabilized platform is realized by parallel motors. The coupling characteristics of it make the design of the control system more difficult and restrict the improvement of its control quality. On the basis of kinematics analysis, a dynamic model is constructed by Lagrange method. Then based on the typical coupling structure, the coupling degree analysis and calculation of the Double Input Double Output (DIDO) coupling system are carried out. The principle of decoupling control is further analyzed, and a feed-forward compensation decoupling control strategy is proposed. Finally, simulation verification is carried out. The following conclusions are drawn from the study and calculation:1) The studied stability platform is essentially a DIDO coupling system with power coupling and position coupling, and the effect of position coupling is to harmonize the power coupling;and 2) The proposed feed-forward compensation decoupling strategy can decouple the power and position completely and achieve the desired control effect.

In the flight environment of civil aviation. there are various high-energy particles. which may result in the effect of Single Event Upset (SEU) when acting on avionic equipment's SRAM-based FPGA. and cause device logic fault and system breakdown.Hence. the corresponding design of anti-single-event-upset must be taken into consideration for the SRAM-based FPGA in airborne equipment.According to the basic structure of AFDX switch. the concrete design measures are given for anti-single-event-upset of the FPGA based AFDX switch. and simulation-based verification is conducted by using the method of fault injection.It is shown that the method can effectively enhance the performance of resisting single event upset in FPGA based AFDX switch. thus can improve the reliability and stability of system in the application of aviation flight.

With the increasing degree of modularization and integration of avionic systems, the ARINC 818 standard is used to make the new-generation avionic system transmit large-capacity digital video information at high speed in real time, which requires higher quality for the transmitted signal.A new Continuous Time Linear Equalizer (CTLE) is designed based on the ARINC 818 protocol for high-speed signal transmission systems.On the basis of the traditional CTLE, the folded cascode topology structure is imposed on the output of the conventional first-order CTLE to form a secondary structure, which can increase the high-frequency gain and achieve the purpose of channel compensation.The simulation results show that: at the rate of 2.125 Gibit/s, the two-level equalizer structure has good compensation ability, and the horizontal opening degree of equalized eye reaches 0.87UI.