The decision-making problem of multi-UAV cooperative air combat based on game iteration is studied.Firstly, based on the combat situation of the two sides and effectiveness parameters, the matrix game method is used to establish the confrontation payoff game model of the two sides to obtain the payoff matrix.Then, according to the definition of the mixed strategys Nash equilibrium and its derivation process, the method for solving the mixed strategys Nash equilibrium in air combat using game iteration is obtained and the steps are given.Finally, simulation examples have verified the feasibility and effectiveness of the method, which provides a valuable reference for solving multi-UAV air combat strategies.

For the evaluation of target stealth camouflage effect in a single infrared image, based on the classical Structural Similarity (SSIM) algorithm, the gradient value is extracted by the improved quad-direction gradient operator, and the gradient difference function is established to replace the structural similarity function in SSIM algorithm.An image similarity evaluation algorithm based on the Improved Gradient Similarity (IGSIM) is proposed.Based on the algorithm, the experiment of determining the number of equal blocks in an infrared image and the experiment of evaluating the targets infrared stealth camouflage effect are designed.The results show that the evaluation index obtained by the algorithm is more consistent with the visual characteristics of human eye compared with that of the classical algorithms, and the algorithm can better deal with the evaluation of the targets infrared stealth camouflage effect in a single infrared image.

Smeared Spectrum (SMSP) jamming is a kind of active jamming on Linear Frequency Modulated (LFM) signal radar.The jamming signals are highly similar to LFM signals in time domain and frequency domain.When the jamming signals enter from the mainlobe, it is more difficult for radar to detect the targets.To solve the problem, the hybrid polarization antenna reception system is adopted, the differences of jamming signals in adjacent Pulse Repetition Intervals (PRI) are used, and the time-domain channel and the polarization-domain channel are combined to be a joint channel to obtain the mixed signals.Then, the blind source separation algorithm is applied to separate target signals from jamming signals, and finally the target signals are obtained by pulse compression, so that the goal of anti-jamming is realized.Simulation results show that this method can effectively extract target signals from the mixed signals.

Anomaly detection of hyperspectral image is one of the important research contents in processing onboard satellite.Based on the traditional RX algorithm, a hyperspectral image anomaly detection method is proposed by use of incremental learning and the hierarchical method.Incremental learning is used to update the detector model.When generating a new covariance matrix, there is no need to calculate the covariance matrix of all samples, which avoids repeated data calculating and inverse matrix solving.The hierarchical method is used to suppress the background and preserve target spectrum, which effectively improves the performance of hyperspectral image target detector.The experimental results show that: 1)Compared with SAM algorithm and the traditional RX algorithm, this algorithm has the highest detection probability, and its detection result is the closest to the ground target; and 2)The computation complexity of this algorithm is reduced by an order of magnitude, the running time is reduced by 0.215 s compared with SAM algorithm.Therefore, the anomaly detection algorithm proposed here has higher detection speed and occupies less onboard resources, which is superior to the traditional RX algorithm.

In order to give full play to the performance of the missile, it is necessary to optimize the design of multiple parameters such as missiles mass and flight sequence angle under the conditions of meeting technical and tactical indicators.Genetic Algorithm (GA) is very effective for multi-parameter optimization problem.In this process, ballistic integration is needed to calculate the range of all individuals in the population, but the model of ballistic integration is complex, which is not conducive to global search.This paper uses Back-Propagation Neural Network (BPNN) to fit the process of numerical integration when calculating the range, improves the individuals fitness calculation in GA, and completes the design of the missiles overall parameters under the given range index.The simulation results show that: the computational accuracy of the trained neural network meets the requirements, the computational speed of the improved GA is significantly increased, and its local search capability is strengthened.

In the process of countering the mainlobe interference by using the traditional blocking matrix, the interference signal is suppressed, but the complex envelope of the expected signal will also change, which will cause certain signal loss, affect the Signal-to-Noise Ratio (SNR) of the output, and is not conducive to subsequent signal processing.To solve this problem, the construction method of the traditional blocking matrix is improved, and a new algorithm based on the improved blocking matrix is proposed.The received signal is preprocessed by the improved blocking matrix, and then adaptive beamforming is performed.Simulation results show that, under the same conditions, this algorithm reduces signal loss, increases the Signal-to-Interference-plus-Noise Ratio (SINR) gain of the array output, and has better anti-interference performance compared with the traditional blocking matrix.

As for network security situation prediction, an Improved Genetic Algorithm and Particle Swarm Optimization (IGAPSO) is proposed to optimize Extreme Learning Machine (ELM) neural network, so as to obtain higher prediction precision and faster convergence rate.Firstly, the inertia weight and the learning factor in GAPSO are improved to realize self-adaptation at different stages of execution by defining the dynamic exponential function.Secondly, as for the fixed crossover rate and mutation rate in GAPSO, an adaptive crossover and mutation strategy is proposed.Finally, the IGAPSO is used to optimize the initial weights and deviations of ELM.IGAPSO not only ensures the diversity of the population, but also improves the convergence rate of the algorithm.The simulation results show that the fitting degree of IGAPSO-ELM for network security situation prediction can reach 0.99, and the convergence rate is greatly improved compared with that of the contrast algorithms.

For shipboard helicopters, the platforms operational performance is directly affected by operational environment support.To solve current serious problems in operational environment support, this paper introduces the idea of modularization to build the shipboard helicopters operational environment support system covering three levels of plan, function and load.The key technologies such as requirement analysis, plan system generation, coding system, optimal plan configuration and plan decision evaluation are studied.According to different combat missions, the operational environment support plan for shipboard helicopters can be provided precisely and rapidly.

Gravity and geomagnetic field are the main references of aircraft attitude measurement, and their shortcomings in high acceleration environment are hard to overcome.In this paper, the 3D structure vector of electromagnetic wave is taken as the reference of aircraft attitude measurement.According to the relationship between the attitude of airborne electromagnetic vector receiving array and the received electromagnetic signal, the steering vector is established.MUSIC algorithm is used to search the attitude parameters in 3D to maximize the signals spatial spectrum, so that the attitude estimation of aircraft platform is realized.Since excessive computation of MUSIC algorithm will cause slow attitude estimation, an improved Genetic Algorithm (GA) is applied to spatial spectrum search.Simulation results show that the algorithm speeds up attitude estimation while ensuring the accuracy and stability.

An improved algorithm for 3D path planning based on Basic Theta* algorithm is proposed.In view of the helicopters flight environment in low altitude, the 3D models of threat sources are established,which make full use of the known information of the environment, and take the flight length and threat cost as influence factors.The algorithm adopts the variable-weight evaluation function to improve search efficiency, and reduces invalid checks of node visibility in the process of route search to improve the speed of calculation.Finally, a flyable route between the start point and the end point is obtained.The simulation results show that the improved Theta* algorithm can obtain better path planning results in a shorter time compared with A* algorithm and Theta* algorithm.

Operation scheme evaluation is a key link in assisting operational decision-making, which quantitatively evaluates to which extent a planned operation can accomplish a predetermined operational objective.Current evaluation methods are mainly based on the classic multi-attribute decision theory, but these methods unilaterally focus on subjective or objective factors when allocating index weights.To solve the problem, an evaluation model of operation scheme based on adaptive allocation of latent weights is proposed.The model implements the weighted sum strategy within the framework of Bayesian network, and integrates both subjective and objective information in a nonlinear manner according to a given sample set and sample labels, so as to adaptively allocate latent weights of the indexes, and then calculates the evaluation results of the new scheme using the posterior inference mechanism of Bayesian method.Experiments have also verified the effectiveness of the model.

For the online implementation of infinite horizon optimal control for continuous linear systems, an online Q-learning algorithm is designed under the condition that the system dynamics are completely unknown.Based on the Hamiltonian function and the optimal cost function in the infinite horizon optimal control theory, the Q function of the continuous linear system is constructed.An Actor/Critic approximator structure is designed by using the integral reinforcement learning method.With asymptotic stability of the closed-loop system and convergence to the optimal solution, the parameters of the Q function are estimated online.The 6th-order linear system model of the turbocharged engine is numerically simulated, and the results show that both the Critic weight and the Actor weight asymptotically converge to the optimal value, and the model-free optimal control is realized.

With the continuous development of Synthetic Aperture Radar (SAR) technology, radar image target recognition has become an important research direction.In recent years, Deep Learning (DL) technology has been widely used in radar image target detection and recognition.However, small data sample size and unbalanced data sample categories have become important factors limiting the application of DL to SAR image target recognition.The DL-based algorithms of SAR image target recognition are analyzed.Firstly, the common data sets for SAR image target recognition and multi-angle SAR image target recognition methods are introduced.Then, the problems of small data sample size and unbalanced sample categories in SAR image target recognition are summarized respectively.Finally, the remaining problems in SAR image target recognition and the research plan of the next step are discussed.

Accurate calibration of focal length deviation is the key to ensure that the star sensor outputs high-precision attitude information, and optical distortion is the primary source of focal length deviation.According to the geometric model of star sensor imaging and the optical distortion model, the mathematical model describing the coupling relationship between optical distortion and focal length deviation is established, and the expression describing the relationship between the optical distortion and the equivalent focal length deviation is derived.The reference distortion threshold δrmax of navigation star point is proposed for focal length deviation calibration.The simulation results show that the calibration accuracy of focal length deviation when selecting the star points that are all within the distortion threshold is 5 to 10 times higher than selecting those that are partly or all outside the distortion threshold, which reduces the influence of optical distortion on calibration accuracy of focal length deviation to a great extent.

In monocular image depth estimation, current unsupervised learning methods have inaccurate estimation results and fuzzy edges.To solve the problems, an unsupervised monocular depth estimation network that combines multi-scale feature information with semantic information is proposed.The network not only introduces layer connection from the encoder to the decoder to realize the extraction and fusion of features of different scales, but also adds a semantic layer of multiple parallel dilated convolutions between the encoder and the decoder to enlarge the receptive field and make the result more precise.Finally, training and testing are conducted on the KITTI data set.The results show that all the error indicators are lower than that of the current unsupervised learning methods.The accuracy of image prediction reaches 91%, 96.8% and 98.7% respectively under the three ratio thresholds, which exceeds that of all the other supervised and unsupervised methods.The improved method makes the edges clearer and the levels more distinct.

Due to excessive oncoming flow, high total pressure and high total temperature in the test section of a hypersonic wind tunnel, it is difficult to design the layout of optical measurement equipment.To solve this problem, a layout scheme of the binocular vision measurement system as well as the attitude measurement method is designed, which is suitable for hypersonic wind tunnels.Through the high-speed cameras and LED light sources arranged inside and outside the test section, the 6-DOF data of the test model is collected.In order to verify the rationality of the layout, static tests are conducted, and the mark-point technology in the binocular vision system is adopted to calculate the displacement and attitude of the model.In the experiment, a comparison with the actual feeding mechanism on the change of the model’s attitude is conducted, which verifies the feasibility of the layout scheme of the binocular vision system.Within a distance of 1.2 m and a space of 0.5 m3, the test error of rotation attitude is lower than 0.08°, and the test error of displacement is lower than 0.05 mm.

In complex environments with narrow roads and many obstacles, Rapidly-exploring Random Tree* (RRT*) algorithm has the problems of high sampling randomness, low efficiency, slow convergence speed and tortuous path.In combination with an improved collision check mechanism RRT* algorithm, an improved Bidirectional RRT* (Bi-RRT*) algorithm is proposed.The algorithm first performs node sampling in the dynamic target area to reduce execution time.Then, the target bias strategy is introduced to the node expansion stage to reduce the randomness of tree growth, and the bidirectional tree growth mechanism is adopted to improve the convergence speed.Finally, redundant nodes on the initial planning path are eliminated to increase the smoothness of the path and reduce the memory requirement of this algorithm.Experiments are conducted in simulation environment.The results show that the improved Bi-RRT* algorithm reduces the randomness of execution and the number of iterations, and the convergence to a smooth path is realized in a shorter time.

In avionics systems, there are two configurations of airborne integrated modular processor platform: one is LRM configured in a centralized cabin and the other is ARINC600 LRU installed on a bracket.As for the two configurations, their characteristics are described respectively.Based on the status quo of airborne optical interconnection technology, its engineering realization modes of optical-electric transformation and high-density interface transmission are introduced.The verification results of current optical interconnection methods are described, and then the direction of future improvement is discussed.

In order to compensate for parameter uncertainties and disturbances, a quadrotor aircraft speed control method based on Linear Active Disturbance Rejection Control (LADRC) is proposed.Firstly, the dynamic model of the quadrotor aircraft is analyzed.As for the quadrotor aircrafts barycenter in the inertial frame, linear speed of z axis is regulated via the first-order LADRC, and linear speeds of x and y axis are regulated via proportion control and feedforward control.The asymptotic stability of the three axes speed-loop control system is proved via the Lyapunov function.Then, the calculations of the virtual attitude angles are analyzed.The second-order LADRC is adopted to regulate three attitude angles tracking, which proves the stability of the second-order LADRC system.Finally, simulation experiments show that: compared with that of the PD control system, the linear speed tracking of three axes is faster with no overshoot and less tracking errors, and the convergence rates of three attitude angles are faster with less tracking errors.

For the problem of position and attitude tracking control of coaxial octorotor UAVs, a neuro-adaptive sliding mode control approach is proposed in consideration of model uncertainties and external disturbances.Firstly, the dynamic system of coaxial octorotor UAV is divided into two subsystems, that is, a fully actuated subsystem and an under-actuated subsystem.Then, the model parameter uncertainty and external disturbance terms are estimated by using neural network, and a proper sliding mode controller is designed.Based on the designed controller and Lyapunov stability theory, the state trajectories of the two subsystems can be driven onto the corresponding sliding mode surface asymptotically, which ensures that the position and attitude tracking control of coaxial octorotor UAV can be performed.Finally, abundant simulation results have verified the effectiveness of the proposed approach.

The fusion of visual sensor and lidar is a hotspot of research at present, and its actual effect is superior to that of a single sensor.In existing fusion algorithms of visual sensor and lidar, the feature points for positioning are not sufficient, so that the positioning accuracy is not high enough.To solve the problem, this paper makes full use of the depth information provided by lidar, and proposes a multi-strategy SLAM algorithm based on fusion of vision and laser.Before estimating inter-frame posture, the depth value of the feature points in the last frame is judged.According to the three kinds of judgment results, that is, all the feature points have depth information, part of the feature points have depth information, and none of the feature points has depth information, different pose estimation strategies are adopted respectively.Finally, the algorithm is tested on the public data set KITTI, and the experimental results show that the algorithm effectively improves the positioning accuracy and robustness.

For 3-DOF helicopter system, the problem of fault estimation based on adaptive observer is studied.Firstly, a nonlinear model of a 3-DOF helicopter is established, and a fault estimation method based on proportional adaptive observer is studied.Furthermore, in order to improve the dynamic performance of fault estimation of 3-DOF helicopter system, a proportional-derivative adaptive observer is constructed.The robustness of the proposed fault estimation method is guaranteed by introducing H∞ performance, and the observer gains are calculated by using Linear Matrix Inequalities (LMI).Finally, the designed fault observer is simulated and verified by 3-DOF helicopter simulation system.The results show that, the proportional-derivative adaptive observer has higher accuracy and speed in fault estimation compared with the proportional adaptive observer.

Traditional radar emitter identification suffers from low accuracy, bad real-time performance and low robustness under the condition of low SNR. To solve the problem, a radar emitter identification algorithm based on random forest is proposed. This algorithm takes Carrier Frequency (CF), Pulse Width (PW) and Pulse Recurrence Interval (PRI) as features of identification. Firstly, random sampling is conducted on the priori sample set to obtain multiple training sets. Secondly, the training sets are used to build a number of decision tree classifiers. Finally, the decision tree classifiers are used to identify new features and the final identification results are obtained by voting. Simulation results show that this algorithm has good robustness and real-time performance even under low SNR, which can effectively identify radar emitter on the battlefield.