In order to obtain better guidance performance, a dual-control adaptive sliding mode guidance law of modified proportional guidance and flight velocity control is proposed by using the added flight velocity control freedom of engine flow adjustment missile.The modified proportional guidance is designed with zero overload at the end of the trajectory.Based on the analysis of its instantaneous miss distance, the angle of sight and flight velocity control quantity are selected as sliding surfaces.Furthermore, the adaptive sliding mode control method is adopted to derive the velocity control guidance law to reduce the miss distance.The simulation results show that, compared with proportional guidance and modified proportional guidance with speed retaining, the designed adaptive sliding mode guidance law has smaller miss distance, smoother trajectory and smaller overload requirement, thus realizing the active control of missile flight velocity.

In a complex battlefield environment, it is difficult to grasp the timing of the battle.Therefore, when completing the “intelligence reconnaissance and destroying strike” mission, it is necessary to quickly connect the Surveillance UAV (SUAV), Combat UAV (CUAV) and Ground Control Station (GCS), so as to realize the timely interaction of information between combat nodes.An algorithm based on Voronoi-Sparrow Search Algorithm (V-SSA) is proposed to realize safe and effective assignment and deployment of Relay UAVs (RUAV).Firstly, the mission scenarios of the research are introduced.Secondly, mathematical modeling is conducted for the deployment of RUAV nodes, and the specific implementation method of RUAV deployment based on the V-SSA is verified.The relay node deployment method uses the V-SSA to select the location of the relay node, so as to ensure that the minimum number of RUAVs and the farthest distance from the key target are the objective function.The simulation results show that the proposed relay deployment method can effectively establish the global battlefield communication network, and improve the self-safety protection of the RUAV.

Aiming at the problems of RRT* algorithm in path planning in complex obstacle environment, such as blind search, redundant nodes, unsmooth path and easy approach to obstacles, a Magnified Obstacle-based Bidirectional Dynamic Goal Bias RRT* (MOBDB-RRT*) algorithm is proposed.Firstly, the obstacles are magnified to ensure a safe running distance for the robot.Then, based on RRT* algorithm, a bidirectional dynamic goal bias strategy is introduced, which reduces the time of searching the path and improves the planning efficiency of the algorithm.Finally, pruning algorithm and cubic Bézier curve are used to optimize the planned path, so as to generate a shorter and smoother path.The simulation results show that the improved RRT* algorithm is superior in path planning efficiency and path quality.

In order to realize the goal of improving the efficiency of autonomous trajectory planning during long-range penetration of aircraft, in view of the characteristics of radar threat distribution under the background of this mission, the subdivision grid theory is used to organize the grid environment, and the A* algorithm is improved from the underlying characterization mode.The code of the split grid is used to characterize the organizational structure to obtain the orientation information, thereby imposing a penalty factor in a targeted manner, and the cost calculation method of the actual moving path is improved based on the directivity.Then, according to the characteristics of radar threat distribution, the sub-nodes are found by using the variable step size in sections based on the variation of comparative coding bit.The simulation results show that the node computation of the improved algorithm is greatly reduced, and under the condition that the total cost value of the trajectory sought is approximate, the improved algorithm can always obtain a feasible trajectory more quickly in the environment with dense threat distribution, which is well suited to the background of aircraft penetration.

Aiming at the shortcomings of the basic Grey Wolf Optimization (GWO) algorithm, such as insufficient population diversity and easy to fall into local optimum, an improved grey wolf optimization (ISIAGWO) algorithm based on information sharing search strategy is proposed from the perspectives of chaos initialization and information sharing among populations. Firstly, the Iterative chaotic mapping is used to initialize the population to ensure the diversity, and the adaptive dynamic operator is introduced to increase the weight of outstanding individuals; Secondly, the information sharing search strategy is used to update the population to effectively avoid the algorithm falling into local optimum; Thirdly, eight benchmark functions are tested for optimization and the proposed algorithm is compared with other advanced swarm intelligence algorithms.The experimental results show that ISIAGWO algorithm has significantly improved the accuracy and robustness of the solution; Finally, ISIAGWO algorithm is applied to solve the classic traveling salesman problem, so as to prove the practicability of the algorithm.

Aiming at the problem that UAV target tracking is easily affected by factors such as the change of angle of view, occlusion and background clutter, a multi-feature spatial-temporal regularized correlation filters for UAV tracking is proposed.Firstly, the saliency feature is introduced into the spatio-temporal regularized correlation filter tracking framework, which is combined with the features of color, gray level and histogram of oriented gradient to improve the diversity of target appearance representation.Secondly, the peak sidelobe ratio is utilized as the weight to measure the peak strength of correlation response maps for different features, and the weighted features are combined for noise reduction, so as to realize the final weighted fusion in the response layer and improve the tracking accuracy.Finally, it is compared with 12 classic trackers on the public data set UAV123@10FPS.The experimental results and analysis show that the proposed method achieves favorable results in both tracking accuracy and success rate.

A robust sliding mode synchronization control method under the Delta operator framework is proposed for nonlinear chaotic systems under DoS attack and network failure.Firstly, based on the Delta operator theory, a unified model for continuous and discrete time nonlinear chaotic systems under the Delta operator framework is established.Secondly, the corresponding linear sliding mode surface is designed for the Delta operator chaotic synchronization error system, and the sufficient conditions for the existence of the sliding mode surface are given by using the Linear Matrix Inequality (LMI) method.Thirdly, the sliding mode controller is designed for the synchronization error system, so that it can reach the sliding mode surface quickly in finite time and realize the asymptotic synchronization of the nonlinear chaotic system.Finally, the simulation results show that under the designed controller, the synchronization error tends to zero and the asymptotic synchronization is realized, which shows the feasibility and effectiveness of the proposed method.

Aiming at the problems of heterogeneous fog distribution and uneven illumination of underwater images, an image enhancement method is proposed based on global feature dual attention fusion generative adversarial network.Firstly, the convolutional layers are used instead of average pooling layers to continuously down-sample the input images and extract the global features.Secondly, the global feature dual attention fusion block is constructed, which is adaptive to the changing water environment and can enhance underwater information with various dissemination degrees more significantly.Finally, conditional information is incorporated as a restriction in training to increase the networks stability.The results of experiments demonstrate that the proposed algorithm outperforms the classic and latest algorithms and has fine functionality.

Aiming at the problem of poor adaptability and applicability of traditional error compensation methods for air defense radar, a Back Propagation (BP) neural network optimized based on improved particle swarm optimization algorithm is constructed, which can estimate radar error more stably and accurately and compensate for radar measurements, so as to better improve radar detection accuracy.Firstly, the convergence factor and dynamic adaptive adjustment of inertia weight are introduced to improve the global optimization ability and convergence speed of particle swarm optimization algorithm.Secondly, the improved particle swarm optimization algorithm is used to optimize the initial weight and threshold of BP neural network, improve the estimation accuracy of BP neural network and reduce the training time.The actual measurement data of a certain radar is used for simulation and verification.The simulation results show that the accuracy and error fluctuation of the range, azimuth and pitch angle after compensation are greatly improved.Compared with the traditional method, the compensation effect is better, and the engineering applicability and popularization applicability are stronger.

In this paper, density peak clustering is introduced into radar signal sorting, which can quickly find the cluster center without determining the number of clusters.The concept of density entropy is proposed by referring to potential entropy, and the cutoff distance selection algorithm of kernel function in clustering is optimized.The problem of how to automatically judge the cluster number and cluster center point is studied, the threshold function and judgment rules are designed.The allocation-merging criterion of the original algorithm is also optimized.The simulation results show the effectiveness of the algorithm.

As outboard active centroid jamming has obvious effect on tracking and penetration of anti-ship missile, considering the process of anti-ship missile penetration and ship maneuvering, a dynamic model of attack-defense confrontation between anti-ship missile and ship outboard active centroid jamming is established by using the principle of outboard active decoy jamming, and a discrimination model of outboard active centroid jamming is designed.The maneuvering modes of anti-ship missiles at different incident angles are studied.Through simulation experiments, the tactical effects of different anti-ship missiles turn-on distances and incident angles against outboard active centroid jamming are analyzed, which has important reference significance for reasonably determining the specific parameters of anti-ship missiles terminal guidance radar.

Aiming at the difficulty of acquiring SAR image data, a multi-scale generative adversarial network based on single-image training is proposed, and it is applied to the augmentation of SAR aircraft images.Because the original generative adversarial network sets a single-scale convolution kernel, only the feature distribution of the image under the fixed receptive field is obtained.Therefore, integrating multi-scale group convolution into the adversarial network can mine the distribution features of the image from different scales, and increase the detailed information of generated image.The result is that 400 new SAR aircraft image samples are obtained by training, and the augmented data set is verified by Faster R-CNN and image quality assessment indexes.The experimental results show that the quality assessment indexes of generated images meet the requirements of image detection.Using Faster R-CNN algorithm combined with data augmentation of generative adversarial network, the average detection accuracy is improved from 73.5% to 77.6%.

Aiming at the problem that the traditional target detection algorithm has poor detection effect on small vehicle targets in aerial images, a vehicle detection algorithm based on improved Faster R-CNN is proposed.Based on the original Faster R-CNN network, this method combines FPN as the basic network model—FFRCNN, ResNet-50 is applied instead of original VGG-16 as the main backbone network for multi-feature fusion, and Focal Loss function is used to correct the imbalance between positive and negative samples.On the basis of improving the network, atrous convolution is used to fuse the feature information of multi-scale space, which can improve the receptive field of the network and better collect the context information of the image.The experimental results show that the average accuracy of the improved detection algorithm reaches 93.8%, which is 19.2% higher than that of the original FFRCNN network and has better robustness.

Aiming at the problem of unexpected obstacle avoidance during UAV flying, an improved Artificial Potential Field (APF) strategy is proposed.In the process of obstacle avoidance, the obstacle avoidance maneuver angle is used to design the obstacle repulsion coefficient and the velocity control item is added to jointly improve the traditional APF.In comparison to the traditional UAV obstacle avoidance method of setting the target point, this paper uses the virtual submachine to apply the track regression force to the UAV in real time, which ensures that the UAV can return to the original route faster.Furthermore, model prediction is used to adjust the track regression force to avoid the phenomenon of jitter, and then the UAV obstacle avoidance maneuver strategy is designed.The simulation results show that the proposed strategy has higher obstacle avoidance efficiency than the traditional APF method.Finally, as a basic module, the proposed strategy is applied to different complex tasks, which proves that the proposed strategy has high versatility.

In order to solve the problem that the indexes are difficult to describe quantitatively and the index data is uncertain in the process of aviation safety risk evaluation, a fuzzy evidence reasoning method for aviation safety risk evaluation is proposed.Based on the “man-machine-environment” system safety theory, the evaluation index system is constructed from four aspects of “personnel, aircraft, environment and management”.On this basis, the safety risk value and reliability structure representation of each evaluation index are determined based on triangular fuzzy number and its operation rules, and then the overall safety risk value is calculated by using the evidence reasoning algorithm, so as to determine the safety risk level.Finally, an example is given to verify the effectiveness and rationality of the method.

In order to deeply understand the cascading failure effect of the airline network under deliberate attack, the invulnerability of the airline network under different load conditions is systematically studied.Firstly, the node load and capacity of the airline network are defined, the cascading failure process of the airline network is modeled, and three states of nodes are defined, that is, normal, congestion and failure.Then, according to the operation characteristics of the airline network, three kinds of initial loads are proposed, and their effects on the survivability of the airline network under two kinds of load reallocation schemes are studied.The simulation results show that when the actual flight volume is taken as the initial load of the node, the airline networks invulnerability is better than that under the traditional initial load allocation scheme.Compared with the traditional load allocation scheme, the excessive load reallocation scheme can effectively alleviate the effects of cascading failure.The results show that the initial load scheme based on the actual flight volume can effectively test the rationality of the traditional load allocation model.The results also reflect that the traditional initial load settings cannot fully reflect the characteristics of the network nodes, and further optimization is needed to better reflect the survivability of the real network.

The Miniature Inertial Measurement Unit (MIMU) is vulnerable to environmental influences, and has the shortcomings of non-linear output and low accuracy under severe angular and linear motion.Considering that the traditional polynomial calibration method is difficult to compensate for the dynamic error accurately, deep learning method was used to model and compensate for the overall dynamic error of MIMU.The shallow neural network and the deep recurrent neural network were used to establish the overall dynamic error model of MIMU.The calibration process based on three-axis temperature control position and rate turntable was designed, so that angular motion and temperature changes were applied to the three axes of the turntable of the inner, middle and outer axis simultaneously.An MIMU multi-factor influence error training set was also built.The experimental results showed that:1) The shallow neural network model has a slight improvement on the error compensation effect compared with the traditional model, and the deep recurrent neural network model can reduce the residual mean and mean square error significantly through compensation; 2) The effect of Gated Recurrent Unit (GRU) neural network model is the best, and there are fewer parameters to be trained and low computational burden.

Aiming at the problem of low accuracy of multi-scale pedestrian target detection in low-illumination environment, a pedestrian detection method based on YOLOv5s and infrared and visible light image fusion is proposed.Firstly, the Generative Adversarial Network (GAN) is used to make a data set of the fused visible light and infrared images, and the influence of external factors on pedestrian detection can be reduced by combining the advantages of two kinds of images, and then, the SENet channel attention module is introduced into YOLOv5s to make the network pay more attention to the highlighted target, so as to improve the accuracy of pedestrian detection.Finally, the structure of YOLOv5s network is optimized, part of the convolution layers are deleted, and the activation function is modified to maintain the high real-time performance of the algorithm.Experimental results show that a detection model with higher mAP can be obtained by using the fusion image data set for training than using the visible light data set or the infrared data set.The improved SE-YOLOv5s algorithm effectively improves the mAP of pedestrian detection while maintaining the high real-time performance of the original algorithm.

To solve the problems of deep network gradient disappearance and image information loss caused by superimposition of simplex convolution layer when extracting image information in current image super-resolution reconstruction algorithms, an image super-resolution reconstruction algorithm based on multi-scale and residual network is proposed.The proposed algorithm uses multi-scale dense connection convolution kernel instead of the simplex accumulated convolution kernel, to fully extract the low-resolution image input information and realize reuse of channel feature dimension.The residual network is used to supplement the lost image information at multiple levels and suppress the gradient problem of the deep network model, which helps the whole network model to adaptively update the weights in the process of reverse propagation.In the end, the final reconstructed image is output through nonlinear mapping.Experiments show that:1) The peak signal-to-noise ratio and structural similarity of the proposed algorithm on the test set are improved compared with that of the contrast algorithms; and 2) In comparison with the current mainstream algorithms, the proposed algorithm obtains a reconstructed image with richer detail information and clearer edge texture.

In the early stage of fire, different amounts of smoke are often generated, so the high-precision and sensitive detection of smoke plays an important role in preventing the spread of fire.A SmokeNet algorithm based on optical flow estimation and target detection is proposed to detect smoke.The algorithm firstly converts the color space of the input image, then estimates smoke spreading by using the optical flow estimation algorithm LiteFlowNet, and eliminates the interference of moving objects by using the target detection algorithm YOLOv4.Finally, the smoke area size, shape and spreading track in the image can be obtained via noise reduction,so that the smoke can be evaluated.In the indoor smoke evaluation experiment, the method achieved 93.53% detection accuracy.

The piezoelectric micro-motion rod is a key component of the active focusing system of optical equipment, but the nonlinear piezoelectric hysteresis and complex electromechanical coupling effects seriously affect the output displacement accuracy of the piezoelectric micro-motion rod,thus affecting the performance of the optical equipment.In order to realize high-precision control of the piezoelectric micro-motion rod, a multi-field coupling dynamic model of hysteresis is established.Furthermore, a robust composite controller is designed for the piezoelectric micro-motion rod.The composite controller consists of the robust H∞ feedback controller and the feedforward compensator based on the inverse Bouc-Wen hysteresis model.The proposed control method can compensate for the effects of piezoelectric hysteresis and improve the control accuracy while ensuring the robustness of the system.Finally, the experimental system of the piezoelectric micro-motion rod is designed to verify the effectiveness of the proposed modeling and control method.The experimental results validate that the proposed robust composite control method can realize high-precision control of the displacement of the piezoelectric micro-motion rod.

Synthetic Aperture Radar (SAR) image Automatic Target Recognition (ATR) technology is one of the key technologies of artificial image interpretation, which aims to isolate the influence of inherent noise, obtain the potential characteristic information of the target in the region of interest, and provide strong data support for target recognition.In order to improve the accuracy of target recognition in high-resolution SAR images, focusing on the problems of speckle suppression and feature extraction in algorithm design, an automatic target recognition framework for SAR images is designed by combining the traditional Constant False Alarm Rate (CFAR) detection algorithm and the latest research of the Deep Convolutional Neural Network(DCNN).The experiment is based on MSTAR standard data set, and the results of target recognition show the effectiveness of the model.

When the UAV flies in the attitude mode, the attitude angle error fluctuates greatly.According to the complementary characteristics of magnetometer, accelerometer and gyroscope, an Adaptive Unscented Kalman Filter (AUKF) algorithm is proposed to optimize the MEMS sensor data.The attitude quaternion and gyro drift are taken as state variables, and the output of accelerator and magnetometer is taken as measurement variables.The gradient descent algorithm is used to optimize the key parameter of Unscented Kalman Filter, namely, process noise covariance, so as to improve the accuracy of attitude calculation.The analysis of actual flight data shows that the proposed method has the highest accuracy compared with conventional Kalman filter and traditional unscented Kalman filter, and can ensure flight stability of small UAVs in various situations.