Due to its advantages of low cost,consumable,distributed deployment,agility and flexibility,UAVs have shown great success in many civil fields.However,due to the limitation of its intelligence,there are still significant challenges in how to autonomously and safely complete tasks under complex adversarial conditions.Aiming at the problems of intelligence and safety in UAV mission planning,based on safe reinforcement learning,a UAV intelligent planning method called SAC-Lagrangian is proposed.Considering the radar threats,no fly zone safety constraints and ground-to-air missile(SAM) countermeasure conditions,the mission planning problem is modeled as a Constrained Markov Decision Process (CMDP),which is transformed into a dual problem through Lagrangian multiplier method.The maximum entropy Soft Actor-Critic(SAC) algorithm is used to approximate the optimal policy,ensuring that the agent can maximize the expected return under the safety constraints.Compared with other baseline algorithms,simulation results show that the proposed method can ensure the safety while ensuring the task performance,adapt to the dynamical changing scenarios,and achieve a task completion rate of 96%.Therefore,the proposed method is efficient,robust and safe.

The spectral consistency between the camouflage target and the background is an important indicator for evaluating the optical camouflage effect.Under the optical reconnaissance threat of random time and arbitrary direction,changes in the angle of sunlight incidence and different observation angles result in corresponding changes in the spectral characteristics of the observed target and background,leading to spectral dynamic differences that make it difficult to accurately evaluate the fusion degree between disguised targets and background.To address this issue,a spectral consistency analysis method based on Bidirectional Reflectivity Distribution Function (BRDF) is proposed.The model parameters are obtained through generalized inverse matrix regression fitting,and a kernel coefficient consistency index is constructed to quantify the difference between disguised targets and background parameters,achieving dynamic spectral consistency analysis from multiple angles.The model is compared with traditional evaluation methods through experimental verification, and the results show that the evaluation results of traditional consistency analysis methods fluctuate with changes in angles,while the proposed method provides stable evaluation results for dynamic spectral consistency.This indicates that the method is more reliable and reasonable than the traditional methods, which has reference significance for the evaluation of optical camouflage.

In order to address the issues of poor robustness and low control accuracy of attitude control systems for ducted UAVs in narrow spaces due to influence of turbulance,the self-developed coaxial twin-rotor ducted UAV is used as the research object for kinematic and dynamic modeling.Based on this,modeling errors and external disturbances during flight are considered.A cascade method of PID and Adaptive Disturbance Rejection Control (ADRC) is used to design the attitude control system of the ducted UAV.A simulation model is built for contrast experiment of the UAV under ideal and turbulent wind disturbance conditions.Finally,flight tests are conducted on a real aircraft.The results indicate that the designed attitude control system has good robustness and anti-interference performance.

Aiming at the problem of area coverage reconnaissance path planning for fixed-wing UAVs,a concave area coverage path planning method based on improved Ant Colony Optimization(ACO) is proposed.The vertical load reconnaissance model is established,and parallel coverage strategy is introduced,the flight direction is determined by rotating calipers algorithm.For the irregular concave polygon region,a multi-strategy concave polygon conversion method combining concave point removal and region decomposition is proposed.The coverage path planning problem is transformed into an optimization problem of searching the optimal strip traversal order,and Dubins turning paths are established.The problem is solved by improving the heuristic function in ACO,introducing adaptive pheromone volatilization coefficient and 3-opt algorithm.The experimental results show that the proposed method can effectively realize the concave area conversion and obtain a shorter area coverage reconnaissance path,which can provide theoretical support for the research of autonomous regional reconnaissance of unmanned reconnaissance aircraft.

Aiming at the problems of poor detection accuracy and large amount of computation in the existing SAR ship target detection methods,a lightweight ship target detection method based on YOLOv5 and GhostNet is proposed.The GhostConv and GhostC3 modules of the lightweight network GhostNet are introduced to improve the backbone network of YOLOv5,achieving a significant reduction in model computation.The CBAMC3 module is introduced in the neck network to adjust attention during the feature fusion stage and achieve accurate target detection.In addition,the EIoU loss function is introduced to improve the regression accuracy and rate of convergence of the prediction box.The test results on the public dataset indicate that the improved algorithm significantly reduces the number of parameters and model volume while maintaining high accuracy,making it an ideal lightweight ship detection model for SAR images.

In the target tracking system in complex environment,due to the influence of random pulse interference,modeling error,unknown outliers and other factors,the process noise and measurement noise of the system model show complex non-Gaussian heavy-tailed characteristics.A method based on the KL Divergence (KLD) minimization in the distributed fusion framework is proposed.Firstly,a priori model including many parameters such as target state,process noise and measurement noise is constructed as a studentt t distribution.Secondly,KLD minimization solves the problem of distance in fitting approximate distribution to the real distribution,and improving the accuracy of studentt t modeling.Finally,the Covariance Intersection (CI) fusion strategy is adopted to realize the fusion and correction of local platform state estimation.The simulation results show that the proposed algorithm has higher estimation accuracy compared with the traditional NKF,STF and MCCKF algorithms.

In order to improve the performance of target recognition in SAR images,this paper proposes a method combining Kernel Sparse Representation-based Classification (KSRC) and augmented dictionary based on traditional Sparse Representation-based Classification(SRC).The KSRC introduces nonlinear kernel function on the basis of SRC,so as to improve the representation ability of the classifier for nonlinear data relationships.By using original training samples,the augmented dictionary expands the original dictionary through noise addition and partial occlusion to improve its adaptability to typical Extended Operating Conditions (EOC).At the same time,with the help of KSRC,the augmented dictionary further improves the coverage of other related EOCs,thus the effectiveness of the proposed method for other EOCs can be upgraded.Experiments carried on the MSTAR dataset under Standard Operating Conditions (SOC) and EOCs including noise interference and partial occlusion shown the superior performance of the proposed method.

In view of the shortcomings of UAV path planning in 3D complexity environment such as search stagnation and convergence to local optimization,an improved chimp optimization algorithm based on multiple strategies is proposed.To address the shortcomings of the optimization accuracy of the chimp algorithm,a convergence factor nonlinear update strategy is introduced to balance the global search and local development capabilities.A weight factor is designed to avoid the blindness of individual following and the assimilation of individuals in the later stage of iteration for improving the search accuracy.A golden sine Levy flight guidance mechanism is designed to prevent falling into local optimization due to the gradual dilution of diversity.The improved chimp optimization algorithm is used to solve the UAV path planning problem for constructing a path planning model by use of 3D terrain map of the UAV flight environment,designing a multi-constraint flight cost function and using it as a fitness function,and iteratively solving the UAV 3D path planning scheme.The results show that the improved algorithm can search for a safe obstacle avoidance path with lower trajectory cost,and the search accuracy is higher than that of analog algorithms.

Ordinary neural networks are difficult to generate infrared and visible light fusion images that conform to human vision,and the network model is complex and occupies too much memory.The existing Generative Adversarial Network (GAN) framework is improved.Firstly,deep convolution and point by point convolution are integrated into the generator,and a convolutional network with small convolution kernels is designed to reduce network parameters.Secondly,mask processing is applied to the source image to reduce the loss of source image information during feature extraction.Then,the processed image and the fused image obtained by the generator are jointly input into the discriminator to enhance the networks ability to retain source image information for visible light images.Finally,in the performance evaluation stage,the loss functions are set as gradient loss,adversarial loss,and content loss functions to constrain the fusion image to contain more background information of visible light images and target information of infrared images.The results of simulation experiments on the TNO image fusion dataset show that the proposed algorithm can obtain fused images with rich details and clear targets while reducing network complexity and operational parameters.

Aiming at the problem that the air combat control ability evaluation of combat personnel has more subjective evaluation and less objective evaluation,an air combat control ability evaluation method based on eye movement data is proposed.Combining eye movement elements with air combat control process,a quantitative assessment model of competency is established.Then,the entropy weight-coefficient of variation method is used to combine and assign weights to the four indexes,and the results are generated into a visual radar map of air combat control capability.Finally,the TOPSIS method is used to verify the competency ranking results.The results show that the proposed method can provide an objective evaluation of the air combat control capability,and also establish a standardized basis for the future evaluation of the air combat control capability of combat personnel.

Parallax has a significant impact on the observation and aiming of a telescopic system.The traditional parallax detection methods mainly rely on the subjective interpretation of human eyes and the detection accuracy is low,and the parallax adjustment quantity cannot be given quantitatively.Based on the prototype improvement of parallax detection by theodolite,a new parallax detection method for the telescopic system with two optical paths is designed.It receives the outgoing rays on both sides of the objective lens of the telescopic system through two optical paths respectively,detects the angular parallax of the telescopic system by relying on image interpretation technology,and obtains the line parallax of the telescopic system through formula.The mathematical relationship between the offset of cross division center and the parallax value of image interpretation is established according to the physical model.The system can realize the parallax detection of a wide range of optical calibers with high efficiency,high precision,automation and other advantages,and the detection accuracy is as high as 11.9″.

Aiming at the problems of edge loss and color distortion in the process of dehazing,a multiple dehazing network based on structure-texture decomposition is proposed.According to the theory of image decomposition,the haze image is divided into structure layer and texture layer.Structure layer image contains most haze and structure information,a deep network is used for dehazing.The texture layer image contains detailed information such as texture edges,a shallow network is used to enrich the texture information.The network uses multi-scale convolution to improve the robustness of the feature map,and jump connection to reduce the number of parameters of the operation.To avoid color distortion caused by error in image reconstruction,the color visibility recovery module is used for color compensation.The experimental results show that the proposed algorithm promotes the restoration of edge details,retains the original image color, and performs well in objective indicators and visual effects.

In complex combat environments,pilots need to quickly and accurately acquire and process a large amount of information.The rationality of the design of the cockpit display and control interface directly determines the pilots work efficiency and decision-making quality.From the perspective of human-machine ergonomics,this paper systematically analyzes the design elements and principle requirements of the human-machine interaction interface for military aircraft cockpit display and control,and elaborates on how to incorporate these principle requirements into the design process to guide the application of human-machine ergonomics in the top-level design of aircraft cockpit display and control systems.

In response to the limitations of the Informed-RRT* algorithm in path planning,such as slow convergence speed,inadequate targeting and non-smooth trajectory,a bidirectional regional sampling RRT* algorithm is proposed.Firstly,the bidirectional greedy search approach is introduced to expedite the identification of sampling points while simultaneously adapting the expansion rules governing the growth of the random tree.This dual-pronged strategy not only accelerates the search process but also enhances its alignment with specified objectives.Secondly,following the initial solution establishment,a heuristic sampling region is introduced to proximate to trajectory nodes,and the path length is continuously iteratively optimized through node reconstruction strategy within this region.Finally,a combination of intermediate point interpolation and cubic spline curve techniques is employed to smooth the path.Simulation results demonstrate that the proposed algorithm can generate fewer nodes,lower costs and smoother paths in different environment maps with less runtime.

ng SAR images for ship detection,it is inevitable to be affected by speckle noise,and nearshore ship detection is easily overwhelmed by complex background signals.A ship detection algorithm RBox-YOLO based on edge feature fusion network is proposed.Using YOLOv8 as the baseline network,the edge of Canny operator is optimized to enhance the contour edges in the image,forming a more complete object boundary.An FDN module based upon coordinate attention mechanism is introduced to fuse denoised images to improve the ability of capturing key information in complex background.The CAU module,which combines bilinear interpolation method with attention mechanism,reduces the detail feature loss caused by upsampling.In addition,a loss function on the basis of rotating frame is used to enhance the ship detection effect under complex background.The experimental results show that RBox-YOLO not only maintains the real-time detection speed of YOLOv8 algorithm,but also improves the average accuracy by 8 percentage points.It is preliminarily concluded that RBox-YOLO algorithm has good detection performance and high application value.

An improved algorithm based on YOLOv5 is proposed to address the problem of poor detection due to complex image background and too small target under UAV vision.Firstly,the algorithm proposes a Filter Separation Feature Extraction (FSFE) structure,which inputs the filter separated image into the neural network in parallel with the original image,strengthens the networks extraction of important information both globally and in detail,the output feature map is spatially adaptively fused to prevent the problem of semantic information fragmentation during fusion,and enables the network to pay more attention to the information of key layers.Secondly,a small target detection layer is added,and the SPD convolution module is utilized to enhance feature learning to improve detection performance.Finally,the CA feature enhancement module is embedded in the C3 module to mine and preserve important semantic information during feature extraction.Experimental results based on the VisDrone 2019 dataset show that mAP@0.5 and mAP@0.5∶0.95 of the improved algorithm increases by 8.3 and 6.1 percentage points respectively,and the accuracy and recall increases by 5.1 and 4.5 percentage points respectively,improving the precision of small target detection and reducing the probability of missed and false detection,which is significant for realizing the UAV visual small target detection.

In the field of electronic countermeasures,Digital Radio Frequency Memory (DRFM) can quickly store and reproduce radio frequency signals,and effectively interfere with advanced radar systems.In order to improve the survivability of radar,an anti-DRFM spoofing signal is designed.Firstly,the principle of DRFM jammer is analyzed,and a new chaotic system is designed by improving the traditional Chebyshev chaotic mapping.Then,the Linear Frequency Modulated Continuous Wave (LFMCW) is modulated between pulses by using a new chaotic system,and an anti-DRFM spoofing signal is designed.Finally,the randomness and complexity of the improved chaotic mapping and the anti-interference performance of the designed signal are verified by simulation analysis.The simulation results show that the improved chaotic mapping has strong randomness and high complexity,and the designed signal has good anti-interference performance and anti-DRFM spoofing effect.