To improve the real-time and adaptive performance of intelligent automobile trajectory tracking,a variable universe fuzzy adaptiv model predictive control (VUFAMPC) method is proposed.The strong tracking square root cubature Kalman filter (ST-SRCKF) algorithm is used to estimate the lateral force of tire to obtain the real-time tire lateral stiffness values.And then,based on the traditional model predictive control method of trajectory tracking,a variable universe fuzzy control (VUFC) method is used to design a variable universe fuzzy model predictive controller (VUFMPC),and the estimated tire lateral stiffness is used as the parameter for VUFMPC to achieve the adaptive correction of controller parameters by combing it with ST-SRCKF algorithm,thereby obtaining VUFAMPC.Last,the comparative analysis and verification are made through hardware-in-the-loop experiment.The experimental result show that,compared to FMPC and AMPC,the overshoot of VUFAMPC is reduced by 11.1% and 18.8%,and the transition time is reduced by 46.6% and 67.9%.The trajectory tracking test shows that,compared to FMPC、VUFMPC and AMPC,the maximum tracking error of VUFAMPC is optimized by 73.9%,68.7% and 24.8%,and the average tracking error is optimized by 72.3%,56.1%,28%.The conclusion indicates that VUFAMPC has better real-time and adaptive performance,while effectively balancing the trajectory tracking accuracy and driving stability of intelligent automobile.

The negative obstacles pose a significant threat to the driving safety of unmanned ground vehicles (UGVs) when they operate in unstructured environments.However,in the existing research on negative obstacle detection,the camera-based detection methods are not well suited for the detection of negative obstacles in unstructured environments with poor lighting conditions and complex backgrounds; and LiDAR(light detection and ranging)-based detection methods are mostly designed based on the characteristics of mechanical LiDAR harnesses and have poor universality.Therefore,a highly compatible and scalable negative obstacle detection method is proposed,which can adapt to different types of LiDAR and achieve high-precision negative obstacle detection.The proposed method involves preprocessing the point cloud data,extracts the ground regions of interest,and performs point cloud pose correction.The adaptive resolution polar coordinate rasterization technology is used to enhance the spatial representation capability of point cloud data.A negative obstacle grid feature descriptor is designed,and the potential negative obstacle regions are extracted from multiple features such as the hollow characteristics,height differences,and minimum height of point clouds.A multi-frame fusion strategy is introduced,and the high-precision negative obstacle surface occupancy range is outputed through map reprojection and Bayesian rule-based probability updates.The experimental results show that the proposed method is applicable to LiDARs with different scanning modes,and can effectively identify the negative obstacle areas in complex unstructured environments.

The underwater unmanned vehicle (UUV),as an intelligent underwater operational system,generates a noise during operation,that can significantly interfere with the flank array sonar.The current mainstream adaptive beamforming techniques are primarily designed to suppress the far-field interference and have insufficient adaptability to the strong near-field interference caused by UUV self-interference.To address this issue,this paper proposes a UUV self-interference suppression algorithm based on covariance matrix reconstruction.The covariance matrix of the UUV self-interference exhibits high temporal stability under constant operating conditions due to the rigid connection between the flank array and the UUV.Leveraging this characteristic,the peoposed algorithm first estimates the self-interference covariance matrix with pre-collected data,and then utilizes this matrix to modify the covariance matrix of far-field interference and noise.Consequently,an interference-plus-noise covariance matrix is constructed to compute the weights of the adaptive beamformer.Experimental results indicate that,compared to current mainstream adaptive beamforming methods,the proposed algorithm performs better when the input signal-to-noise ratio exceeds 0dB.Moreover,compared to matrix filtering methods,it can enhance interference suppression effect by approximately 6dB.

To solve the problem of power distribution in unmanned diesel-electrical power system，this paper proposes an improved constant power control strategy of inverter.The active power control command of the strategy is obtained by calculating the frequency deviation and droop coefficient，and its reactive power control command is obtained by calculating the voltage deviation and droop coefficient.The control strategy can be used to directly introduce the dynamic changes in the frequency and voltage of diesel-engine generator into the inverter control，equally distribute the power to the inverter and diesel-engine generator when operating in parallel by properly setting the droop coefficients，and deal with the generator overload or reverse power caused by sudden loading or unloading in unmanned systems.Simulation and experiment verify that the proposed control strategy is feasible and effective.

For the collaborative optimization problem of multiple unmanned aerial vehicles (UAVs) jamming the ground communication networks，this paper aims to minimize the maximum power of multiple jamming UAVs under the constraint of suppressing the entire communication network.The complex interference resource allocation problem is transformed into a traditional combinatorial optimization problem with constraints.A hybrid particle swarm optimization (HPSO) algorithm based on particle swarm optimization(PSO) is designed，which simultaneously optimizes the three-dimensional positions of UAVs and the jammer power by treating them as particle positions，and incorporates a weapon threat model. HPSO algorithm，sparrow search algorithm and genetic algorithm arecompared through simulation experiment.The results show that HPSO algorithm has faster convergence speed and can efficiently find suitable UAV deployment solutions.In mission scenarios with different numbers of UAVs，HPSO algorithm demonstrates effective jamming performance and strong stability，making it a more suitable algorithm for optimizing the task of multi-UAV cooperatively jamming the ground communication networks.The research findings provide an effective solution and algorithm for multi-UAV cooperatively jamming the ground communication networks，which is of great significance for enhancing unmanned cyber-electronic warfare capabilities.

Rotary-wing unmanned aerial vehicles (UAVs) have been widely used in various fields such as power inspection，agricultural and forestry protection，and counter-terrorism rescue operation due to their exceptional flexibility and multifunctionality.However，the two-dimensional coverage issue caused by irregular terrain in complex planar areas presents significant challenges for path planning.To address this issue，this paper proposes a coverage path planning technology for UAVs in complex planar regions.During the region decomposition phase，an improved boundary-cutting decomposition (BCD) method is utilized to incorporate a directional tolerance parameter Δφ during the merging of sub-regions to eliminate the redundant elongated areas generated by decomposition，which significantly reduces the frequency of backtracking in subsequent path planning.In the coverage path generation phase，four innovative coverage modes (FF/FR/RF/RR) are designed，and the endpoints are dynamically configured.A multi-mode traveling salesman problem (MM-TSP) is constructed between sub-regions.By combining a candidate arrangement set Π_ϵ generated from a relaxed TSP with a dynamic programming strategy，the computational complexity is reduced from O(n！·4ⁿ)to O(|π_ϵ|·n·4²)，enhancing both computational efficiency and task coverage efficiency.Experimental results demonstrate that this coverage path planning technology reduces the average number of waypoints by 27.24%，shortens task distance by 26.12%，and decreases estimated task time by 26.20% across five scenarios：urban，airport，mountainous area，city，and forest.These findings validate the significant advantages of this technology in reducing the path redundancy and improving the task efficiency in complex planar regions，providing an efficient solution for the coverage path planning of rotary-wing UAVs.

The complex nature of missions executed by the Armed Police Force imposes higher requirements on the precision and control of strike effects of firearms.This paper focuses on the design and key technologies of intelligent non-lethal electromagnetic constant kinetic energy strike weapon systems.A “human-weapon integration” precision strike model is constructed by integratiing artificial intelligence，electromagnetic launch，and constant kinetic energy strike theories to achieve the dynamic control of non-lethal kinetic energy output.A hierarchical framework and operational mechanism based on the “perception-decision-execution” architecture are proposed by reviewing the global development status of intelligent firearms and analyzing the design necessity and intelligent connotation of weapon systems.Key technologies including line-of-sight stabilization，trajectory computation with terminal energy threshold，and electromagnetic launch control are thoroughly analyzed.The future development directions and potential contradictions of weapon systems are presented，transitioning from theoretical research to practical combat applications and providing effective support for equipment upgrading and operational effectiveness enhancement of Armed Police Force.

To address the challenges of extreme scale variations，dense occlusion of small targets，and complex background interference in unmanned aerial vehicle （UAV）-based target detection，this paper proposes a cross-layer dynamic detection network based on an improved YOLOv10 for the detection of small target via UAV aerial photography.A dual-branch cross-layer feature fusion pyramid network for replacing the original pyramid network is designed to resolve the problem of insufficient detail preservation for small targets in traditional methods.A channel-shuffling depth-wise upsampling module is developed，which combines channel shuffle operations with depth-wise separable convolutions and enhances the edge features of small targets through high-frequency residual connections.An end-to-end dynamic detection head is adopted to replace the original detection head，and a dynamic weighting mechanism is introduced，which enables the adaptive adjustment of feature representations at each position based on contextual information.Experimental results show that the proposed detection network achieves mAP0.5 of 53.3% and mAP0.5：0.95 of 33.2% on the VisDrone 2019 validation set，which are improveed by 12.7% and 9% ，respectively，compared to YOLOv10s，while reduces the model parameters by 23.7% and achieves an FPS of 79.The proposed algorithm significantly enhances the detection accuracy while maintaining excellent inference speed.

In complex geographical environments such as uncertain area，hostile environment and restricted area on the battlefield，conducting the search and rescue operation for persons in distress faces numerous challenges including low efficiency and poor positioning accuracy.How to utilize emerging equipment such as unmanned aerial vehicles （UAVs） to improve the efficiency of search and rescue operations for the persons in distress has become a research hotspot across the world.A searching and positioning UAV micro-swarm system composed of three multi-rotor UAVs is designed.This system is mainly composed of quadcopter UAVs，electro-optical search payloads，broadband ad hoc network radios，and UAV ground stations，etc.The day and night search and collaboration control are realized by single UAV and muliti-UAV collaboration.The miniaturized design of UAV makes it easy for the operator to carry with a backpack，facilitating forward reconnaissance to determine the locations information and the condition of the injury.The system can precisely locate the persons in distress and has the advantages of being less restricted by terrain and environment and having high search efficiency in the area.UAVs can achieve multi-mode networking communication among multiple unmanned platforms，and the ground control station has the functions of mission planning，setting the mission routes for multiple UAVs，and controlling UAVs to fly in formation，which can enhance the efficiency of battlefield search and rescue through the multi-UAVs collaboration.

In a typical integrated power system （IPS），the generator sets，transmission lines and distribution network are integrated in a relatively confined space，which leads to the tight coupling between the devices and accelerates the propagation of cascading failures.A risk assessment model of cascading failures in unmanned platform IPS is established based on pattern search theory and risk assessment theory，which is used to analyze the cascading failure of IPS under severe operating environment and frequent mode switching.Based on the model，a risk assessment is performed on a typical MVDC IPS，and the influences of different operation conditions and reliability parameters on the cascading failure risk are also analyzed.The proposed risk assessment model is validated through risk assessment and anlysis.Based on the risk assessment results，the weak links of the system are identified，and the corresponding prevention strategies are proposed by optimizing the equipment structure.The proposed cascading failure prevention strategies are verified through both simulation and experiment.

The quadruped unmanned combat platform holds significant military application value in future warfare due to its exceptional mobility and adaptability to complex terrains. A rigid-flexible coupled launch dynamics model is established to investigate the impact of shock loads on the vibration characteristics of the platform and firing accuracy. The amplitude, angular displacement and angular velocity variations of muzzle center point around x-axis and z-axis under different shock loads are analyzed through numerical simulation. The firing dispersion characteristics are evaluated using a six-degrees-of-freedom external ballistic model, and the live-fire tests are made on unmanned combat platforms with and without a bidirectional buffering device. The results show that the amplitude of the muzzle center point around the x-axis and z-axis during five-round bursts is significantly reduced, the vibration levels decreases, and the angular velocity tends to stabilize without the continuous increase observed in fixed connections after installing the bidirectional buffering device. The radius of 100% dispersion circle (R₁₀₀) is reduced to 86.4mm with a decrease of 34.6%. Live-fire test data indicates that R₁₀₀ for single-shot and five-round bursts is 75.7mm and 94.5mm, respectively, with the reductions of 21.1% and 32.8%. The test data are in good agreement with the simulated results, validating the accuracy of the numerical simulations. This confirms that the designed damping device effectively suppresses firing-induced vibration, significantly improving the firing stability and accuracy of the quadruped unmanned combat platform. The research findings provide technical support for the structural optimization design of unmanned combat platforms.

Addressing challenges in drone swarm damage assessment—such as unclear component-level damage mechanisms and insufficient understanding of formation configuration effects—this study proposes an evaluation method integrating high-fidelity component-level damage modeling with formation dynamics analysis. By establishing a damage calculation chain of “physical damage → component failure → functional damage” for quadrotor drones, the structure-effect relationship of target vulnerability is analyzed. Combined with a fragmentation-explosive warhead blast field model, this quantifies the dynamic impact of different damage elements on drone functionality. Furthermore, considering four typical swarm formations (one character, V character, snake, round), multi-scenario simulations were conducted using the standard damage percentage criterion and warhead-target encounter conditions. Results indicate that: For single-drone targets, the number of effective fragment hits negatively correlates with burst distance; Blast waves demonstrate superior damage efficacy at close ranges, with detonation below the drone yielding optimal results; Swarm damage outcomes are influenced by both detonation position and formation type, with formation being the dominant factor—circular formations sustain the highest damage, while serpentine formations exhibit the lowest. This research provides methodological support for damage assessment of fragmentation warheads against drone swarms and offers theoretical insights for tactical formation selection.

The traditional dung beetle intelligent optimization algorithm has good global search capability,but its performance is affected by the initialization parameter settings,which can lead to problems suchas blind spots in the local search,and non-communication between populations,etc.To address the problem of search blind spots in identifying the threats or no-fly areas for 3D trajectory planning of UAVs,a multi-strategy improved dung beetle optimization algorithm is proposed to improve the global trajectory planning capability.The initialization parameters,dung beetle ball-rolling behavior,small dung beetle foraging behavior and dung beetle stealing behavior are improved by using a novel chaotic mapping,a novel Cauchy-Lorenz wandering strategy,an improved triangular wandering strategy,and a novel Cauchy's inverse cumulative distribution function wandering strategy,respectively.The dung beetles of each population are crossed by using an improved longitudinal and transversal crossover strategy,and the ability of the UAV to identify the threat areas and the global trajectory planning is enhanced by the improvement of the multi-strategy.The results show the superiority of theimproved dung beetle algorithm in UAV trajectory planning.The total cost of the improved optimization algorithm is only 57.88% of the cost of the traditional dung beetle intelligent optimization algorithm,which is reduced by 42.12%.Compared with the total costs of the sand cat swarm algorithm,the particle swarm algorithm,the hippopotamus algorithm,and the gray wolf algorithm,the total cost of the proposed algorithm is reduced by 38.37%,38.80%,44.17% and 41.80%,respectively.

In view of the two typical failure modes of breechblock opening-closing mechanism for a naval gun,namely the wear of the key parts and the weakening of spring elasticity,the traditional fault diagnosis methods mainly rely on manual inspection,expert empirical reasoning and theoretical simulation.However,these methods not only take a long time for diagnosis,but also the diagnostic accuracy cannot be guarantee.In order to solve this problem,a fault diagnosis method of Gram angle field combined with convolutional neural network and long short-term memory neural network (GAF-CNN-LSTM) based on sparrow search algorithm (SSA) is proposed by using the deep learning algorithms.Firstly,the original fault signal of the breechblock opening-closing mechanism is collected and preprocessed by the test bench,and the one-dimensional time-series data and two-dimensional image fault dataset are established by the time-frequency analysis method and the Gramian angular field method.Then the fault dataset is input into the LSTM and CNN channels,respectively,and the powerful spatial feature extraction ability of CNN and the time-series feature ability of LSTM mining data are used to extract the features,and the feature informations obtained by the two abilities are fused to output the diagnostic results under the action of the fully connected layer and activation function.Finally,the SSA optimization algorithm is used to optimize the hyperparameters in the GAF-CNN-LSTM network structure to improve the diagnostic accuracy and applicability of the model.The proposed SSA-GAF-CNN-LSTM fault diagnosis model is verified by the test data.The result shows that the proposed fault diagnosis model can not only diagnose the fault type of the breechblock opening-closing mechanism for naval gun more accurately,but also has stronger generalization ability and anti-interference ability,which effectively improves the fault diagnosis performance of the breechblock opening-closing mechanism.

To address the issue of evading a medium-long range air-to-air missile in beyond visual range (BVR) air combat,a feasible maneuver strategy for unmanned aerial vehicle (UAV) possessing the capability of high overload lateral maneuver within a short period of time to evade the medium-long range air-to-air missile in terminal guidance is proposed based on the characteristics of UAV equipped with lateral rocket boosters.A model of UAV and medium-long range air-to-air missile is established.The influences of lateral overload activation time,lateral overload size and lateral overload direction on the miss distance of a missile are investigated.The experimental results show that the UAV utilizes the maximum available lateral overload within the sustainable time to evade the air-to-air missiles.When a missile is approaching head-on,UAV climbs to evade it,while the same direction climbing and diving maneuvers are used to against the missile approaching from the front.

Aiming at the problem that the stabilization system of unmanned vehicle-mounted guns is affected by the complex nonlinearity and random disturbances during moving,a composite control strategy combining active disturbance rejection adaptive control is proposed.The electromechanical coupling dynamic equations of the unmanned vehicle-mounted gun stabilization system, which take into accountconsidering the actuator dynamics and model uncertainties,are established.Based on the backstepping approach,the adaptive control is ingeniously integrated with the extended state observer (ESO),constructing the parameter adaptation laws to update the unknown parameters of the system online.A dual-channel ESO is used to estimate the matched and unmatched disturbances in real time and provide feedforward compensation.Since the parameter uncertainties of the stabilization system are mainly addressed by adaptive technology,the learning burden of ESO is further reduced,improving the tracking performance of the stabilization system during moving and avoiding the impact of high-gain feedback.The stability analysis based on the Lyapunov function indicates that the asymptotic control of the vehicle-mounted gun can be achieved when only constant disturbances are present,and even in the presence of time-varying uncertainties,the prescribed transient performance and tracking accuracy can still be ensured.Comparative co-simulations and simulation tests demonstrate the effectiveness and feasibility of the active disturbance rejection adaptive composite control strategy.

Trajectory tracking is a crucial functionality of the autonomous driving control system.The vehicle dynamics model has a significant impact on trajectory tracking performance,however,there is a conflict between model complexity and solving efficiency,often leading to insufficient tracking accuracy under nonlinear conditions.To address this challenge,this paper proposes a model predictive control method based on Gaussian process regression (GPR) for trajectory tracking.A simplified model is used to ensure solving efficiency,and GPR model is employed to compensate for the vehicle model,thereby enhancing the trajectory tracking performance.First,a vehicle state fusion estimation method based on the single-track dynamics model is developed to obtain the GPR compensation model.A trajectory tracking error model is developed.Based on the trajectory tracking error from vehicle dynamics model,the iterative equation for GPR error compensation within the predictive horizon is derived to dynamically compensate for model errors in the vehicle state prediction for achieving the trajectory tracking control.Finally,a real-vehicle validation platform is constructed to validate the proposed method under typical driving conditions.The proposed method is compared with other predictive control methods without GPR compensation.The results show the proposed method achieves a significant improvement in trajectory tracking accuracy.Specifically,the lateral and heading errors are reduced by 33.3% and 27.9%,respectively.Furthermore,the vehicle comfort performance is also improved,and the mean lateral acceleration and yaw rate are reduced by 17.1% and 21.7%,respectively.

Application of reinforcement learning in unmanned aerial vehicle (UAV) air combat faces the challenges of which the rigid reward functions and single models are used to handle the complex tasks difficultly in high-dimensional continuous state spaces.This severely limits the decision-making generalization capability in dynamic and of algorithm varied situations.Addressing the aforementioned issues,an autonomous decision-making framework with the deep double Q-network (DDQN) and deep deterministic policy gradient (DDPG) algorithms is proposed,which integrates the essence of hierarchical and distributed architectures.Based on the advantage differences between the opposing forces in various situations,a series of DDPG algorithm models with different reward function weight combinations are designed to construct a bottom-level distributed deep deterministic policy gradient (D³PG) decision-making network.The DDQN algorithm which excels in handling discrete action spaces is introduced to construct a top-level decision-making network.It allows for autonomous selection and switching to the most suitable bottom-level policy model based on real-time situation changes,thereby achieving the instant adjustment and optimization of decisions.To further enhance the realism and challenge of combat environment,a self-play mechanism is introduced into the DDPG algorithm training to construct an enemy decision-making model with high intelligence.The experimental results demonstrate that UAVs equipped with the proposed algorithm achieve a maximum win rate of 96% in adversarial engagements against intelligent opponents,which is increased by more than 20% compared to those of baseline algorithms such as D³PG.Moreover,it consistently defeats the opponents under various initial conditions,confirming the effectiveness and advancement of the proposed algorithm.

The traditional dung beetle optimization algorithm (DBO) exhibits the poor stability and insufficient optimization ability in the trajectory planning of unmanned aerial vehicles (UAVs) in complex environments,DBO Optimization Algorithm with Group-based Optimization and Adaptive t-Distribution (GOTDBO) is proposed.Based on the DBO algorithm,the GOTDBO algorithm combines the composite population initialization strategy,the adaptive disturbance global exploration strategy and the adaptive t-distribution disturbance strategy,effectively enhancing the global exploration and local exploitation capabilities of the algorithm and improving the convergence speed of the algorithm.The smoothness and safety of the trajectory are further optimized by constructing an objective function that comprehensively considers the total flight length,corner curvature and maximum flight direction change,and introducing the penalty function method to handle no-fly zones and other constraints in the path,the smoothness and safety of the trajectory are further optimized.Experimental results show that,in terms of the flight range,When the GOTDBO algorithm is applied to route planning in scenarios with different complex environments,it can plan compact and efficient routes,performs excellently in terms of maximum range,and effectively improves the economy of endurance.In terms of threat avoidance,the trajectory planned by the GOTDBO algorithm has the least number of approaches to threat areas,thus ensuring higher flight safety.In terms of altitude control,the degree of altitude deviation is low,enabling stable and accurate altitude control.Although the GOTDBO algorithm is comparable to other algorithms in the trajectory smoothness,it has significant advantages in multiple core indicators.It is energy-saving and efficient,safe,and reliable in UAV trajectory planning,and has high application value and broad prospects.

Unmanned aerial vehicle (UAV) formation can execute complex collective tasks and reduce the risk and operation difficulty of a single UAV. A distributed formation compound control method with the leader-follower structure is designed based on the prescribed-time stability theory and multi-agent consensus theory for the control of UAV swarm formation in three-dimensional scene.Firstly,a multi-UAVs kinematic model is established by analyzing the relationship between the actual input and the equivalent control.In order to enhance the robustness of UAVs against external disturbances,a prescribed-time convergent extended state observer is designed to achieve online estimation of disturbance based on active disturbance rejection control theory.Furthermore,considering that only the followers connecting to the leader can access the leader's state information,a prescribed-time convergent distributed estimator is introduced to rapidly estimate the leader's state.On this basis,a prescribed-time convergent consensus formation control algorithm is proposed combined with the outputs of the observer and the estimator,and the prescribed-time stability of closed-loop system is proved by Lyapunov theory.The simulated results validate the effectiveness of the proposed method.The research results show that the proposed control method can achieve the stable cooperative control of UAV formation within a preset time in the presence of external disturbances.

As a new type of aircraft,the unmanned aerial vehicle (UAV) is gradually integrating into the modern weaponry system and becoming an indispensable and important part of the military field.In order to equip UAV with a safe landing decision-making system that can autonomously perform landing tasks without ground marking,this paper proposes a phased autonomous location selection technique based on multi-sensor data fusion from coarse to fine.The rough landing point search is realized based on the semantic segmentation of the image information.After guiding the UAV to reduce the flight altitude,the terrain parameters are calculated from the elevation value of point cloud information to construct a terrain cost map,and the semantic information of the image is fused by considering the category of a terrain to complete the fine landing point search.The experimental results show that the proposed location selection technique can well delineate the safe and dangerous areas,and enables UAVs to autonomously arrive at a safe landing position.Meanwhile,the comparative analysis of the decision-making in the fine landing point search stage with the fitted point cloud plane verifies that the technique can save decision-making time to a greater extent and improve the efficiency of location selection.

Amphibious vehicle is a mobile platform capable of operating in both terrestrial and aquatic environments,and has significant application value and development potential in both military and civilian fields.The development history of amphibious vehicles is reviewed,and the characteristics and development trends of different types of amphibious vehicles are compared and analyzed.The key technologies for the navigation of amphibious vehicles on water are expounded from three aspects:modeling and simulation,high-speed amphibious vehicle design,and navigation control.The difficulties and challenges in achieving the unmanned operation of amphibious vehicles on water are discussed based on the research progress of unmanned technology for amphibious vehicles,and the future research direction of amphibious vehicles is prospected.

Unmanned aerial vehicle (UAV) remote sensing detection plays an important role in military reconnaissance,and the polarization detection is to utilize the polarization changes generated by the interaction between polarized light and target to improve the target contrast.However,in complex scenes,the small targets are less distinguishable from the background due to their similar features and the insufficient spatial information,resulting in difficulties in detection.To this end,a polarization camouflaged small object detection (PCSOD)-YOLO algorithm is proposed,and an efficient layer attention module-coordinated attention (ELAM-CA) and a spatial pyramid pooling cross stage partial channel-3D weights attention (SPPCSPC-3DWA) module are designed to capture the polarization features and semantic information of target,enhancing the ability to understand the contextual information.A dynamic small target detection head is designed to enhance the ability to extract the features of small targets through dynamic convolution,and the detected results of small target are outputted using the feature information from different scales and the multi-channel feature information.A polarization image of camouflaged small objects (PICSO) dataset is constructed for the camouflaged small target polarization images.Experiments on the PICSO dataset show that the proposed method can effectively detect the camouflaged small targets,with mAP_0.5 and mAP_0.5:0.95 reaching 92.4% and 47.8%,respectively.The detection rate reaches 60.6 frames per second,meeting the real-time requirements.

Fuzzy cognitive map (FCM),as a kind of knowledge graphical soft computing model with both fuzzy reasoning and neural network-like features,is highly compatible with the development direction of the third generation of artificial intelligence driven by both knowledge and data,and has been widely used in various fields.Firstly,the basic concepts and principles of FCM theory are introduced,the latest research progress and major existing problems of FCM are systematically analyzed from the three aspects of construction method,learning algorithm and extension model,and the main directions and key contents of the future research are summarized.Secondly,the application research of FCM in unmanned systems (single unmanned systems,multi-unmanned systems,unmanned-manned systems) are comprehensively reviewed and summarized.Finally,the key research contents and main research ideas of FCM in unmanned systems applications in the future are discussed in depth by analyzing in detail the technical requirements of single unmanned systems driven by autonomous intelligence,multi-unmanned systems driven by swarm intelligence,and unmanned-manned systems driven by mutual trust intelligence.

A multi-dimensional enhanced particle swarm optimization algorithm (MDEPSO) is proposed to address the problem of insufficient global search capability and susceptibility to local optima in the 3D trajectory planning process of unmanned aerial vehicles using classical particle swarm optimization algorithms.This algorithm first introduces improvement factors to dynamically adjust inertia weights in various stages of particle optimization,enhancing population adaptability and overcoming local optima; Secondly,relying on dynamic constraint equations to enhance learning factors promotes more efficient information sharing between particles and improves the algorithm’s self-learning ability; Subsequently,the advantages of orderly integration of chaos initialization and elite reverse learning evolution strategies were utilized to re plan the particle swarm evolution process,enhance the balance and diversity of particles in the iterative process,and improve the convergence accuracy of the algorithm.In the experiment,through horizontal comparison of test functions and vertical application in complex 3D task scenarios,the multi-dimensional enhanced particle swarm optimization algorithm showed an improvement in the UAV trajectory planning ability compared to the classical particle swarm algorithm in the new multi-dimensional objective function indicators.It demonstrated good effectiveness and competitiveness among the five comparison algorithms.

In air-ground collaborative systems,the coordinated landing of unmanned ground vehicles (UGVs) and unmanned aerial vehicles (UAVs) is of paramount importance for extending the task scenarios of heterogeneous intelligent agent clusters.Current trajectory optimization-based autonomous landing methods couple the temporal and spatial dimensions by designing the optimal control laws for joint trajectory optimization.However,the optimization objective function design is relatively complex,and it cannot fully utilize the actuator’s optimal performance.A novel spatio-temporal decomposition planning (STDP) method is proposed to address the excessive coupling of time and space in traditional trajectory optimization methods.The STDP method optimizes the landing trajectories separately in spatial and temporal dimensions,enabling UAVs to adopt more aggressive flight strategies in complex scenarios.Furthermore,the objective function is meticulously designed to account for the UAV’s landing time and motor power consumption model,formulating a second-order cone programming problem to expedite the solution process while ensuring high-quality and efficient solutions.Simulated results indicate that,compared to spatio-temporal coupled planning methods,the STDP method generates the trajectories that closely adhere to kinematic constraints,substantially reducing the task completion time and enhancing the mission efficiency.Additionally,the empirical tests in real-world scenarios confirm the reliability and efficacy of STDP method in practical application.

A dynamic imaging model for UAV(Unmanned Aerial Vehicle)-borne line-array LiDAR is proposed to address the scarcity of point cloud datasets in UAV imaging scene.A pillar-voxel-based point cloud target detection algorithm is proposed to verify the authenticity of the simulation data.The typical target models and the typical scenarios with rugged terrain,camouflage,and vegetation cover are established based on a virtual simulation platform.A laser point cloud dataset is generated by using the point cloud simulation model,UAV motion model,stitching imaging and distortion correction methods.The dataset is annotated by using a completeness judgment method based on overlapping areas.A pilar-voxel feature extraction module is used to process the target’s top features.The point cloud target detection algorithm is trained using the annotated simulation dataset,and evaluated on a real dataset obtained from equivalent experiments.The proposed algorithm achieves an accuracy rate of 93.2% on the real dataset.The evaluated result indicates that the simulated data effectively reflect the true characteristics of the targets,and the dynamic imaging model has high credibility.

The near-space hypersonic gliding vehicle (HGV) poses a significant threat to existing defense systems due to its ultra-high velocity,extreme maneuverability,and superior penetration capabilities.To address the challenges in tracking and predicting HGV trajectories during interception,this paper presents an intelligent trajectory prediction method based on aerodynamic acceleration estimation.The maneuver patterns and aerodynamic variation laws of HGV are systematically analyzed according to the HGV motion model.On this basis,three critical parameters,i.e.,aerodynamic lift acceleration,drag acceleration and bank angle control,are identified as trajectory prediction variables for replacing the unknown terms in the HGV motion model.A dynamics tracking model based on aerodynamic acceleration estimation is developed to use the radar measurement data and the unscented Kalman filter (UKF) for real-time tracking and estimation of these parameters.These estimated parameters are then used as inputs to train a long short-term memory (LSTM) network,which captures the temporal relationships and variation patterns in the prediction parameters.The trained LSTM network is used to iteratively forecasts future aerodynamic accelerations,which are integrated with the numerical solutions of motion equations to extrapolate HGV trajectories.Numerical simulations confirm that the proposed method achieves high prediction accuracy and robust stability in predicting the trajectories of non-cooperative HGVs.

Constructing an environment map is a crucial prerequisite for navigation,and a comprehensive and detailed map can effectively assist in planning the optimal motion paths for unmanned ground platforms.To address the issues of data redundancy and difficulty in distinguishing the terrain structures in traditional map construction methods,a lightweight map processing and staircase area classification method for indoor navigation of unmanned ground platforms is proposed.The method first extracts the traversable area for unmanned platforms to reduce data redundancy and removes the outliers based on the distribution characteristics of stair surfaces.Subsequently,a map construction algorithm incorporating edge smoothing is used to generate multi-layer grid maps with clear boundaries,regular shapes,and distinct levels.Then,the stair environment features are extracted,and a Naive Bayes classification algorithm with Laplace smoothing is employed to distinguish and label the structures such as steps and turning platforms.The experimental results show that the maps generated by this method maintain high resolution while reducing the data volume by an order of magnitude compared to traditional point cloud maps,and the macro-precision rate of map classification reaches 91.3%.Compared with conventional methods,the proposed method can construct more lightweight multi-layer grid maps with terrain classification labels,providing safe and efficient navigation support for unmanned ground platforms.

In order to improve the design ability of modified double base(MDB)propellant formulations, different intelligent algorithms were used for the formulation design and optimization of the burning rate property. By comparing several typical intelligent algorithms, including genetic algorithm(GA), differential evolution algorithm(DE), particle swarm optimization algorithm(PSO)and whale optimization algorithm(WOA), the intelligent formulation design and optimization of MDB propellants containing RDX were carried out. The results reveal that compared to the statistical optimization method, the intelligent optimization algorithm is simple to be operated and runs more quickly, while the 14 optimal formulations fitted by differential evolution algorithm have the best and more stable burning rate performance. Under the same pressures, the predicted burning rates of these 14 optimized formulations are significantly higher than the experimental values of the prepared formulations. Finally, a set of intelligent algorithm software for predicting the burning rate of modified double base propellants containing RDX is formed by coupling the difference algorithm with the support vector regression model. Based on the intelligent algorithm software, new formulations with the superior burning rate property could be deduced.

In order to ensure the safety of unmanned underwater vehicle (UUV) when performing the complex tasks such as coastline patrol, collision avoidance of large vessels, and traversing dense islands and reefs, an elliptical modeling obstacle avoidance method based on control barrier function (CBF) is proposed. A control barrier function containing the heading angle constraints is designed on the basis of elliptical modeling obstacles, a quadratic programming (QP) problem with constraints is constructed, and a closed-form obstacle avoidance guidance law is proposed by combining with the guidance law in obstacle-free environment. The simulated results verify the effectiveness of the proposed method, which globally satisfies the safety and stability requirements. The proposed method has practical application value for the safe navigation of unmanned underwater vehicle in complex marine environment.

For the inaccurate track cost estimation in task allocation for multi-agent systems, a track cost calculation method based on extended rapidly-exploring random tree is proposed to rationally plan the motion trajectories of agents and improve the accuracy of track cost estimation. In order to solve the problem of premature contracting of dominant agents in improved contract net algorithm, an agent bidding transformation mechanism is proposed to make the dominant agents participate in the task bidding for multiple times and achieve the balance of task load between agents in a system. The simulated results show that the proposed track cost calculation method can be used to accurately calculate the trajectory between agent and target, and the trajectory between target and target. The agent bidding transformation mechanism solves the resource waste caused by the premature contracting of dominant agent, and the time of the agents to complete all tasks is reduced by 6.54%. However, when dealing with the dominant agent problem, the new mechanism will increase the bidding rounds of the entire task allocation.

Timely and effective identification of aircraft flight patterns is crucial in monitoring task. However, the existing flight pattern recognition methods have limitations in practical applications due to strong subjectivity and single pattern, which limits the flight monitoring capability in complex situations, and in turn leads to imprecise pattern boundary positioning and low recognition accuracy. For this reason, a flight pattern intelligent recognition method based on sensitive boundaries and long flight sequences is proposed for the intelligent recognition of flight states. In order to better explore the spatial relationships of multi-modal flight parameters, an adaptive graph embedding is designed. A denoising depth multi-scale autoencoder is proposed for the flight patterns at different durations, as well as the classification-weighted focal point loss and regression-joint spatio-temporal intersection loss for mitigating model loss. In order to verify the superiority of the proposed method, the real parameters of several civil flights covering 11 flight patterns are collected, and a flight state dataset is constructed by manual labelling. The results show that the proposed model is able to automatically distinguish different flight patterns in consecutive flight sorties and accurately extract the mode boundaries without any pre-processing or post-processing, with an identification accuracy of 99.07%. The intelligent recognition method can effectively improve the recognition accuracy and the flight state recognition of sensitive boundaries.

In response to the significant reduction in target positioning accuracy caused by severe nonlinear factors affecting UAV electro-optical platforms, an algorithm based on improved mutant firefly algorithm-particle filter (IMFA-PF) is proposed for UAVs to accurately locate ground targets. Firstly, the state equations and measurement equations for target observation from UAV electro-optical platform are established. And then the IMFA-PF algorithm is utilized to estimate the geographic locatio of a target, and the interaction patterns among particles are altered by introducing multiple mutation strategies and an elasticity mechanism, thereby addressing the particle degradation issues caused by severe nonlinear factors and excessive optimization. Finally, the effectiveness of the algorithm is verified through a one-dimensional nonlinear unstable simulation system and actual flight experiments. Experimental results indicate that the proposed algorithm can improve the particle distribution’s resilience to observational nonlinearity and effectively tackle particle degradation issues, showing better robustness and positioning accuracy compared to the existing positioning methods.

The sea surface corner reflector exhibits extremely strong radar echo characteristics. It creates the false targets that interferes continuously in the time domain and generates a deceptive situation to bring a significant challenge to the precision strike capability of a seeker. To address this issue, the immunity of infrared sensors to corner reflector interference is leveraged, and an intelligent recognition algorithm for corner reflectors in multi-target scenarios based on radar-infrared feature-level fusion is proposed. The target interference in infrared images is preliminarily discriminated by YOLOv8 network. The high-confidence target images can directly output the recognition results. The low-confidence target images are individually cropped for target correlation using radar-infrared angular information and radar feature extraction. The radar features and infrared images are input into a dual-channel fusion network, achieving the secondary recognition of low-confidence targets. The measured data validate that the recognition accuracy of the proposed method exceeds 96%. The research work has significant reference value for corner reflector interference recognition.

In the context of large-scale unmanned aerial vehicle (UAV) swarm cooperative flight scenarios, the high computational time consumption in swarm path planning is caused by frequent path conflicts. Aiming at the problem above,a large-scale UAV swarm path planning method based on reinforcement learning conflict resolution is developed. A dual-layer planning architecture, comprising a high-level layer of conflict resolution and a low-level layer of path planning, is constructed to reduce the spatial and temporal dimensions of path conflicts. At the high-level layer of conflict resolution, a conflict resolution strategy network based on the Rainbow deep Q-networks (DQN) algorithm training framework is designed. This network transforms the resolution process of each path conflict into the action selection process of left and right tree nodes of a binary tree. This approach maps different conflict resolution sequences to their outcomes, thereby reducing the traversal of tree nodes and improving the efficiency of conflict resolution. At the low-level layer of path planning, the time dimension is incorporated into the spatial collision avoidance strategy. A re-planning jump point search (ReJPS) method based on a node re-expansion mechanism is proposed, which increases the feasible planning domain and enhances the ability to resolve the path conflicts. Simulated results indicate that, compared to the path planning methods based on the conflict-based search (CBS)+A^* and CBS+ReJPS, the proposed method reduces the average planning time by 86.64% and 19.65%, respectively, while maintaining comparable optimality.

In order to improve the obstacles avoidance ability of intelligent vehicles,an trajectory planning and control method of intelligent vehicles is proposed based on velocity obstacle model.The proposed method is used to establish a velocity obstacle model for intelligent vehicles by combining the velocity obstacle method and obstacle expansion method,The motion uncertainty of dynamic obstacles in the velocity space is transformed into the positional uncertainty,and the safety margin is adaptively adjusted by obstacle size and relative velocity.To balance trajectory tracking accuracy and driving stability,a fuzzy model predictive controller (FMPC) is designed based on the equation of state for vehicle,the fuzzy control principle and the model predictive control principle.A simulation model is established to verify the effectiveness of the proposed method.The simulated results show that the proposed method can be used to avoid the multiple random static and dynamic obstacles,and the reference trajectory can be quickly and smoothly tracked after obstacles avoidance.Based on the analysis of obstacles avoidance stability,it is concluded that the target speed is 100km/h,the maximum lateral speed is 4.01km/h,the maximum yaw rate is 20.8°/s,and the maximum centroid side slip angle is 2.32°,which meet the requirements of driving stability.The proposed method effectively improves the obstacle avoidance ability and driving stability of intelligent vehicles.

Aiming at the issue of optimal energy consumption of UAV time-varying formation under multiple constraints,an optimal control method based on an dynamic parameter extended high-order control barrier function is proposed.The method is used to improve the existing high-order control barrier function.In consideration of the motion states of UAV and obstacles,an the extended state dynamics model is established.The safety constraints of dynamic target obstacle avoidance are expanded on the extended state dynamics model,and the extended high-order control barrier function used to establish dynamic target obstacle avoidanceis obtained.For the feasibility and conservatism of the extended high-order control barrier function constraints,the adaptive rules of the class K function parameter are designed to obtain the dynamic parameter extended high-order control barrier function,which improves the feasibility and reduces the conservatism for the solution of constraint optimization problem.The control barrier function and the extended high-order control barrier function are used to establish the multi-constraints,and an objective function is established by the consistency condition and Hamiltonian analysis.The energy consumption optimization problem of time-varying formation is transformed into a multi-constraint optimization problem,and the quadratic programming is used to solve the problem to obtain the optimal control input.Simulated results demonstrate that the proposed method has superior obstacle avoidance and lower energy consumption compared to the fixed parameter high-order control obstacle functions and the artificial potential field.

For the interception issue of multiple intruders in area defense missions,a multi-UAV sequential capture algorithm is proposed by taking into account the temporal relationship between pursuit tasks and the overall interception effectiveness.The temporal and spatial rewards are constructed based on the long-term planning benefits and short-term execution effects of the tasks,which serve as the optimization objectives for task allocation and execution,respectively,and the dynamic and real-time solutions are achieved for complex game-theoretical problems.A reachability-set-based approach is used to describe the advantage levels of both attackers and defenders,and a deep Q-network is introduced to estimate the temporal rewards for tasks and then guide task allocation.The single attacker pursuit-evasion game problem is solved based on the spatial reward of task,and an optimal control strategy is presented for task execution in a continuous action space.Simulated results show that the peoposed algorithm optimizes the temporal and spatial rewards to facilitate the effective cooperation among multiple UAVs,enhances the capture success rate of the defenders,and has an increased scalability.

To solve the problems of slow convergence rate and continuous controller updates in formation control, this paper proposes a prescribed-time formation control method based on event-triggering mechanism for second-order multi-agent systems.Based on a prescribed-time acceleration observer, the followers can estimate the acceleration state of leader within the prescribed time.Moreover, a prescribed-time formation controller based on the event-triggering mechanism is designed to enable the followers to keep up with the leader within the prescribed time.The event-triggering mechanism proposed avoids the continuous updates of controller.Through the rigorous theoretical analysis, it is proved that the proposed method can be used to achieve the prescribed-time formation control for the multi-agent systems without Zeno behavior.Simulated results indicate that the proposed prescribed-time formation control method based on the event-triggering mechanism can make the multi-agent systems form a desired formation configuration within a preset time, and reduce the update frequency of controller to save resources.The feasibility and extensibility of the proposed control method are further verified based on the formation flight test results of quadrotor UAVs.