In the future intelligent war, new challenges for detection, perception, command and decision of air defense and anti-missile weapon systems are required and have been put forward. Focusing on intelligent perception countermeasure to empower future air defense and anti-missile combat equipment, the concept and connotation of intelligent perception were analyzed. Centered on the OODA(Observe, Orient, Decide and Act) ring and the typical combat process, the main forms of sensing countermeasures were elaborated, including warning and counter warning, detection and counter detection, and tracking and counter tracking. The core scientific problem was identified, which was feature concealment and recognition. The main application scenarios of perception countermeasure technology in the OODA ring were revealed. Finally, the key technologies that need to be developed to enhance the intelligent level of perception concealment for air defense and anti-missile weapon systems have been proposed.

Facing the future with a highly dynamic, strong confrontation and rejection combat environment, intelligence level has become the core factor restricting the UAV mission/environment adaptability and combat effectiveness. Focusing on autonomous decision-making and planning technology, which is a difficult, rapidly updated and broad application prospect, the main research achievements and future development direction of autonomous decision-making and planning technology were reviewed. Firstly, the necessity of autonomous decision-making and planning technology research was elaborated based on the development and current situation of unmanned aerial vehicles. Secondly, this study summarized and analyzed the current research results of autonomous decision-making and planning technology. Finally, the development direction of autonomous decision-making and planning technology of UAVs was prospected.

The DETR series networks based on the Transformer architecture keep pushing the boundaries of object detection accuracy and speed in computer vision. However, non-cooperative object detection applications based on infrared images face challenges because of environmental complexity and poor image quality. To solve this problem, a novel object detection algorithm with high detection accuracy was proposed in this study, utilizing the Deformable DETR as the baseline. Initially, an image enhancement module called CLAHE-GB was designed to enhance the image process on infrared images, and it was effectively integrated with Deformable DETR. Subsequently, the algorithm was pre-trained on a large-scale general dataset. Then, data augmentation and transfer learning methods were developed to retrain the parameters of the detection head network using a self-made dataset of small infrared images of aerial objects. Finally, a comprehensive result analysis was conducted. The results show that the proposed algorithm can successfully achieve promising image enhancement effects and detection accuracy on infrared image data.

Aiming at the issue of weapon target assignment for a many-to-many interception in the air confrontation environment, this study has proposed a multi-target intelligent assignment method based on reinforcement learning. Under the many-to-many interception engagement scenario, a mathematical model of target assignment was established based on the engagement posture evaluation. By introducing the concepts of target threat degree and interception effectiveness degree, the interception urgency of each target and the interception capability characterization of each interceptor were fully reflected, allowing a comprehensive evaluation of the engagement posture of the attacking and defending sides. Based on the target assignment model, the target assignment issue was built up using a Markov decision process and was trained to be solved by a reinforcement learning algorithm using deep Q-network. Relying on the self-learning and reward mechanism under environment interaction, the dynamic generation of optimal assignment schemes was effectively realized. A many-to-many interception scenario was created and its effectiveness was verified through mathematical simulation, and the result shows that the trained target assignment method satisfies the requirements of continuous and dynamic task assignment in many-to-many interception.

Using small drones to detect urban targets such as vehicles at low altitudes in cities has gradually become a mainstream means. Given the existing problems of low detection accuracy of single-mode detection networks affected by visible light detection, inability to work at night and the blurred edge of infrared detection targets in actual scenes, this paper has proposed a multi-modal UAV detection algorithm based on image fusion and deep learning network. Firstly, based on the DUT-VTUAV visible-infrared registration data set and TIF image fusion algorithm, a multi-mode fusion data set was built up. Secondly, by comparing the detection accuracy, speed and several parameters of the existing YOLO series network, the lightweight network YOLO v5n which was most suitable for the mobile deployment of UAVs was decided. Finally, a multi-modal fusion detection algorithm was produced by combining an image fusion algorithm and a target detection model. Comparative experiments on vehicle data sets successfully show that compared with single-mode detection, the detection accuracy of the proposed algorithm is effectively increased, with mAP up to 99.6%, and a set of visible-infrared image fusion detection can be completed within 0.3s, indicating the high real-time performance.

This paper has proposed an intercept guidance law based on hierarchical reinforcement learning to solve the three-dimensional maneuvering target intercept guidance problem with constraints of low acceleration ratio and bearings-only measurement. The aforementioned problem was initially modelled using a Markov decision process model, where a heuristic reward function was applied considering both the energy consumption and the missile-to-target line of sight (LOS) angular rate. Besides, the policy of two levels was built up with the lower-level policy generating the required guidance command and being supervised by subgoals that were instructed by the higher levels, allowing the convergence of the LOS angular rate and guaranteeing the successful interception against a maneuvering target. Simulation results have validated the superiority of the proposed method compared with the augmented proportional navigation guidance law in terms of intercept accuracy and hit probability, and its required acceleration ratio is much lower.

The hypersonic vehicle has the characteristics of large maneuverability, high speed, long-range and strong threat. It has become a critical weapon development in many countries, and the identification of its motion behavior can significantly support defense and interception. In this study the hypersonic vehicle motion behavior recognition problem under the complex constraints of a no-fly zone was investigated. First, parametric motion equations that can fully describe the behavioral patterns of hypersonic vehicles were developed by analyzing the motion characteristics of hypersonic vehicles， and the trajectory planning problem under the complex constraints was acquired by optimization. Then, the hypersonic vehicle motion behavior recognition algorithm was designed based on deep learning. Finally, simulation experimental results were achieved and analyzed, presenting the effectiveness and generalization ability of the proposed algorithm.

Focusing on the difficulties in formation position control caused by uncontrollable axial overload in the glide phase of a hypersonic glide vehicle, a cooperative formation control method using consistent formation control and prediction correction theory was proposed. According to the cooperative guidance architecture, the distributed cluster formation control strategy was developed based on the multi-agent consistency theory. The coordination variables were acquired using the absolute and relative positions of each aircraft in the formation, while the basic formation control commands were achieved using the dynamic inverse method. The lateral manoeuvre control command was established to adjust the formation, and the two-stage predictive correction control algorithm was applied to realize the lateral manoeuvre control of the formation. The simulation results show that the proposed method can realize formation generation and maintenance of the underactuated hypersonic vehicle cluster, and performs significant formation adjustment ability and robustness to initial position errors.

3. Aeronautics Computing Technology Research Institute， Xi’an 710119， Shaanxi， China） Abstract: Reconfiguration is widely used by integrated electronic systems to enhance its fault tolerance and stability. It involves transforming a system from a faulty state to a normal state using a series migration actions based on a pre-defined reconfiguration blueprint after fault occurred. Considering the existing functional diversification and structural complexity of integrated electronic systems, it is crucial to enhance the fault tolerance and stability of the system. However, the current manual reconfiguration and conventional reconfiguration algorithms, two methods for designing reconfiguration configuration blueprints, are challenging to the fault tolerance and stability requirements of integrated electronic systems. This study has integrated the deep reinforcement learning algorithm to determine the reconfiguration blueprint model for the integrated electronic system fault situation and has proposed the Prioritized Experience Playback-based Competitive Deep Q-Network algorithm (PEP_DDQN). Utilizing the prioritized experience playback mechanism and SUMTREE's batch sample extraction technique, the proposed algorithm has built a competitive deep Q-network reconstruction algorithm based on deep reinforcement learning. Experiment results demonstrated that the PEP_DDQN method can outperform traditional reinforcement learning Q-Learning and DQN algorithms in generating higher-quality blueprints. It also exhibits better convergence performance and solution speed.

Focusing on path planning with designated point constraints in unstructured environments, this study has proposed a two-stage solution method which can improve the algorithm for each stage. In the first stage, map modelling was commissioned on typical unstructured environments. To solve the problems of contacting obstacles and winding paths in the A-Star algorithm, a new obstacle safety distance method was established and a line optimization strategy was developed to smooth the path. In the second stage, the processes of solving the designated-points problem were elaborated in detail, modelling a travel salesman variant problem and extending various optimization algorithms to this scenario. Due to the existing methods’ difficulty in effectively solving designated-point problems, an improved Genetic Particle Swarm Optimization (IGPSO) algorithm was proposed, including a hierarchical random initialization, an improved crossover method, and a mutation operator. Finally, comparative experiments were conducted where the significant advantages of the improved algorithm in optimal solution success rate, running time, and number of iterations were verified.