This paper investigates noncooperative online games in multi-agent systems with time-delay feedback constraints. In these games, all players individually minimize their own time-varying costs, yet these costs are coupled through the joint strategies of the entire group. The gradient feedback available to the player is subject to unknown, heterogeneous delays that vary across players and time instants. To address these challenges, a distributed online gradient-descent algorithm is developed to seek the time-varying Nash equilibrium in the game. The algorithm is proven to achieve a sublinear dynamic regret bound via an appropriate learning rate, provided that the path variations grow sublinearly. Finally, an online electricity market game with time-varying delays is provided to illustrate the effectiveness of the algorithm.

To address the detection and assessment of military target damage, this paper proposes a two-stage damage assessment method based on deep learning and fuzzy analytic hierarchy process. Firstly, in the Yolov5-based dual-stage target detection subsystem with coordinate attention mechanism, a key region extraction mechanism is employed to extract damage components, and a damage component classifier based on the combination of intersection over union, Hungarian linear matching, and decision tree is used for classification and quantification of damage severity. Then, in the damage assessment subsystem based on the triangular fuzzy analytic hierarchy process, multiple damage assessment weight systems and damage tree criteria are designed to comprehensively consider the damage features and categories extracted from the previous stage, achieving online real-time damage assessment for targets. Experimental results show that on various simulated datasets with multiple interference factors, the Yolov5-based dual-stage target detection subsystem with the attention mechanism achieves an average improvement in detection accuracy of over 3.6% when compared to classical target fine-grained recognition algorithms. It demonstrates better performance in target key region extraction, and damage classification and assessment, providing strong support and reference for target damage assessment and military operation decision making.