A Multi-Agent Deep Reinforcement Learning Approach with Communication Resource Optimisation for Cooperative UAV Swarm Perception and Multi-Target Detection in Emergency Rescue

The rapid localisation of survivors in disaster-affected areas remains a critical challenge due to complex terrains and limited communication infrastructure. Existing multi-UAV cooperative search methods suffer from insufficient perception of target distribution, inefficient use of scarce communication resources, and redundant coverage that degrades exploration efficiency. To address these issues, this paper proposes a multi-agent deep reinforcement learning framework based on the Multi-Agent Proximal Policy Optimisation (MAPPO) algorithm under centralised training with the decentralised execution (CTDE) paradigm. A detection-augmented observation mechanism is designed to encode target distribution information without introducing additional trainable parameters. A lightweight mean-pooling communication strategy is developed to optimise communication resource utilisation, reducing per-agent bandwidth to 64 bytes per step while preserving effective inter-agent coordination. Furthermore, a composite reward function is constructed to balance target detection, exploration, and redundancy suppression. Simulation results demonstrate that the proposed method achieves a recall of 0.70 and improves flight efficiency by 28% compared with the Grid Search baseline, while consuming over two orders of magnitude less communication bandwidth than representative feature-sharing approaches. The results support the use of the framework as a communication-resource-efficient solution for cooperative UAV swarm perception in bandwidth-constrained emergency rescue scenarios.