Because of the shortcomings of traditional waterway inspection in coverage, response speed and resource utilization, this paper puts forward an algorithm for dynamic change detection and unmanned inspection decision optimization of waterway driven by remote sensing images, and constructs an intelligent inspection system with closed loop of perception-decision-execution. Firstly, ST-CDNet, a multi-source data change detection model integrating optical and SAR images, is designed, and the spatio-temporal attention mechanism is introduced to realize high-precision and multi-scale dynamic perception of waterway water, coastline and obstacles. Secondly, the unmanned patrol decision algorithm DP-DDPG based on reinforcement learning (RL) is proposed, which takes the change detection results as the state input, combines the environmental context information, and optimizes the patrol path through the dynamic priority experience playback mechanism to realize the dynamic deployment and efficient utilization of patrol resources. The experiment is based on the inland waterway remote sensing data set from 2022 to 2024. The results show that ST-CDNet achieves 0.911 and 0.837 in F1 score and IoU index respectively, which is superior to the existing mainstream methods. The average response time of DP-DDPG algorithm in simulated patrol task is shortened to 73 minutes, and the total value of change discovery is significantly improved. This study effectively breaks through the bottleneck of the separation of change detection and inspection decision-making, and provides a feasible technical path for intelligent waterway management.
Dongliang Cao (Sun,) studied this question.