Ship collision avoidance is the primary problem to be solved in intelligent navigation. In situations where multiple ships encounter each other, only by collaborating and jointly planning collision avoidance strategies can collision risks be effectively reduced. The navigation environment of ships is complex and ever-changing, and the fusion process of multi-source visual information collected by multimodal visual cameras is complicated. The feature extraction is not precise enough. Based on this, this article designs an intelligent collision avoidance system for ships based on multi-source fusion perception and deep collaborative decision-making. The system combines Mask R-CNN (Mask Region based Convolutional Neural Network) and ResNet in visual perception to accurately extract target features from image information, achieving high-precision and low latency target recognition under harsh weather conditions. In terms of collision avoidance decision-making, a ship intelligent collision avoidance decision-making model based on deep reinforcement learning (DRL) algorithm is proposed, which is based on adversarial dual deep Q-learning (Dueling DQN) and the establishment of ship domain models. When designing the reward function, factors such as collision avoidance rules (COLREGs) are fully considered to ensure the compliance and rationality of collision avoidance decisions. Simulation experiments show that the system has improved in both object detection and decision response.
Li et al. (Sun,) studied this question.