Abstract: Due to the complex and dynamic underwater environment, underwater unmanned systems face challenges of unpredictability and incomplete perception, which makes it difficult for them to accurately and efficiently accomplish autonomous decision - making tasks. Traditional methods highly rely on complete perception data and map information. However, limited by the dynamic characteristics of the underwater environment, it is difficult to construct effective map information in real - time, thus leading to limited efficiency of underwater unmanned systems in executing tasks such as underwater detection, resource exploration, and environmental monitoring. To address the above challenges, this paper proposes a double-layer decision-making method based on expert knowledge and deep reinforcement learning. This method can effectively enhance the adaptive ability of unmanned systems in underwater intelligent decision-making and significantly improve the efficiency of task execution. Specifically, an autonomous decision-making strategy generation method is first proposed to enhance the adaptive ability of underwater unmanned systems in unknown scenarios, further strengthening their autonomous decision-making level in complex environments. Secondly, a double-layer autonomous decision-making method is put forward. By enhancing the robustness of the system, it effectively ensures navigation safety. Finally, a multi - module design method is proposed to achieve the decoupling of each functional module, effectively improving the research and development efficiency of underwater unmanned systems. Taking the unmanned underwater vehicle (UUV) as the research object, experimental results show that the success rate and the convergence speed of the average reward value of the method in this paper outperform various benchmark methods in the simulation scenarios of UUV autonomous navigation and obstacle avoidance, providing a solid theoretical support for autonomous decision - making in real-world scenarios.