RIASSUNTO
This paper analyses the time-frequency spectrum characteristics of a diver’s inspiration signals collected by a passive sonar and presents a passive detection method of divers by taking the diver’s inspiration signals as the keywords of a detection model, which is built by data synthesis and convolutional neural network (CNN), gated recurrent unit (GRU), and other deep learning algorithms. Through combining with the periodic characteristics of the diver's respirations, the method provided in this paper can detect the inspiratory signals quickly and realize the hierarchical recognition of divers, which is helpful to the rapid response of a diver warning system. The feasibility of this method is verified by a sea experiment, and it can be used for reference in the study of passive diver detection.