Method and system for identifying marine natural earthquake in real time through small-sample lightweight deep learning

The invention provides a small-sample lightweight deep learning real-time marine natural earthquake identification method and system, and the method comprises the steps: inputting sound pressure signals recorded by a snorkeling type ocean seismograph into a plurality of trained lightweight deep learning networks, and outputting whether a marine natural earthquake occurs or not; a voting mechanism is adopted for output results of the multiple networks, and the voting results serve as final recognition results; the training of the lightweight deep learning network comprises the following steps: preprocessing sound pressure signals recorded by a snorkeling ocean seismograph; adopting a supervised learning mode to carry out multiple times of model training; the preprocessing comprises the steps of uniformly sampling the data to a fixed frequency and a set length, and carrying out 0.05-3Hz band-pass filtering processing on the data. The method has the advantages that natural seismic signals can be accurately recog