Research on Seismic Signal Classification Based on Time-Frequency Map and Deep Learning

Abstract

Time frequency analysis is an effective method of processing non-stable signals, and this paper attempts to introduce the combination of time-frequency graph and deep learning neural network to study the classification of seismic signals. The short-term Fourier transformation, wavelet transformation and S transformation were used to obtain the timefrequency graph in seismic signal processing, and the good performance of image feature extraction ability was used by deep learning neural network, and Resnet50 was used to classify the results. The combination of timefrequency graph and deep learning neural network can effectively classify seismic signals, and from the classification accuracy, the combination of S transformation and Resnet50 network reached 89.82%, which is more superior.