Classification method of infrasound events based on the MVIDA algorithm and MS-SE-ResNet

Abstract

The verification of nuclear test ban necessitates the classification and identification of infrasound events. The accurate and effective classification of seismic and chemical explosion infrasounds can promote the classification and identification of infrasound events. However, overfitting of the signals of seismic and chemical explosion infrasounds easily occurs during training due to the limited amount of data. Thus, to solve this problem, this paper proposes a classification method based on the mixed virtual infrasound data augmentation (MVIDA) algorithm and multiscale squeeze-and-excitation ResNet (MS-SE-ResNet). In this study, the effectiveness of the proposed method is verified through simulation and comparison experiments. The simulation results reveal that the MS-SE-ResNet network can effectively determine the separability of chemical explosion and seismic infrasounds in the frequency domain, and the average classification accuracy on the dataset enhanced by the MVIDA algorithm reaches 81.12%. This value is higher than those of the other four types of comparative classification methods. This work also demonstrates the effectiveness and stability of the augmentation algorithm and classification network in the classification of few-shot infrasound events.