动态水声环境中的 SE_ResNet 模型目标识别方法

The SE_ResNet network, which uses an adaptive weighting method with convolutional operation channels to learn different feature information to increase the robustness of the network and make it adaptive to recognition targets in different sound field environments, was proposed to solve the crucial issue of ensuring superior recognition stability in the absence of sea area information. Experiments were performed on the basis of two different sets of measured underwater acoustic datasets. Experiment 1 showed that the SE_ResNet network outperforms other networks in the recognition task within the range of -20 dB to 20 dB signal-to-noise ratio (SNR). Experiment 2 illustrated that the SE_ResNet network has a high recognition rate for data with low interclass similarity at high SNR and has an excellent recognition effect on datasets with high similarity. The results showed that the proposed SE_ResNet network has a strong generalization capability.