Underwater acoustic target recognition method based on CT and residual CNN

This paper presents an underwater acoustic target recognition method using Chirplet transform (CT), with the residual convolutional neural network (CNN) as the classifier. The method involves decomposing signal by the empirical mode decomposition (EMD), denoising based on principal component analysis (PCA) algorithm, extracting CT spectrum features of underwater acoustic targets, and using a ResNet18 model for recognition. The results of different models, including support vector machine, ordinary CNN, VGG19, and ResNet18 are compared. The results show that the denoise method based on PCA can effectively reduce noise and redundancy. Compared to other features, the recognition accuracy of CT spectrum is better. The CT-ResNet18 achieves the best recognition performance. While this method is used in ship recognition, it can be applied to other target voice recognition, such as marine mammals.