Classification Method for Ship-Radiated Noise Based on Joint Feature Extraction

In order to address the problem of poor recognition performance from single signal features in ship identification and to enhance the accuracy of Convolutional Neural Networks (CNNs) in underwater acoustic target recognition, this paper proposes a method of joint feature extraction for ship target identification, combining energy features extracted from wavelet decomposition and frequency domain features extracted from Mel filters. Subsequently, two types of CNNs are constructed to train the joint features, evaluating the recognition performance of the joint features for ship targets. Through result analysis, it is found that the joint features can effectively identify ship targets. When compared to the recognition performance of the single Mel frequency domain features, the recognition accuracy of the joint features is significantly higher, providing a useful reference for underwater acoustic target recognition.