Vessel Type Classification Utilizing Underwater Acoustic Data and Deep Learning

Identifying underwater vessels using acoustic data is crucial for various marine applications. Precise classification enhances maritime safety and supports marine ecosystem preservation, along with informed resource management. This paper proposes a lightweight convolutional neural network (CNN) designed to classify underwater vessels by analyzing their audio signals. Our methodology involves analyzing vessel acoustic data through continuous wavelet transform (CWT), which effectively extracts features by evaluating the time-frequency characteristics of the audio signals. These preprocessed features are then sent to a novel CNN that efficiently identifies the patterns and classifies the various underwater vessels. We comprehensively evaluate our model on the publicly available ‘VTUAD: Vessel Type Underwater Acoustic Data’ dataset, which consists of five classes. Furthermore, we benchmark our approach against several leading CNN architectures, including ConvNeX, DenseNet, Efficient-Net, InceptionNet, MobileNet, NASNet, ResNet, VGGNet, and Xception, in terms of accuracy, number of parameters, and model size. The proposed lightweight CNN model demonstrates significant efficacy, achieving an accuracy of 93.53% and a size of 19.5 MB, thereby demonstrating its superior classification performance and efficiency compared to existing state-of-the-art methods.