Multi-Target Detection in Underwater Sensor Networks Based on Bayesian Deep Learning

Abstract

Underwater target detection and its development have an important role in advancing marine science and technology. However, the complex and dynamic underwater environment poses challenges for detecting non-cooperative targets. This paper focuses on the problem of detecting and recognizing multiple non-cooperative targets in USNs. Specifically, the generative model is firstly utilized to learn the probability distribution of underwater signals, and then Bayesian fusion of active and passive measurements is utilized to achieve target detection. Along with this, a Bayesian deep learning classification framework is employed to categorize multiple targets. Compared to the traditional statistical detection methods, our method excels in hading underwater complexity and dynamics. In addition, unlike traditional deep learning, our classification framework combines Bayesian inference with deep learning to quantify environmental uncertainty. This approach helps the model perform more robust detection and improves the management of noise and uncertainty. Experimental and simulation analysis demonstrate the effectiveness of Bayesian deep learning methods in solving the challenges of underwater target detection. These findings highlight the potential of our approach in enhancing sensing and surveillance capabilities in complex underwater environments.