Classification of Underwater Target using Deep learning and Active Sonar Systems

الملخص

Underwater Acoustic (UA) target identification is a challenging and crucial sonar system task in marine remote sensing operations, particularly when complicated sound wave propagation characteristics are present. Most conventional machine learning (ML) algorithms generally encounter difficulties while trying to develop the costly recognition model for huge data analysis. In this paper utilized the task is to train a network to discriminate between sonar signals bounced off a metal cylinder and those bounced off a roughly cylindrical rock. We suggest a method utilizing a dense CNN model for identifying targets underwater. All previous feature maps are intelligently reused in the network architecture. to maximize classification rates under a range of compromised circumstances while achieving little computational expense. Our classification model beats classical ML approaches, achieving an overall accuracy of 96.77% based on the experimental findings assessed on the real-world data set of active sonar. The objective is to train a network to distinguish between sonar signals that bounce off roughly cylindrical rocks and those that bounce off a metal cylinder.