Target Recognition and Classification of Underwater Airplane and Sunken Ship Images Based on Convolutional Neural Network with Side-scan Sonar

The traditional analysis of side-scan sonar images faces several challenges, such as the strong subjectivity of manual interpretation, low efficiency, high labor costs, time consumption, and over-reliance on experience. Additionally, machine learning methods require manual feature selection, leading to a lack of adaptability and robustness. To address these issues, this study attempts to employ Convolutional Neural Networks (CNN) for the analysis of side-scan sonar images. This method can automatically extract features from side-scan sonar images of underwater aircraft and sunken ships and accomplish classification and recognition. In the study, we used a total of 1140 pre-processed side-scan sonar images of the seabed, underwater aircraft, and sunken ships from the Seabed Objects (ships and aircraft) dataset to train and test the constructed CNN model. The experimental results demonstrate that the trained CNN model can accurately classify and recognize side-scan sonar images of the seabed, underwater aircraft, and sunken ships, with an average accuracy rate reaching as high as 93.70%. This confirms that the method has high efficiency, high accuracy, and strong robustness, significantly enhancing the classification and recognition ability of side-scan sonar images of underwater aircraft and sunken ships.