Recognition and Classification of Seabed Geological Images by Side-Scan Sonar Based on Convolutional Neural Networks

The classification and recognition of traditional submarine bottom sonar images are mainly carried out by manual recognition, which is inefficient and consumes a large amount of labor and time costs, and also leads to a decrease in recognition accuracy due to the subjective factors of the recognizer. In this paper, a convolutional neural network is introduced into the sonar side-scan image of seafloor geology, and the traditional manual identification is changed to machine learning features and automatically completing the classification work. The convolutional neural network model is trained and tested by using 1200 pre-processed images of 578 seafloor substrate images in the Seafloor dataset, and the results show that the trained CNN model can accurately recognize and classify the side-scan sonar seafloor substrate images with an accuracy of 99.22%, which is a highly efficient, accurate, and robust method, and is able to effectively increase the accuracy of the side-scan sonar seafloor substrate images. The method is efficient, accurate, robust, and can effectively improve the level of recognition and classification of side-scan sonar seabed substrate images.