Sonar Image Translation Using End to End Network for Underwater Recognition

Abstract

With the rapid development of sonar technology, the classification of underwater sonar images is of great significance. In order to recognize underwater sonar images more effectively, side scan sonar is used for data acquisition. After pre-processing, six types of sonar images are selected. Combining with the characteristics of underwater sonar images, relevant theories and methods of convolutional neural network are studied to establish an underwater sonar CNN recognition model, which achieves 98.5% recognition success rate. Experiments show that the recognition model can improve the accuracy, speed and robustness of underwater sonar image classification, and can meet the practical application requirements.