Side-scan Sonar Image Rough Recognition and Feature Matching Based on CNN and SIFT

Abstract

Compared with traditional single-beam side-scan sonar device, current side-scan sonar have greatly improved the stability, resolution and definition of sonar image, which leads to a large increase in the amount of data, so it is impossible to completely identify target artificially. During the side-scanning sonar operation, sonar equipment is affected by ocean currents, wind waves, etc., resulting in posture instability and image distortion. The complex underwater environment and the interference of its own will also cause a lot of noise in the collected sonar image. The problems above have brought great difficulties for target recognition. This paper utilizes LeNet5 convolutional neural network (CNN) to perform automatic rough recognition of side-scan sonar images, and then uses scale-invariant feature transform (SIFT) feature matching for further recognition. The processing of experimental data shows that the accuracy and efficiency of target recognition are good using these two algorithms.