A Boundary-Guided and Lightweight Object Detector for Sonar Images

The complex underwater environment results in low resolution, poor contrast, and serious noise interference of sonar images. Although many denoising methods are used in sonar images, edge information blur has always been an important problem in sonar image denoising. In this paper, we propose a sonar image processing module based on edge guidance to transform sonar images into binary images with targets highlighting and clear boundaries, thereby proving that this module is helpful to improve the accuracy of sonar image detection tasks. In addition, we introduce the Transition Block module to decrease the parameters of the YOLOv7 detection model and improve the detection efficiency of the model. A comparative test on a public data set shows that our detector has better performance in accuracy and detection efficiency under four widely used evaluation indexes.