Crosstalk Noise Detection and Removal in Multi-beam Sonar Images Using Convolutional Neural Network

With its long operating range and ability to be used in a turbid environment, sonar sensor is mainly used to explore underwater environment. As a result, many algorithms based on the sonar have been developed. However, in the case of a multi-beam sonar, its mechanism causes crosstalk noise around the underwater object, which degrades the accuracy of these algorithms. In this paper, we propose a method to remove crosstalk noise from a sonar image by detecting the region where the crosstalk noise occurred using a convolutional neural network and filling the detected region with adjacent pixel values. The proposed method could accurately and effectively detect and remove crosstalk noise in a given sonar image. Therefore, the accuracy of the sonar-based algorithms such as a 3-D reconstruction of underwater terrain can be improved.