The convolution neural network based agent vehicle detection using forward-looking sonar image

Abstract

We propose the convolution neural network (CNN) based underwater object recognition and detection solution using forward-looking sonar image for localization of agent vehicle (small ROV). We designed the multiple underwater vehicle system with the autonomous underwater vehicle (AUV) equipped with the forward-looking sonar and the agent vehicle linked to the AUV through a tether cable. For localization of agent vehicle, we use the forward-looking sonar images. The CNN based object recognition algorithm trained the agent vehicle's sonar image and detected it in the overall sonar images. We found that the CNN algorithm successfully recognized the agent vehicle in the sonar image. The goal of our research is to propose a solution to apply the CNN based recognition algorithm to the underwater robotics. Finally, it shows the elevated recognition rate in the underwater and we can get the agent vehicle's localization data.