Automatic Visual Inspection of a Net for Fish Farms by Means of Robotic Intelligence

Autonomous underwater vehicles (AUV) have been utilized to address the issue of net inspection in the fish farm industry. This approach can improve the accuracy, precision, and cost-effectiveness of net inspection, while also providing a safer alternative to human divers. However, creating precise inspection trajectories without GPS represent a significant challenge. In this paper, we propose a system that overcomes this challenge by controlling the distance between the net and the AUV using an optical camera and a supervised learning method. Specifically, we train a convolutional neural network to predict the distance between the net based on offline data. We then implement a feedback controller using the CNN predictions to guide the AUV’s movements. We evaluate our approach through experiments conducted in a 2x2x2m cube at CIRTESU, Castellón de la Plana, Spain. The results demonstrate the effectiveness of our proposed solution in addressing the problem of accurate net inspection without GPS, in controlled lighting conditions, and using a net that helps to run the computer vision algorithm. Further work will focus on adapting the algorithm to a more realistic scenario.