Autonomous Underwater Robotic System for Aquaculture Applications

TL;DR

This paper presents an autonomous underwater robotic system utilizing deep learning for real-time detection of aquaculture net defects, aiming to enhance inspection efficiency and reduce operational costs.

Contribution

It introduces a novel integrated system combining deep learning and feedback control for autonomous underwater net inspection in aquaculture.

Findings

Effective detection of biofouling, vegetation, and net holes.

Real-time inspection capability demonstrated.

Reduced need for human divers and ROVs.

Abstract

Aquaculture is a thriving food-producing sector producing over half of the global fish consumption. However, these aquafarms pose significant challenges such as biofouling, vegetation, and holes within their net pens and have a profound effect on the efficiency and sustainability of fish production. Currently, divers and/or remotely operated vehicles are deployed for inspecting and maintaining aquafarms; this approach is expensive and requires highly skilled human operators. This work aims to develop a robotic-based automatic net defect detection system for aquaculture net pens oriented to on- ROV processing and real-time detection of different aqua-net defects such as biofouling, vegetation, net holes, and plastic. The proposed system integrates both deep learning-based methods for aqua-net defect detection and feedback control law for the vehicle movement around the aqua-net to obtain a clear sequence of net images and inspect the status of the net via performing the inspection tasks. This work contributes to the area of aquaculture inspection, marine robotics, and deep learning aiming to reduce cost, improve quality, and ease of operation.