Active vision-based real-time aquaculture net pens inspection using ROV.

Area of Science:

• Marine technology
• Aquaculture engineering
• Robotics

Background:

• Aquaculture is crucial for sustainable protein supply.
• Timely detection of net pen damage is a significant challenge.
• Underwater inspections using optical cameras are hindered by poor visibility.

Purpose of the Study:

• To develop a robust method for clear underwater net inspection using ROVs.
• To improve the safety and efficiency of aquaculture net monitoring.
• To enable automated damage identification in dynamic marine environments.

Main Methods:

• Utilized a remotely operated vehicle (ROV) equipped with optical cameras.
• Implemented image processing techniques to track mean gradient features for optimal view selection.
• Integrated a convolutional neural network (CNN) with a Proportional Integral Derivative (PID) controller for precise pose control.
• Trained the CNN offline using supervised learning for distance set-point simplification.

Main Results:

• The proposed system successfully obtained clear net images in challenging underwater conditions.
• The ROV maintained a consistent relative pose to the fishnet, ensuring stable image acquisition.
• Experimental validation in both controlled and real-world fish farm environments demonstrated the method's efficacy.

Conclusions:

• The developed ROV-based system effectively addresses visibility issues in underwater net inspection.
• The combined CNN-PID approach provides accurate pose control for clear image capture.
• This technology enhances the safety and efficiency of aquaculture net damage assessment.