A Two-Mode Underwater Smart Sensor Object for Precision Aquaculture Based on AIoT Technology.

Area of Science:

• Aquaculture Technology
• Underwater Imaging
• Artificial Intelligence in Fisheries

Background:

• Precision aquaculture requires non-intrusive monitoring of fish in various water conditions.
• Traditional methods struggle with turbid or low-light environments and crowded fish schools.

Purpose of the Study:

• To develop and validate a two-mode underwater surveillance camera system for monitoring fish in aquaculture.
• To enhance fish quantity and size estimation using AI in challenging underwater environments.

Main Methods:

• Development of a two-mode camera system integrating sonar and stereo RGB imaging.
• Application of machine learning (Mask R-CNN, CNNs, semantic segmentation) for sonar image analysis.
• 3D spatial alignment of sonar and RGB data for AI-driven fish annotation.
• Cloud-based AIoT system for data collection and analysis in aquaculture tanks and net cages.

Main Results:

• AI functions accurately estimate fish quantity and size distribution from sonar images.
• Overlapping fish instances in crowded schools were effectively analyzed using machine learning.
• Integrated sonar and RGB data improved fish annotation accuracy.
• The system demonstrated feasibility and suitability for remote underwater fish metric estimation.

Conclusions:

• The developed smart camera system offers a robust solution for non-intrusive fish monitoring in aquaculture.
• AI integration significantly improves the accuracy of fish stock assessment in challenging underwater conditions.
• This technology supports precision aquaculture by providing reliable, remote sensing of fish populations.