Related Experiment Video
Updated: Feb 28, 2026
09:32
Development of New Methods for Quantifying Fish Density Using Underwater Stereo-video Tools
Published on: November 20, 2017
9.8K
Cost-Effective Fish Volume Estimation in Aquaculture Using Infrared Imaging and Multi-Modal Deep Learning
Like Zhang1, Yanling Han1, Ge Song1
1College of Information Technology, Shanghai Ocean University, Shanghai 201306, China.
Sensors (Basel, Switzerland)
|February 27, 2026
Summary
This study introduces a low-cost infrared camera system for accurate fish volume estimation in aquaculture. The innovative pipeline enables scalable biomass monitoring, supporting sustainable seafood production.
Area of Science:
- Aquaculture technology
- Computer vision
- Biomass estimation
Background:
- Accurate fish volume estimation is crucial for sustainable aquaculture but traditional methods are invasive and costly.
- Existing non-invasive techniques often require expensive multi-sensor systems, limiting scalability.
- There is a need for cost-effective, non-invasive solutions for real-time biomass monitoring in dense aquaculture tanks.
Purpose of the Study:
- To develop a cost-effective infrared (IR)-only pipeline for reconstructing depth and Red Green Blue (RGB) data from low-cost IR videos.
- To enable scalable and accurate fish biomass monitoring in dense aquaculture environments.
- To reduce hardware costs for fish volume estimation while maintaining high accuracy.
Main Methods:
- Developed an IR-only pipeline with five integrated modules: IR-to-depth estimation, IR-to-RGB generation, detection and tracking, instance segmentation, and volume estimation.
- Utilized contour-guided attention, texture-conditioned injection, cross-modal fusion, depth-guided branches, and trajectory-depth Transformer fusion.
- Incorporated specific losses like smoothing loss, water-adaptive loss, and deformation-adaptive loss for improved performance.
- Trained the system on a dataset of 166 goldfish across 124 videos, with 8-16 fish per tank.
Main Results:
- Achieved a Mean Absolute Error (MAE) of 0.85 cm³ and a coefficient of determination (R²) of 0.961 for fish volume estimation.
- Outperformed state-of-the-art methods by 19-41% in accuracy.
- Reduced hardware costs by 80% compared to existing multi-sensor setups.
- Demonstrated robustness in dense tank conditions with 8-16 fish per tank.
Conclusions:
- The proposed IR-only pipeline offers a cost-effective and accurate solution for fish volume estimation and biomass monitoring in aquaculture.
- This technology advances precision aquaculture, enabling better feed optimization and health monitoring.
- The system promotes environmental sustainability by supporting efficient resource management in response to rising global seafood demand.
