Related Experiment Video
Updated: Jun 9, 2025
05:51
VisioTracker, an Innovative Automated Approach to Oculomotor Analysis
Published on: October 12, 2011
11.0K
Computer vision in aquaculture: a case study of juvenile fish counting
Krishna Moorthy Babu1, Daniel Bentall1, David T Ashton2
1The New Zealand Institute for Plant and Food Research Limited, Lincoln, New Zealand.
Journal of the Royal Society of New Zealand
|October 23, 2024
Summary
Machine learning models, Single Shot Detection (SSD) and Faster R-CNN, significantly speed up juvenile fish counting in aquaculture. Optimized models achieved high accuracy, reducing costs and enabling faster assessments for Australasian snapper.
Area of Science:
- Aquaculture
- Computer Vision
- Machine Learning
Background:
- Counting juvenile fish in aquaculture is labor-intensive and costly.
- Existing methods rely on manual counting, limiting throughput.
Purpose of the Study:
- To evaluate the efficacy of machine learning models for automated juvenile fish counting.
- To augment manual counting methods for Australasian snapper (Chrysophrys auratus).
Main Methods:
- Utilized two deep learning architectures: Single Shot Detection (SSD) and Faster R-CNN.
- Tuned model parameters (confidence thresholds, non-maximal suppression) and applied bias correction with Poisson regression.
- Validated model accuracy against manual counts.
Main Results:
- Optimized SSD and Faster R-CNN models achieved Mean Absolute Percent Errors (MAPE) below 10%.
- SSD model demonstrated superior performance with MAPE below 5%.
- Machine learning methods offer rapid assessment, complementing slightly more accurate manual counts (MAPE=1.56).
Conclusions:
- Machine learning provides a viable, high-throughput solution for juvenile fish counting in aquaculture.
- This study is a foundational step towards real-time fish counting and phenotypic data collection.
- Further development can enhance automation in aquaculture production programs.
