The salmon industry faces a challenge: the loser fish syndrome. These fish grow more slowly than their healthy counterparts, which affects farm profitability. Unfortunately, identifying “loser fish” can be subjective and time-consuming for human workers.
A new and interesting research conducted by scientists from the Norwegian University of Science and Technology (Norway), University of Campinas (Brazil), Wageningen University and Research (Netherlands), and the University of Bergen (Norway) presented and compared computer vision systems for the automatic detection and classification of “loser fish” in images of Atlantic salmon taken in marine cages.
What is loser fish syndrome?
Loser fish are smaller and thinner than their healthy counterparts. They often exhibit:
- Low energy: They swim slowly and appear lethargic.
- Poor health: They have reduced muscle mass and lack typical fat reserves.
- Abnormal behavior: They often swim alone near the surface or the walls of the cage.
These fish can become stressed and even transmit diseases to healthy fish. Unfortunately, the cause of loser fish syndrome is still unclear.
AI: A powerful tool for monitoring salmon health
Artificial Intelligence is being employed in the salmon industry for monitoring fish growth and health and detecting deformities in salmon, among other activities. The new study investigated how AI, specifically computer vision (CV), can be used to identify loser fish in images captured in salmon cages. Here’s how it works:
- Image analysis: Cameras installed in cages capture images of the salmon.
- Object detection and classification: AI algorithms analyze the images, identify individual fish, and classify them as “healthy” or “losers” based on their size, shape, and swimming behavior.
Computer Vision (CV) to detect loser fish
CV systems can automatically detect and classify these stunted fish by analyzing images of Atlantic salmon captured in marine cages.
The study compares three different computer vision systems:
- YoloV7: This state-of-the-art system combines fish detection and classification in a single step.
- YoloV5: Similar to YoloV7, but an earlier version.
- Two-stage approach: This system uses separate models for detection and classification, offering more flexibility. Uniquely, it employs an ensemble of classifiers, combining multiple models to improve accuracy.
Why is the ensemble approach unique? This study is the first to use an established ensemble of classifiers for identifying loser fish in salmon farms. Essentially, it combines the strengths of multiple classification methods to achieve a more robust result.
How well did the AI perform?
The AI systems analyzed images of salmon in marine cages, classifying them as healthy or “losers.” Here’s how they performed:
- YoloV7: Led the competition with an impressive 86.3% precision (correctly identifying loser fish) and a 78.35% F1 score (balancing precision and recall).
- YoloV5: Performed well with a precision score of 79.7%.
- Two-stage approach: Showed promise with a precision of 66.05%.
Interestingly, human evaluation also revealed some subjectivity, highlighting the benefit of AI for consistent monitoring.
Benefits of AI for salmon farming
Implementing AI-based loser fish detection could be cost-effective and allow continuous monitoring, helping to:
- Identify loser fish more quickly and consistently compared to manual efforts.
- Improve tracking of loser fish abundance throughout the production cycle.
- Potentially reduce economic losses by enabling earlier intervention and management strategies.
Sharing the knowledge
The researchers generously made their annotated salmon image dataset publicly available. This allows other scientists to build on this research and further refine AI models for loser fish detection.
Conclusion
The study explored the use of computer vision (AI) to detect and classify loser fish in salmon farm images automatically. According to the study, both the “end-to-end” AI and “two-stage” models showed promise in identifying loser fish.
However, current models have issues with fish posture variations and require additional testing under different lighting conditions and sizes.
The study paves the way for broader adoption of AI in aquaculture. By automating tasks like identifying loser fish, AI can contribute to more efficient and sustainable fish farming practices.
Reference (open access)
Banno, K., Gonçalves, F. M. F., Sauphar, C., Anichini, M., Hazelaar, A., Sperre, L. H., Stolz, C., Aas, G. H., Gansel, L. C., & Da Silva Torres, R. (2024). Identifying losers: Automatic identification of growth-stunted salmon in aquaculture using computer vision. Machine Learning With Applications, 16, 100562. https://doi.org/10.1016/j.mlwa.2024.100562
