Aquaculture faces challenges such as sea lice infestations, diseases, and deformities in fish, which affect their health and commercialization.
In this context, Artificial Intelligence (AI) is ready to revolutionize the aquaculture industry. The master’s thesis conducted by Yeronis Assefa and Mohamed Ali, from the University of Agder, aimed to develop an AI-based solution to identify irregularities and deformities in farmed salmon through the use of ‘key point’ detection.
The research delves into how Artificial Intelligence (AI) can revolutionize aquaculture by detecting irregularities in fish using cutting-edge technologies.
The Power of AI in Aquaculture
AI has been used in the aquaculture industry for precision feeding, monitoring and controlling water quality, disease detection, and improving fish farm management; however, the new study focuses on using AI for early detection of fish deformities.
By utilizing key point detection and geometric analysis, AI can identify fish with irregularities or deformities, allowing for quick intervention and disease prevention. This translates to healthier fish, better yields, and reduced economic losses.
The AI Toolkit for Sustainable Aquaculture
AI-related technologies play a significant role because they can process large amounts of data, aiding in more informed decision-making. Tools that include AI can be grouped into:
- Deep learning and Convolutional Neural Networks (CNN): These AI techniques excel in image recognition, making them ideal for analyzing fish health and identifying anomalies.
- Object detection algorithms: YOLO and RCNN key point detection models identify specific features on fish bodies, enabling precise anomaly detection.
- Large datasets and machine learning: Training AI models on extensive labeled fish image datasets allows them to “learn” and improve their accuracy in identifying even subtle anomalies.
The Study
Yeronis Assefa and Mohamed Ali’s research proposes a solution: automatic detection of deformities in salmon fish through key point detection. This advanced technique uses AI models to identify specific points on the fish body, such as fins and eyes. By analyzing these points, the system can detect deviations from the norm, indicating possible deformities.
By analyzing these key points, the AI model can identify subtle deviations from the normal morphology of fish, which could indicate deformities.
The research focuses specifically on:
- Finding Solutions: Developing methods to identify fish with irregularities through key point detection and geometric analysis.
- Creating Efficiency: Utilizing AI models like YOLO and RCNN key point detection for rapid and accurate detection.
- Real-world Impact: Evaluating the practicality and effectiveness of AI solutions in real fish farms.
Deep learning methods based on Convolutional Neural Networks (CNN) have revolutionized visual recognition. Techniques like YOLO, Faster R-CNN, and Stack Hourglass excel in object identification and their key points (e.g., fins, eyes) in images.
The Winning Models: YOLO vs. R-CNN
The study tested two leading AI models: YOLO and R-CNN on a database containing 469 annotated images, where each image contains 20 key points. Both models impressed with their high precision rates:
- YOLO: With an average overall precision of 98.3%, YOLO excels in efficiently identifying key points.
- R-CNN: While slightly less accurate in detection (96% average overall precision), R-CNN shines in precisely aligning key points, providing detailed information on each point’s location.
Beyond Detection: Geometric Analysis
Yeronis Assefa and Mohamed Ali’s research doesn’t stop at mere detection. By analyzing the spatial relationships between these key points, the system can assess factors such as:
- Triangle Formations: Deviations from normal triangular shapes formed by fins and body parts may indicate deformities.
- Slopes and Angles: Unusual angles in the spine or body curvature could indicate issues.
- Pairs of Distances: Analyzing distances between key points can reveal anomalies in size or growth issues.
Additionally, the study categorizes deformities into three main areas: jaw, size (thin or thick), and spine. This specific focus allows for targeted interventions, such as adjusting feeding regimes or selective breeding to address concerns.
Conclusion
“Our findings have demonstrated a significant improvement in the accuracy and precision of deformity detection in fish and have established a new practice in the field by bridging the gap between cutting-edge technology and traditional aquaculture practices,” conclude the researchers.
This research paves the way for a future where AI enables aquaculture to be more efficient. By identifying and addressing deformities early on, we can ensure the health and quality of salmon and ultimately contribute to food security and economic well-being. It is important to note that different efforts are being made to reduce salmon deformities, mainly regarding their feeding.
AI is not just a fashion word; it’s a powerful tool that is shaping the future of aquaculture.
Reference (open access)
Assefa Yeronis and Mohamed Ali. 2023. Identification of Irregularities in Salmon Fish in Aquaculture by Using Key Point Detection. Master’s theses in Information and Communication Technology. Faculty of Engineering and Science. University of Agder. 55 p.
