AI technology has been proven as a valuable tool to collect behavioral data and individual size measurements of the reared stocks in finfish aquaculture, thus avoiding invasive methods involving fish sampling and handling (e.g., behavioral analysis1). Similar integrated systems have been developed for the embryos, larvae, and adults of zebrafish (Danio rerio), a valuable model species in biological research (e.g., developmental biology, ecotoxicology, genetics, fish biology2,3) In this study I developed an object detection and analysis model (Zid-AI), based on narrow AI technology which minimizes the needs for pricy equipment and can be done in any mobile phone camera, to automatically monitor the size and behavior of the individual zebrafish in the holding aquaria.
A total of 3607 zebrafish images from various free sources were annotated and augmented to form a diverse dataset of 13999 images. The YOLO V9 architecture was used to build the object detection model. For software testing, an experimental setup of 10 small tanks was used, each containing 3 fish (of small, medium and adult size). The size of the fish was estimated (mean total length, TL, of multiple measurements) by the Zid-AI software and the use of two ArUco markers placed in the tank. The accuracy of the Zid-AI output was estimated by comparing the results with the real fish TL (following anesthetization and digital photography).
The model’s precision-recall value (% of true positive fish identification) ranged from 83.0% to 87.6% (Figure 1). The trials on the accuracy of TL estimation are in progress. The Zid-AI system shows significant progress in non-invasive zebrafish monitoring, achieving reliable fish detection. However, visual occlusion and lens distortion pose challenges. Future work should focus on refining length measurement accuracy and expanding the dataset for more robust results. Testing in diverse environments and integrating real-time monitoring will further enhance its utility in zebrafish research.
References