A Practical YOLO Approach to Classifying Thai Freshwater Snails of Economic Significance

Abstract

Freshwater snails are a valuable economic resource in Thailand, but species identification remains challenging due to morphological similarities that impact pricing, traceability, and aquaculture management. This study assesses an automated freshwater snail classification system using three YOLO variants trained for 100 epochs on 4,610 annotated images of six economically important species. The models were evaluated using precision, recall, mAP50, mAP50–95, inference time, and model size, revealing clear performance trade-offs. YOLOv9-tiny achieved the highest detection accuracy with an mAP50–95 of 0.9738 but incurred the largest model size and slowest inference. In contrast, YOLOv11-nano delivered the fastest inference and smallest footprint, though with lower accuracy (mAP50–95 of 0.8849), making it suitable for resource-limited or edge deployments. YOLOv8 provided a balanced alternative, offering competitive accuracy (mAP50–95 of 0.9708) with moderate computational cost. Misclassification most occurred between Bellamya sp. and Bellamya reticulata, particularly for juvenile specimens, highlighting the difficulty of distinguishing morphologically similar species and the need for more diverse training data. Overall, the results demonstrate the effectiveness of YOLO-based models for automated snail species identification, with strong potential for applications in aquaculture management, market standardization, and supply chain traceability. Future work will focus on real-world deployment, expanding datasets across diverse environments, and integrating explainable AI to improve model transparency and user trust.