Smart IoT-based visual detection system for aquaculture monitoring using YOLO and edge computing

Abstract

Aquaculture is crucial for global food security, offering sustainable protein amid environmental challenges and declining wild stocks. This study develops a smart IoT-enabled real-time monitoring system for aquaculture farms, focused on detecting and counting post-larval redclaw crayfish (Cherax quadricarinatus). The system integrates a Raspberry Pi 4 Model B with a high-resolution camera and the YOLOv5s deep learning model, performing local image processing through edge computing to reduce latency and network load. A factorial experimental design evaluated system performance across four groups varying by molting status (molted vs. non-molted) and environment (covered vs. open-air ponds). Results showed the highest detection accuracy in non-molted crayfish under open-air conditions (F1-score = 0.93), with precision, recall, and mAP@0.5 exceeding 90%, while molted crayfish in covered ponds had the lowest scores (F1-score = 0.85). Statistical analyses confirmed significant effects of both molting and lighting on detection performance and their interaction (p < 0.05). Robustness tests demonstrated model stability under noise and variable lighting, with F1-scores remaining above 0.80. The system provides a scalable, cost-effective solution that improves operational efficiency, reduces manual labor, and supports sustainable aquaculture by delivering timely alerts for abnormal crayfish behavior, enabling proactive farm management.