Research on Turtle Target Detection System Based on YOLO Algorithm

The implementation of neural networks and computer vision algorithms has been identified as a critical factor in reducing costs, enhancing efficiency, and alleviating labor shortages in agricultural settings. This study introduces the computer vision methods to address the practical challenges of detecting and counting in the breeding of yellow-margined box turtle, with a particular focus on iterative optimizations of the YOLOv5s version to enhance its suitability and efficiency. The experiments conducted a comprehensive evaluation of a series of strategies incorporating advanced loss functions, including Focal-EIoU, SIoU, and WIoU. Additionally, the study employed the BiFPN feature fusion method and incorporated a CoordinateAtt attention mechanism. Following a comprehensive examination of the impact of distinct modules on the outcome, the study identified a YOLOv5s-CA-Focal-EIoU algorithm that demostrated superior performance compared to other models. This algorithm exhibited a 2.0% improvement in mAP50:90 (from 0.614 to 0.634) compared to the baseline network. As part of this study, a user-friendly graphical interface was developed for the optimized YOLOv5s model, which enables real-time detection of turtle images and videos. The GUI serves to reduce the barrier to algorithm usage, allowing breeders and researchers to easily apply intelligent technologies in turtle breeding management.