A YOLO-Based Model for Detecting Stored-Grain Insects on Surface of Grain Bulks

Abstract

This study addresses the challenge of accurately and efficiently detect-ing tiny stored-grain insect pests on grain bulk surfaces, a critical task for integrated pest management (IPM). Existing detection models often struggle with small insects and require high computational resources. To overcome these limitations, the researchers developed YOLO-SGInsects, an enhanced YOLOv8s-based model incorporating a tiny-object detection layer, an asymptotic feature pyramid network, and a hybrid attention transformer module. Trained and tested on the GrainInsects dataset, which includes six insect species, the model achieved a mean average precision (mAP) of 94.2% and a counting root-mean-squared error (RMSE) of 0.7913, outperforming other mainstream detection models. The results demonstrate that YOLO-SGInsects can effectively detect and count tiny insects on grain surfaces, providing a valuable technical basis for improving IPM in granaries. This advancement has significant societal value as it enhances food security by enabling more effective pest control in grain storage facilities. Future research will focus on deploying the model on edge devices for mobile applications.