Enhancing Stored Grain Pest Detection with Improved YOLOv8-BiFPN Model

Grain pests in storage present a major risk to the world’s food safety. The research delves into a multi-scale pest identification technique utilizing machine vision and introduces an enhanced YOLOv8-BiFPN model. Utilizing the GrainPest dataset comprising 500 images of diverse grains and pests, the YOLOv8-BiFPN demonstrated superior performance compared to the baseline model. With random dataset partitioning, YOLOv8-BiFPN achieved a 94.8% mAP, improving by 2.57% over the original. The study suggests that random partitioning is sufficiently effective without manual intervention. The improved YOLOv8-BiFPN shows promise in detecting pests against complex backgrounds and small targets, offering a more effective approach for pest management in storage. Future work will focus on enhancing classification accuracy and finding optimal dataset partitioning strategies.