GAPyolo11: A Lightweight and Accurate YOLO Variant for Crop Pest Detection Using Edge Devices

Crop pests are a core bottleneck restricting the improvement and efficiency enhancement of agriculture. The traditional manual inspection mode has problems such as high missed detection rate and slow response, making it unable to meet the demands of modernized and precise agricultural pest control. Due to the limited computing resources of agricultural edge devices such as mobile phones and drones, as well as detection challenges in the field such as the high proportion of small targets and complex backgrounds, this paper proposes an improved model GAPyolo11 (GhostConv And C3K2 -PPA YOLO11), which is developed on the basis of YOLO (You Only Look Once) v11 and jointly improved by integrating the lightweight module GhostConv and the feature enhancement module C3K2-PPA. It reduces redundant computations through the “main feature and cheap transformation” strategy of GhostConv and strengthens the capture of pest features by incorporating the multi-branch attention mechanism of C3K2-PPA. The experiment shows that, using a real agricultural dataset containing 101 types of pests, GAPyolo11 achieves an mAP@0.5 (mean Average Precision at an Intersection over Union threshold of 0.5) improvement of 16.7% compared to the original YOLOv11n, with an 18% reduction in parameters, a 35% decrease in computational cost, and a detection speed of 20 ms per frame. Under the experimental conditions used in this study, the proposed GAPyolo11 model achieves feasible inference performance for deployment on edge devices such as mobile phones and agricultural drones and realizes real-time and precise pest identification within the test scenarios. It provides potential technical support for rapid field diagnosis and variable-rate pesticide application for precision plant protection, which may help reduce pest control costs and lower the risk of pesticide abuse.