KFP-YOLO: A Lightweight Detection Model for Korla Fragrant Pear Diseases and Pests Detection Toward Edge Deployment

Korla fragrant pear diseases and pests detection faces challenges such as significant object scale variation, multi-organ target confusion, and limited computational resources for real-time inference on edge devices. To address these issues, this study proposes a lightweight object detection model, KFP-YOLO, based on YOLO26n. A multi-organ dataset, the Korla fragrant pear diseases and pests dataset (KFP-PDD), was collected, covering pear leaves, fruits, and flowers and containing 11 classes including healthy and diseased samples. To reduce computational cost while maintaining effective feature representation, an ADown lightweight downsampling module is introduced. A C3-PD feature extraction module is designed by integrating Partial convolution and SE attention to reduce redundant computation and enhance feature representation capability. Furthermore, a CFA feature enhancement module is proposed, which incorporates coordinate attention into the multi-scale feature fusion process to improve spatial information modeling and fine-grained feature representation. Experimental results show that, compared with YOLO26n, KFP-YOLO reduces parameters and GFLOPs by 23.0 % and 23.7 %, respectively, while achieving an inference speed of 278.97 FPS. Meanwhile, mAP@0.5 reaches 94.65 %, with only a 0.44 percentage point drop. Ablation studies verify the effectiveness and synergistic optimization of each proposed module. Deployment experiments on the Jetson AGX Orin platform demonstrate strong real-time performance and edge computing adaptability, indicating that the proposed method offers an efficient solution for intelligent orchard diseases and pests monitoring.