An Improved YOLOv8n-Based Approach for Citrus Leaf Disease Detection

To address the issue of low detection accuracy caused by small targets in citrus leaf disease detection, this paper proposes an improved method based on the YOLOv8n detection model. Firstly, we modify the neck layer of the YOLOv8n detection model by introducing a Channel Feature Adaptive Module to capture richer contextual information and multi-scale features, thereby enhancing the detection capability for small leaf targets. Secondly, we replace the original UpSample operation with DySample to enlarge the receptive field, improve resource utilization efficiency, and enhance model flexibility. Finally, the Inner-WIoU loss function is adopted instead of the original CIoU to optimize target localization accuracy and accelerate model convergence. The improved algorithm is validated on a self-built dataset. Experimental results demonstrate that the enhanced model achieves respective improvements of 3.1%, 2.3%, 3.7%, and 2.3% in precision, recall, mAP@50, and mAP@50-90 compared with the original YOLOv8n, showing significant enhancement in detection accuracy for citrus leaf diseases.