Research on Rice Leaf Disease Detection Based on Improved YOLOv8n Algorithm
Liu Zhengfeng
Yang Jiansheng
Zhang Mei
Chen Zhe
Zhang Qunying
Abstract：Rice leaf disease detection is one of the important ways to reduce disease risk and stabilize rice yield.In response to the problems of large parameter size,high computational complexity,and low accura-cy in existing rice leaf disease detection models,this study proposed an improved YOLOv8n model.First,the backbone network of the original YOLOv8n was replaced with a lightweight HGNetv2 architecture,and the Conv module in the HG-Block was replaced with a Ghost module,improving detection accuracy while reducing model size.Next,the residual blocks in the C3 module were replaced with Ghost Bottleneck to creating a new C3Ghost module,which was used to replace all C2f modules in the neck,further reducing model size while maintaining model performance.Finally,a dense prediction channel knowledge distillation technique was em-ployed to enhance the model in a lossless manner.Experimental results indicated that compared with the base-line model YOLOv8n,the proposed improved model reduced the parameter size,weights,and floating-point operations by 39.33%,37.00%,and 28.40%,respectively,while achieved precision and recall of 94.3%and 95.6%,and mAP of 96.7%,significantly outperforming the baseline model.Overall,the proposed improved model could meet the demands for accuracy and lightweight design in rice leaf disease detection tasks in agri-cultural scenarios,demonstrating good development potential and application prospects.
Keywords：Rice leaf disease detectionYOLOv8nModel lightweightingHGNetv2Knowledge distil-lation
Publication Date：2025-09-30
Online Publishing Date：2026-05-22(First online date of this platform, not the publication date of the document)
Pages：9( 164-172 )