基于改进YOLOv8的水稻病害检测算法  

Rice disease detection algorithm based on improved YOLOv8

作　　者：靳新宇 于复兴 索依娜 宋小明 JIN Xinyu;YU Fuxing;SUO Yina;SONG Xiaoming(College of Artificial Intelligence,North China University of Science and Technology,Tangshan 063210,China;Hebei Key Laboratory of Industrial Intelligent Perception,Tangshan 063210,China;College of Life Sciences,North China University of Science and Technology,Tangshan 063210,China)

机构地区：[1]华北理工大学人工智能学院,河北唐山063210 [2]河北省工业智能感知重点实验室,河北唐山063210 [3]华北理工大学生命科学学院,河北唐山063210

出　　处：《江苏农业学报》2025年第3期537-548,共12页Jiangsu Journal of Agricultural Sciences

基　　金：国家自然科学基金项目(32172583)。

摘　　要：为提升对水稻病害的检测性能,本研究提出了一种改进的YOLOv8n检测算法。首先,在颈部网络中引入Slim-Neck结构,采用GSConv(Ghost shuffle convolution)降低计算成本,同时结合基于一次性聚合方法设计的跨阶段部分网络模块(VoVGSCSP)简化计算过程和网络结构,利用相似性感知注意力机制(SimAM)增强模型对病斑细微颜色变化的敏感性,最后将自适应特征金字塔网络(AFPN)模块和头部结构相结合,通过非相邻层的特征融合,精准捕捉病害区域的颜色、形状与纹理。试验结果显示,改进后的模型YOLOv8n-SMAF精确度、召回率和交并比阀值为0.50的平均精度(mAP_(50))分别达到85.1%、79.7%和83.7%。与原始模型YOLOv8n相比,改进后的模型YOLOv8n-SMAF精确度、召回率和mAP_(50)分别提高了3.8个百分点、4.5个百分点和2.7个百分点。与SSD、YOLOv7-tiny、YOLOv10n等其他主流模型相比,YOLOv8n-SMAF模型具有更高的检测精度,尤其在复杂场景下的检测任务中表现出优势。本研究改进的模型为水稻病害的早期预警和精准防治提供了技术支持。To improve the detection performance of rice diseases,this study proposed an improved YOLOv8n detection algorithm.Firstly,the Slim-Neck structure was introduced into the neck network.Ghost shuffle convolution(GSConv)was adopted to reduce the computational cost.At the same time,the cross-stage partial network module based on the one-shot aggregation method(VoVGSCSP)was combined to simplify the calculation process and network structure.The similarity-aware activation module(SimAM)attention mechanism was utilized to enhance the model’s sensitivity to subtle color changes of disease spots.Finally,the adaptive feature pyramid network(AFPN)module was combined with the head structure.Through the feature fusion of non-adjacent layers,the color,shape,and texture of the diseased areas were accurately captured.The experimental results showed that the precision,recall,and mean average precision at an intersection over union threshold of 0.50(mAP 50)of the improved model YOLOv8n-SMAF reached 85.1%,79.7%,and 83.7%respectively.Compared with the original model YOLOv8n,the precision,recall,and mAP 50 of the improved model YOLOv8n-SMAF increased by 3.8 percentage points,4.5 percentage points,and 2.7 percentage points respectively.Compared with other mainstream models such as SSD,YOLOv7-tiny and YOLOv10n,the YOLOv8n-SMAF model had higher detection accuracy,especially showing advantages in detection tasks in complex scenarios.The improved model in this study provides technical support for the early warning and precise prevention and control of rice diseases.

关 键 词：水稻病害 目标检测 YOLOv8 深度学习 图像处理 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]