基于改进YOLOv8的马铃薯叶片病害检测算法

Potato Leaf Disease Detection Algorithm Based on Improved YOLOv8

针对复杂背景环境下马铃薯叶片病害检测精度低的问题,以YOLOv8n为原型,提出了一种非受控环境下的马铃薯叶片病害检测算法:YOLOv8n-Potato。采用CAA-HS-FPN架构替换YOLOv8的颈部网络,用于提高特征融合效率;使用轻量化检测头Sc-Head替换YOLOv8的检测头,使模型轻量化;采用PIoU替换CIoU,降低了锚框回归的代价。与YOLOv8n相比,YOLOv8n-Potato的精确度提高了2.4%,召回率提高了8.4%,mAP50提高了3.6%,mAP50-95提高了1%,GFLOPs减少了23%,模型参数量减少了42%。

To address the issue of low detection accuracy of potato leaf diseases in complex background environments,a potato leaf disease detection algorithm,named YOLOv8n-Potato,was proposed based on YOLOv8n.The algorithm replaces the neck network of YOLOv8 with the CAA-HS-FPN architecture to enhance feature fusion efficiency.Additionally,a lightweight detection head,Sc-Head,is used to replace the detection head of YOLOv8,making the model lightweight.Finally,PIoU is adopted to replace CIoU to reduce the cost of anchor box regression.Compared to YOLOv8n,YOLOv8n-Potato is 2.4%higher in accuracy,8.4%higher in recall rate,3.6%in mAP50,and 1%in mAP50-95,while GFLOPs reduced by 23%and model parameters by 42%.