基于改进YOLO v7的果蔬叶片病害检测算法
Algorithm for Detecting Fruit and Vegetable Leaf Diseases Based on Improved YOLO v7
张一帆 1张梅 1陈杰1
作者信息
- 1. 安徽理工大学 电气与信息工程学院,安徽 淮南 232001
- 折叠
摘要
为了提高对果蔬叶片病斑细小特征的精准定位能力并减小模型复杂度,提出一种基于改进YOLO v7 的果蔬叶片病害检测算法.首先对YOLO v7 模型的主干特征提取网络添加卷积注意力机制模块(CBAM),增强模型在提取病害初期相似特征方面的有效能力;其次,将原本的路径聚合网络(PANet)结构替换为渐近特征金字塔网络(AFPN)来支持非相邻级别的直接交互,在提高检测性能的同时使得模型轻量化;最后,将原始YOLO v7 的CIOU损失函数,更替为XIOU损失函数.试验结果表明,改进后的YOLO v7 算法能够有效对果蔬叶片病虫害进行检测,其平均检测精度为96.4%,比YOLO v3、YOLO v5s、YOLO v7 和SSD模型分别提高了 13.2、0.9、1.3 和18.7个百分点,与YOLO v7 网络模型大小相比减少了 22.4 MB.所提方法为果蔬叶片病害的精准检测提供了一种有效的技术支持.
Abstract
In order to improve the precise localization ability of small features of fruit and vegetable leaf lesions and reduce model complexity,a fruit and vegetable leaf disease detection algorithm based on improved YOLO v7 is proposed.Firstly,a Convolutional Block Attention Module(CBAM)attention mechanism module is added to the backbone feature extraction network of the YOLO v7 model to enhance the model's effective ability in extrac-ting similar features in the early stages of disease;secondly,the original Path Aggregation Network(PANet)structure is replaced with Asymptotic Feature Pyramid Network(AFPN)to support non adjacent level direct in-teraction,which improves detection performance and makes the model lightweight;finally,replace the CIOU loss function of the original YOLO v7 with the XIOU loss function.The experimental results show that the improved YOLO v7 algorithm can effectively detect fruit and vegetable leaf diseases and pests,with an average accuracy of 96.4%,which is 13.2,0.9,1.3,and 18.7 percentage points higher than YOLO v3,YOLO v5s,YOLO v7,and SSD models,respectively.Compared with the YOLO v7 network model size,it reduces by 22.4MB.The proposed method provides an effective technical support for precise detection of fruit and vegetable leaf diseases and pests.
关键词
果蔬病虫害/图像识别/YOLO/v7/特征金字塔/损失函数/轻量化Key words
diseases and pests/image recognition/YOLO v7/feature pyramid/loss function/lightweight引用本文复制引用
基金项目
国家自然科学基金资助项目(52374154)
出版年
2024
NETL
NSTL
万方数据