图像处理的玉米病害识别研究  

Research on maize disease recognition based on image processing

作　　者：龚瑞昆 刘佳 GONG Ruikun;LIU Jia(School of Electrical Engineering,North China University of Science and Technology,Tangshan 063210,China)

机构地区：[1]华北理工大学电气工程学院,河北唐山063210

出　　处：《现代电子技术》2021年第24期149-152,共4页Modern Electronics Technique

基　　金：国家自然科学基金项目(61203343)。

摘　　要：为了快速准确识别病害种类,及时采取防治措施,文中通过图像处理技术对玉米叶部病害进行识别。首先利用中值滤波法去除图像噪声,在Lab颜色空间采用K⁃均值聚类方法分割图像;然后提取病斑区域的颜色、形状、纹理特征,选择HSV颜色空间各通道的一、二、三阶矩共9个特征量作为颜色特征,面积、周长、矩形度、伸长度、似球度作为形状特征,LBP提取图像的局部纹理特征量,提出一种典型相关分析方法对3组特征量进行融合;最后通过支持向量机对融合特征进行分类识别。典型相关分析既能实现特征融合,也可以有效去除特征间的冗余信息。实验结果表明,玉米大斑病、小斑病、灰斑病、锈病的识别率分别为92.5%,92.5%,92.5%,95.0%,平均识别率为93.1%,识别精度优于3组特征量串联融合方法,大大提高了玉米病害识别准确率。In order to quickly and accurately identify the types of diseases and take control measures in time,the maize leaf diseases were identified by means of the image processing technology.The median filter is used to remove image noise,and K⁃means clustering method is used to segment image in Lab color space.The color,shape and texture features of the lesion spot areas are extracted.The first,second and third moments of each channel in HSV color space are selected as the color features,the area,perimeter,rectangularity,elongation and sphericity are used as the shape feature.The LBP is used to extract local texture features measurement,the method of canonical correlation analysis is proposed to fuse the three groups of features quantity,and the fusion features are classified and recognized by support vector machine.The canonical correlation analysis can not only realize feature fusion,but also effectively remove the redundant information between features.The experimental results show that the recognition rates of large leaf spot,small leaf spot,gray leaf spot and rust disease of maize was 92.5%,92.5%,92.5%and 95.0%respectively,and the average recognition rate was 93.1%.The recognition accuracy was better than the series fusion method of three groups of feature quantities,which greatly improved the recognition accuracy of corn diseases.

关 键 词：玉米 病害识别 图像处理 特征融合 特征分类 典型相关分析 

分 类 号：TN911.73-34[电子电信—通信与信息系统] TP391.41[电子电信—信息与通信工程]