基于自适应中心对称局部二值模式的作物病害识别方法  

Adaptive center-symmetric local binary patterns for crop disease recognition

作　　者：王献锋[1] 张善文[1] 孔韦韦[1] WANG Xian-feng ZHANG Shan-wen KONG Wei-wei(College of Engineering and Technology, Xijing University, Xi＇ an 710123, China)

机构地区：[1]西京学院工程技术学院,陕西西安710123

出　　处：《广东农业科学》2016年第9期140-145,共6页Guangdong Agricultural Sciences

基　　金：国家自然科学基金(61473237);陕西省自然科学基础研究计划项目(2014JM2-6096)

摘　　要：基于局部二值模式(LBP)算子在模式识别中直方图维数高、判别能力差、具有冗余信息等缺点,针对作物病害叶片图像的特点,提出一种自适应中心对称局部二值模式(Adaptive Center-Symmetric Local Binary Patterns,ACSLBP)算法,并应用于作物病害识别。该算法能够得到光照和旋转不变性的纹理特征,利用模糊C均值聚类算法对病害叶片图像进行分割,再将分割后的病斑图像进行分块,然后采用自适应阈值提取每个子块的ACSLBP纹理直方图,结合作物病害叶片图像的颜色特征,利用最近邻分类器识别作物病害。在黄瓜4种常见病害叶片图像数据库上进行试验,平均识别率高达95%以上,表明该方法是有效可行的。The local binary pattern （ LBP ） based pattern recognition method has high histogram dimension, poor discriminate ability and redundant information. According to the features of crop disease image, to accurately describe the texture structure of crop disease leaf image, an adaptive center-symmetric local binary patterns （ ACSLBP ） algorithm was proposed for crop disease recognition. The invariance texture features of illumination and rotation were extracted by the algorithm. Firstly, every disease leaf image was segmented using the fussy means-clustering algorithm. Secondly, the segmented spot image was divided into several sub-images and their corresponding ACSLBP texture histogram features were extracted respectively based on the adaptive threshold. Finally, combining the 6 color features of spot image, the crop diseases were recognized by nearest neighbor classifier. The experiments on leaf image database of four common cucumber diseases were implemented. The average correct recognition rate was above 95%. The experimental results verified that the proposed method was effective and feasible.

关 键 词：自适应对称局部二值模式 病害叶片图像分割 作物病害识别 最近邻分类器 

分 类 号：TP391.413[自动化与计算机技术—计算机应用技术] S431[自动化与计算机技术—计算机科学与技术]