复杂背景下植物叶片病害的图像特征提取与识别技术研究  

Image Feature Extraction and Recognition of Plant Leaf Disease in Complex Background

作　　者：张芳[1] 李晓辉[1] 杨洪伟[1] 

机构地区：[1]沈阳农业大学信息与电气工程学院,辽宁沈阳110161

出　　处：《辽宁大学学报（自然科学版）》2016年第4期311-318,共8页Journal of Liaoning University：Natural Sciences Edition

基　　金：辽宁省博士启动基金项目(201501060)

摘　　要：为了减少植物病害给农业生产者带来的损失,提高植物病害的识别率和识别精度,对复杂背景下植物叶部病害的图像特征提取和识别方法进行了研究.采用基于超像素和形状上下文的方法对复杂背景下的黄瓜病害叶片图像进行分割.通过局部二值模式(LBP)、区域平均方差和区域平均熵值等方法,分别从颜色、形状和纹理三个方面提取了植物病害图像的11个典型特征.在对病斑检测器训练时主要使用了两种核函数,分别是线性核函数和高斯径向基核函数.在使用两种核函数进行训练时,需要进行参数优化,采用k-folder交叉验证和网格搜索法来选择最优的参数,并对采用基于径向基核函数和线性核函数的SVM方法的识别结果进行对比分析.结果表明:对于黄瓜白粉病的识别,采用基于径向基核函数的SVM病斑检测器的结果进行黄瓜叶片白粉病的检测的平均正确识别率为98.3%,而采用基于线性核函数SVM病斑检测器的结果进行黄瓜叶片白粉病的检测的平均正确识别率为96.7%,基于径向基核函数的SVM方法要优于基于线性核函数的SVM方法,更适合对黄瓜白粉病的识别研究.说明提出的植物叶部病害的图像特征提取和识别方法能对植物病害进行有效地识别.In order to reduce the losses brought by plant disease to agricultural producers and to improve the recognition rate and accuracy of plant disease, the image feature extraction of plant leaf disease and recognition methods under the complex background were studied. The method based on super pixels and the shape context was used to segment the cucumber disease image based on the complicated background. The local binary pattern （ LBP）, regional average variance and regional average entropy and other methods were used to extract the 11 typical characteristics of plant disease image respectively from the color, shape, and texture. Linear kernel function and Gaussian radial basis kernel function were used in the training of scab detector. The training required parameter optimization, and k-folder cross validation and grid search were used to select optimal parameters. The recognition results of SVM methods based on radial basis kernel function and linear kernel function were compared and analyzed. Experimental results showed that for the recognition of cucumber powdery mildew, the recognition rate of cucumber leaf powdery mildew used by the results of SVM scab detector based on Radial Basis Function was 98.33 %, while the recognition rate used by the results of SVM scab detector based on the linear kernel function was 96.7%. It is concluded that the SVM method based on radial kernel function is better than the SVM method based on linear kernel function, and the SVM method based on radial kernel function is more suitable for the recognition of cucumber powdery mildew. The results showed that the above method proposed about image feature extraction and recognition of plant leaf disease can effectively recognize the plant leaf disease.

关 键 词：植物病害 特征提取 SVM 识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]