基于字典学习的马铃薯叶片病害图像识别算法

Identification Algorithm of Potato Diseases on Leaves Using Dictionary Learning Theory

为实现马铃薯叶片病害识别,达到及时防治的目的,设计了一种基于压缩感知理论的马铃薯病害图像分类方法。采用K-奇异值分解算法(K-SVD)分别构造了马铃薯早疫病、晚疫病、灰霉病叶片图像病害字典,通过正交匹配追踪算法求解测试样本在不同病害字典下的稀疏系数矩阵,并进行图像重构,求解重构均方根误差。利用不同类别字典本身的差异性,测试样本重构时,误差最小的字典即为测试样本所属病害种类。与支持向量机识别算法相比,该方法能够自学习图像特征,大大降低了图像分割和特征提取复杂度。经对比测试,采用字典学习理论进行分类,马铃薯3种叶片病害单一病斑图片综合识别率达到95.33%,高于支持向量机分类识别算法(识别率92%)。

In this study,an algorithm was designed using compressive sensing theory to classify potato diseases for the purpose of identification and prevention of potato leaf diseases timely. Leaf image dictionaries of potato early blight,late blight and grey mold were generated using K-singular value decomposition（KSVD） algorithm. The sparse coefficient matrix of one disease image was then decomposed by the above dictionaries respectively using orthogonal matching pursuit（OMP） and the image was reconstructed.RMSE of each reconstruction was compared and the smallest RMSE was obtained by the related disease dictionary. This method can learn image features automatically and reduce the complexity of image segmentation and feature extraction compared with the method based on support vector machine（SVM）. The recognition rate for the three disease plots reached 95. 33%,higher than the method based on SVM（92%）.