| 摘要: |
| 为实现苜蓿叶部病害的快速准确诊断和鉴别，基于图像处理技术，对常见的4种苜蓿叶部病害（苜蓿褐斑病、锈病、小光壳叶斑病和尾孢菌叶斑病）的识别方法进行探索。对采集获得的899张苜蓿叶部病害图像，利用人工裁剪方法从每张原始图像中获得1张子图像，然后利用结合K中值聚类算法和线性判别分析的分割方法进行病斑图像分割，得到4种病害的典型病斑图像（每张典型病斑图像中仅含有1个病斑）共1 651张。基于卷积神经网络提取病斑图像特征，建立病害识别支持向量机（Support vector machine，SVM）模型。结果表明：当病斑图像尺寸归一化为32×32像素，利用归一化的特征HSV （即特征H、特征S和特征V归一化后的组合特征）构建的病害识别SVM模型最优，其训练集识别正确率为94.91%，测试集识别正确率为87.48%。本研究基于深度学习和SVM所建立的病害识别模型可用于识别上述4种苜蓿叶部病害。 |
| 关键词: 苜蓿 病害 图像识别 特征提取 深度学习 卷积神经网络 支持向量机 |
| DOI：10.11841/j.issn.1007-4333.2017.07.015 |
| 投稿时间：2016-07-11 |
| 基金项目:公益性行业（农业）科研专项经费项目（201303057） |
|
| Image recognition of four different alfalfa leaf diseases based on deep learning and support vector machine |
|
QIN Feng1, LIU Dongxia2, SUN Bingda3, RUAN Liu1, MA Zhanhong1, WANG Haiguang1
|
| (1.College of Plant Protection, China Agricultural University, Beijing 100193, China;2.College of Agriculture and Forestry Science and Technology, Hebei North University, Zhangjiakou 075000, China;3.Institute of Microbiology, Chinese Academy of Sciences, Beijing 100101, China) |
| Abstract: |
| To realize timely and accurately diagnose and identification of alfalfa leaf diseases,automatic recognition of four kinds of alfalfa leaf diseases including common leaf spot caused by Pseudopeziza medicaginis,rust caused by Uromyces striatus,Leptosphaerulina leaf spot caused by Leptosphaerulina briosiana and Cercospora leaf spot caused by Cercospora medicaginis,was investigated based on image processing technology.A sub-image with one typical lesion or multiple typical lesions was obtained by artificial cutting from each of 899 digital images of the four kinds of alfalfa leaf diseases and then was segmented by using a segmentation method integrating with K median clustering algorithm and linear discriminant analysis.After segmentation,a total of 1 651 typical lesion images,each of which only contained one lesion,were obtained for further feature extraction and image recognition of the diseases.Features of the typical lesion images were extracted based on convolutional neural networks and were then used to build support vector machine (SVM) models for image recognition of the diseases.The results showed that the optimal one among the SVM models was built based on the normalized feature HSV,were obtained by merging the normalized features H,S and V while the corresponding original features which was extracted from the normalized lesion images of 32×32 pixels.For this optimal disease recognition SVM model,the recognition accuracy of the training set reached 94.91% and that of the testing set was 87.48%.The results indicated that the image recognition model built based on deep learning and SVM could be applied to conduct the recognition and identification of the four kinds of alfalfa leaf diseases.In this study,some basis and methodological references were provided for the diagnosis and identification of alfalfa diseases and other plant diseases. |
| Key words: alfalfa disease image recognition feature extraction deep learning convolutional neural network support vector machine |
