A Segmenting Method for Greenhouse Cucumber Downy Mildew Images Based on Visual Spectral and Support Vector Machine
MA Jun-cheng1, DU Ke-ming1*, ZHENG Fei-xiang1, ZHANG Ling-xian2, SUN Zhong-fu1
1. Institute of Environment and Development in Agriculture, Chinese Academy of Agricultural Sciences, Beijing 100081, China
2. College of Information and Electrical Engineering，China Agricultural University，Beijing 100083, China
摘要： 针对温室现场环境下采集的黄瓜霜霉病叶片图像中存在光照不均匀和背景复杂的问题，提出了一种基于可见光光谱和支持向量机的温室黄瓜霜霉病图像分割方法。首先，提出了一种基于可见光谱的颜色特征CVCF(combination of three visible color features)及其检测方法，该颜色特征将超红特征(excess red，ExR)、H分量和b*分量三种颜色特征结合，通过设置ExR参数，降低光照条件对ExR的影响，克服了光照不均匀对病斑分割的影响。在CVCF的基础上，结合基于径向基核函数的支持向量机分类器，通过优化分类器参数构建病斑分割模型，获得了温室黄瓜霜霉病图像初始分割结果。在初始分割结果基础上，采用SURF(speeded up robust features)特征及形态学操作，对分割结果进一步优化，消除背景噪声对分割结果的影响，从而获得最终病斑分割结果。为进一步验证方法的有效性，选择了OTSU算法、K均值聚类算法和决策树算法，作对比研究。结果表明，OTSU+H*0.2，K-means+H+b*，DT+H+b*和该研究算法的错分率分别为：19.44%，40.19%，16.27%和7.37%，该算法对温室现场环境下采集的黄瓜霜霉病图像的分割效果明显优于其他对比方法。该方法能够充分克服光照不均匀和复杂背景的影响准确地提取病斑，为病害识别提供了良好的数据来源。
关键词：温室黄瓜；霜霉病；可见光光谱；支持向量机；CVCF；图像分割；SURF特征
Abstract：Aiming at the issues that there may exist uneven illuminance conditions and complicated background on the disease spots images captured in real greenhouse field, this paper presented a segmenting method for greenhouse cucumber downy mildew images based on visual spectral and support vector machine. Firstly, a novel combination of the visible color features and its detection method were presented, based on which the support vector machine and SURF features were integrated to segment the disease spot from images. The combination of the visible color features combined ExR, H component of HSV color space and b* component of L*a*b* color space. Because ExR was very likely to be influenced by illuminance conditions, an ExR parameter was adopted to reduce the influence from illuminance conditions to the disease spots segmentation. On the basis of combination of the visible color features, initial segmentation results can be achieved by using RBF based SVM classifier. Then the initial segmentation results were further optimized by using SURF features to eliminate the background noises. Finally, the segmentation results were compared with K mean clustering, OTSU thresholding and decision tree. The results showed that the accuracy rate of OTSU+H*0.2, K-means+H+b*, DT+H+b* and proposed method were 19.44%，40.19%，16.27% and 7.37%, respectively. The accuracy rate of proposed method was obviously higher than that of the other three methods, which indicated that the proposed method can meet the data requirement of the following diagnosis for greenhouse cucumber downy mildew.
马浚诚，杜克明，郑飞翔，张领先，孙忠富. 可见光光谱和支持向量机的温室黄瓜霜霉病图像分割[J]. 光谱学与光谱分析, 2018, 38(06): 1863-1868.
MA Jun-cheng, DU Ke-ming, ZHENG Fei-xiang, ZHANG Ling-xian, SUN Zhong-fu. A Segmenting Method for Greenhouse Cucumber Downy Mildew Images Based on Visual Spectral and Support Vector Machine. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2018, 38(06): 1863-1868.
[1] MA Jun-cheng，WEN Hao-jie, ZHANG Ling-xian, et al(马浚诚, 温皓杰, 张领先, 等). Transactions of the Chinese Society for Agricultural Machinery(农业机械学报), 2017, 48(2): 195.
[2] LIU Ya-dong, CUI Ri-xian(刘亚东，崔日鲜). Spectroscopy and Spectral Analysis(光谱学与光谱分析), 2015, 35(12): 3480.
[3] Esmael Hamuda, Brian Mc Ginley, Martin Glavin, et al. Computers and Electronics in Agriculture, 2017, 133: 97.
[4] Esmael Hamuda, Martin Glavin, Edward Jones. Computers and Electronics in Agriculture, 2016, 125：184.
[5] Hernández-Hernández J L, García-Mateos G, González-Esquiva J M, et al. Computers and Electronics in Agriculture, 2016, 122：124.
[6] Ma Juncheng, Li Xinxing, Zhang Lingxian, et al. Computers and Electronics in Agriculture, 2015, 111: 92.
[7] Rong Zhou, Shun’ichi Kaneko, Fumio Tanaka, et al. Computers and Electronics in Agriculture, 2014, 108: 58.
[8] Massimo Minervini, Mohammed M Abdelsamea, Sotirios A. Tsaftaris. Ecological Informatics, 2014, 23: 35.
[9] ZHANG Jian-hua, KONG Fan-tao, LI Zhe-min, et al(张建华，孔繁涛，李哲敏, 等). Transactions of the Chinese Society of Agricultural Engineering(农业工程学报), 2014, 30(19): 222.
[10] Bai Xuefei, Wang Wenjian. Neurocomputing, 2014, 136(8): 243.
[11] Elham Omrani, Benyamin Khoshnevisan, Shahaboddin Shamshirband, et al. Measurement, 2014, 55(9): 512.
[12] Herbert Bay, Andreas Ess, Tinne Tuytelaars, et al. Computer Vision and Image Understanding, 2008, 110(3): 346.
[13] Chang Chih-Chung，Lin Chih-Jen. Acm Transactions on Intelligent Systems & Technology, 2011, 2(3): 389.
