可见光光谱和机器学习的温室黄瓜霜霉病严重度定量估算

doi:10.3964/j.issn.1000-0593(2020)01-0227-06

摘要
参考文献
相关文章 (15)

全文: PDF (2236 KB)
输出: BibTeX | EndNote (RIS)

摘要：温室黄瓜霜霉病严重度的准确估算是科学防治霜霉病的前提条件，对于减少农药使用量、提升温室黄瓜品质和农民经济效益具有重要意义。机器学习在植物病害诊断领域的应用越来越广泛，已经取得了丰富的研究成果，病害严重程度的估算萌发了新的思路。利用霜霉病可见光图像并结合机器学习方法，开展温室黄瓜霜霉病严重度快速准确定量估算研究。利用数码相机采集温室黄瓜霜霉病叶片图像并进行预处理，剔除病害图像的背景。以黄瓜霜霉病叶片图像为输入，构建基于卷积神经网络（CNN）的估算模型。利用可见光光谱颜色特征（CVCF）结合支持向量机进行温室黄瓜霜霉病病斑图像分割，采用SURF（speeded up robust features）特征及形态学操作对分割结果进行优化。在获取黄瓜霜霉病病斑分割图像后，提取病斑图像RGB，HSV，L^*a^*b^*，YCbCr和HSI共5个颜色空间15个颜色分量的平均值和标准差2个颜色特征，以及在此基础上结合灰度共生矩阵提取各颜色分量的对比度、相关性、熵和平稳度4个纹理特征，共计90个特征；利用Pearson相关性分析进行特征优选，采用与温室黄瓜霜霉病严重程度实际值相关性高的图像特征构建浅层机器学习估算模型，包括支持基于向量机回归（SVR）的估算模型和基于BP神经网络（BPNN）的估算模型。基于以上3种估算模型开展黄瓜霜霉病严重度定量估算，采用决定系数（R²）和归一化均方根误差（NRMSE）对估算模型准确率进行定量评价。结果表明，模型估算的温室黄瓜霜霉病严重度与实际值之间具有良好的线性关系，其中，基于CNN的估算模型准确率最高，模型的R²为0.919 0，NRMSE为23.33%，其次是基于BPNN的估算模型，其R²为0.890 8，NRMSE为24.64%，基于SVR的估算模型的准确率最低，其R²为0.8901，NRMSE为31.08%。研究结果表明，利用黄瓜霜霉病可见光图像数据，结合卷积神经网络估算模型，实现了温室黄瓜霜霉病严重度的准确估算，能够为温室黄瓜霜霉病的科学防治提供参考，提高病害防治效率，减少农药使用。

关键词：温室黄瓜；霜霉病；严重度；可见光光谱；机器学习；估算

Abstract：Accurate estimation of disease severity for downy mildew of greenhouse cucumber is a prerequisite for scientific disease control. It is of great significance to reduce the use of pesticides and to improve the quality of greenhouse cucumber, as well as farmers’ income. With the application of machine learning in the field of plant disease diagnosis, estimating the severity of plant diseases is gaining concerns. In order to increase the accuracy, this paper used the digital images of greenhouse cucumber downy mildew and machine learning methods to estimate the disease severity for downy mildew of greenhouse cucumber. A digital camera was used to collect images of greenhouse cucumber leaves with downy mildew, whose background were manually eliminated. An estimation model based on Convolutional Neural Network (CNN) was constructed with cucumber downy mildew leaf image as input. The initial symptom segmentation was achieved by using the combination of three visible color features (CVCF) and support vector machine. The segmentation results were optimized by using the speeded up robust feature (SURF) feature and morphological operation. After obtaining the segmentation image of cucumber downy mildew symptoms, the average and standard deviation of 15 color components in five color spaces of RGB, HSV, L^*a^*b^*, YCbCr and HSI were extracted. On this basis, the gray level co-occurrence matrix was used to extract four texture features of each color component, including contrast, correlation, entropy and stability, resulting in 90 features. Pearson correlation analysis was used for feature selection. Shallow machine learning estimation models, including Support Vector Machine Regression and BP Neural Network, were constructed based on the image features with high correlation with the actual severity value of downy mildew of greenhouse cucumber. Based on the three estimation models, the disease severity for downy mildew of cucumber was estimated. The accuracy of the three estimation models was quantitatively evaluated by using Coefficient of Determination (R²) and Normalized Root-Mean-Squared Error (NRMSE). The results showed that there was a good linear relationship between the severity of downy mildew of greenhouse cucumber estimated by the model and the actual values. The model based on CNN achieved the best accuracy, whose R² was 0.919 0 and NRMSE was 23.33%, followed by the model based on BPNN, with R² being 0.890 8, NRMSE being 24.64%, while the model based on SVR was the last, with R² being 0.890 1 and NRMSE being 31.08%. The evaluation results showed that by using the digital images of cucumber downy mildew and the convolution neural network estimation model, the disease severity for downy mildew of greenhouse cucumber could be accurately estimated, which could provide support to the scientific control of downy mildew of greenhouse cucumber and reduce the use of pesticides.

Key words：Greenhouse cucumber; Downy mildew; Severity; Visible spectrum; Machine learning; Estimation

收稿日期: 2018-12-04 修订日期: 2019-04-08

中图分类号:

TN941.1

基金资助: 山东省重点研究发展计划任务（2017CXGC0201），叶类蔬菜产业技术体系北京市创新团队建设项目（BAIC07-20），国家自然科学基金项目（31801264）资助

通讯作者: 马浚诚 E-mail: majuncheng@caas.cn

作者简介: 张领先，1970年生，中国农业大学信息与电气工程学院教授 e-mail: zhanglx@cau.edu.cn

引用本文:

张领先，田潇，李云霞，陈运强，陈英义，马浚诚. 可见光光谱和机器学习的温室黄瓜霜霉病严重度定量估算[J]. 光谱学与光谱分析, 2020, 40(01): 227-232.
ZHANG Ling-xian, TIAN Xiao, LI Yun-xia, CHEN Yun-qiang, CHEN Ying-yi, MA Jun-cheng. Estimation of Disease Severity for Downy Mildew of Greenhouse Cucumber Based on Visible Spectral and Machine Learning. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2020, 40(01): 227-232.

链接本文:

https://www.gpxygpfx.com/CN/10.3964/j.issn.1000-0593(2020)01-0227-06 或 https://www.gpxygpfx.com/CN/Y2020/V40/I01/227

[1] Ma J C, Du K M, Zhang L X, et al. Computers and Electronics in Agriculture, 2017, 142: 110.
[2] WANG Xiang-yu，LI Xin-xing，ZHANG Ling-xian，et al（王翔宇，李鑫星，张领先，等）. Transactions of the Chinese Society for Agricultural Machinery（农业机械学报），2016，47(9)：266.
[3] MA Jun-cheng，DU Ke-ming，ZHENG Fei-xiang，et al（马浚诚，杜克明，郑飞翔，等）. Spectroscopy and Spectral Analysis（光谱学与光谱分析）, 2018，38(6): 1863.
[4] PENG Zhan-wu，SI Xiu-li，WANG Xue，et al（彭占武，司秀丽，王雪，等）. Journal of Chinese Agricultural Mechanization（中国农机化学报）, 2013，(2): 212.
[5] MA Jun-cheng，WEN Hao-jie，LI Xin-xing，et al（马浚诚，温皓杰，李鑫星，等）. Transactions of the Chinese Society for Agricultural Machinery（农业机械学报）, 2017, 48(2): 195.
[6] Wei Y, Chang R, Wang Y, et al. A Study of Image Processing on Identifying Cucumber Disease. In Li D, Chen Y. Computer and Computing Technologies in Agriculture V. CCTA 2011, IFIP Advances in Information and Communication Technology, Vol 370. Springer Berlin Heidelberg.
[7] Zhang S, Wang Z. Neurocomputing, 2016, 205(C): 341.
[8] YE Hai-jian，LANG Rui（叶海建，郎睿）. Transactions of the Chinese Society for Agricultural Machinery（农业机械学报）, 2017, 48(3): 24.
[9] Wang G, Sun Y, Wang J. Computational Intelligence and Neuroscience, 2017, 2017: 2917536.
[10] YANG Chong-shi，WANG Wan-li，LIU Geng-chun，et al（杨崇实, 王万立, 刘耕春，等）. Progress and Practice of Bioassay Method for Cucumber Downy Mildew(黄瓜霜霉病生物测定技术研究进展与实践). Annual Meeting of Pesticide Professional Committee of China Chemical Industry Association（中国化工学会农药专业委员会年会），2000.
[11] LI Jing-zhu，WANG Peng，GENG Chang-xing（李井祝，王鹏，耿长兴）. Journal of Chinese Agricultural Mechanization（中国农机化学报），2017, 38(6): 67.
[12] SUN Jun，TAN Wen-jun，MAO Han-ping，et al（孙俊，谭文军，毛罕平，等）. Transactions of the Chinese Society of Agricultural Engineering（农业工程学报）, 2017, 33(19): 209.
[13] Ferentinos K P. Computers & Electronics in Agriculture, 2018, 145: 311.
[14] Mohanty S P, Hughes D P, Salathé M. Frontiers in Plant Science, 2016, 7：1419.
[15] MA Jun-cheng，DU Ke-ming，ZHENG Fei-xiang，et al（马浚诚，杜克明，郑飞翔，等）. Transactions of the Chinese Society of Agricultural Engineering（农业工程学报）, 2018, 34(12): 186.
[16] Zhang C, Pham M, Fu S, et al. Mathematical Programming, 2018, 169(1): 277.
[17] Jin B S, Zuo Y H, Ma X D, et al. Advanced Materials Research, 2012, 433-440: 5469.
[18] LI Chang-chun，NIU Qing-lin，YANG Gui-jun（李长春，牛庆林，杨贵军）. Transactions of the Chinese Society for Agricultural Machinery（农业机械学报），2017, 48(8): 147.
[19] Ma J, Du K, Zhang L, et al. Computers and Electronics in Agriculture, 2018, 154: 18.