摘要
为实现玉米叶片表面未见明显病症的病害早期检测,提出一种基于高光谱成像技术的玉米弯孢叶斑病早期检测方法。以玉米叶片为研究对象,采用人工接种病菌使玉米感染弯孢叶斑病,在接种后1,2,3,4,5d,每天采集接种病菌叶片30片,正常未接种叶片10片,利用高光谱成像系统获取接种病菌叶片和正常未接种叶片在400~1000nm高光谱图像数据,经过分析接种病菌叶片和正常未接种叶片的原始光谱、原始光谱的一阶导数光谱、平均光谱绝对差值,确定玉米弯孢叶斑病早期检测的特征波段选取区。然后通过显著性检验和相关性分析,将置信区间设为95%,在特征波段选取区确定458.9,481.1,500.8,515.7,525.7, 531.9,534.4,550.7,578.3,604.9,625.2,646.8,677.5,735.3,754.7nm,为玉米弯孢叶斑病早期检测的特征波段。最后,基于选定的特征波段构建玉米弯孢叶斑病支持向量机检测模型。结果表明:利用选取的特征波段作为支持向量机的输入矢量,建立的玉米弯孢叶斑病早期检测模型,通过支持向量机选择的线性核函数、多项式核函数、径向基核函数3种不同的核函数,在接种后的第1d,3种核函数测试集准确率达79%以上,线性核函数在接种第3d,测试集准确率达到88.75%。该研究可以对玉米弯孢叶斑病在未见明显叶斑的早期进行快速、无损检测,为玉米病害的早期检测提供新的思路。
In order to realize early detection of diseases on maize leaf surface without obvious symptoms, we propose an early detection method of Curvularia Lunata based on hyperspectral imaging technology. In this experiment, maize leaves were inoculated artificially to infect Curvularia Lunata. On the 1st, 2nd, 3rd, 4th, 5 thafter inoculation, we collected 30 leaves inoculated and 10 leaves without not inoculated every day. And the hyperspectral image data of inoculated leaves and normal leaves during 400-1000 nm were collected with hyperspectral imaging system. By analyzing the absolute difference of the original spectrum, the first derivative spectrum and the average spectrum of the original spectrum of the inoculated leaves and the normal leaves, the characteristic band selection area for early detection of Curvularia Lunata in maize was determined. Then, by means of significance test and correlation analysis, 458.9, 481.1, 500.8, 515.7, 525.7, 531.9, 534.4, 550.7, 578.3, 604.9, 625.2, 646.8, 677.5, 735.3,754.7 nm were identified as characteristic bands for early detection of Curvularia Lunata in maize. Finally, a support vector machine(SVM) detection model for Curvularia Lunata was constructed based on the feature bands we selected. The early detection model of Curvularia leaf spot of maize was established by using the selected characteristic band as the input vector of support vector machine. Three different kernel functions, linear kernel function, polynomial kernel function and radial basis kernel function, were selected by support vector machine. On the 1 stday after inoculation, the accuracy of the test sets of three kernel function was more than 79%. On the 3 rdday after inoculation, the accuracy of the test set of linear kernel function was up to88.75%. This research can detect Curvularia Lunata quickly and nondestructively before leaf spots appear, and can provide a new idea for early detection of maize diseases.
作者
徐静
苗腾
周云成
邓寒冰
宋平
张聿博
XU Jing;MIAO Teng;ZHOU Yun-cheng;DENG Han-bing;SONG Ping;ZHANG Yu-bo(College of Information and Electric Engineering,Shenyang Agricultural University,Shenyang 110161,China)
出处
《沈阳农业大学学报》
CAS
CSCD
北大核心
2020年第2期225-230,共6页
Journal of Shenyang Agricultural University
基金
国家自然科学基金项目(31501217,31701318)
辽宁省教育厅科学研究青年项目(LSNQN201716)。
关键词
高光谱成像
病害
早期检测
特征波段
支持向量机
hyperspectral imaging
disease
early detection
feature band
support vector machine
