摘要
为了快速、准确地对水稻叶瘟病病害程度进行分级评估,结合定性分析与定量估算,提出了一种基于高光谱成像技术的水稻叶瘟病病害程度分级方法。利用HyperSIS高光谱成像系统采集了受稻瘟病侵染后不同病害等级的水稻叶片高光谱图像,通过分析叶瘟病斑区域与正常叶片部位的光谱特征,对差异较大的550和680 nm波段进行二维散点图分析,提取只含病斑的高光谱图像;然后通过主成分分析(principal component analysis,PCA)方法得到利于褐色病斑和灰色病斑分割的第2主成分图像,采用最大类间方差法(Otsu)分割出灰色病斑;最后结合延伸率和受害率2个参数对水稻叶瘟病病害程度进行分级。试验结果表明:测试的166个不同稻叶瘟病害等级的叶片样本中,其中160个样本可被准确分级,分级准确率为96.39%。该研究为稻叶瘟病田间病害程度评估提供了基础,也为稻瘟病抗性鉴定方法提供了新思路。
Rice blast is one of the important diseases in rice production. Identification and classification of rice blast is mainly completed by visual observation according to image contrast or literal description currently, however, these methods are subjective and inefficient, besides requiring workers with high professional knowledge. In order to quickly and accurately evaluate the disease level of rice leaf blast, a grading method of rice leaf blast based on hyperspectral imaging technology was proposed. Hyperspectrai images of leaf blast at different levels were captured with a HyperSIS hyperspectral system. The logical AND operation was conducted by using the original image and mask image （single leaf image） to remove the background information and a hyperspectral image of a single leaf was obtained. Based on a whole leaf scale, the principal component analysis （PCA） method was used, but the operation results showed that principal component images （PCI） cannot significantly reflect the difference between brown disease spots and gray disease spots. To solve the above problem, a 2-D scatter chart analysis with two sensitive bands （550nm and 680nm） was used to extract hyperspectral images which contained only disease spots （brown disease spots and gray disease spots） through analyzing spectral features of leaf blast areas and normal areas. Based on a disease spots scale, the second principal component image was obtained to identify brown disease spots and gray disease spots by using a PCA method. On this basis, gray disease spots were efficiently identified using an Otsu method. And disease levels of grade 1 or 2 and above grade 3 were classified based on whether there existed gray disease spots. Combined with two parameters （elongation and infestation rate）, the disease level above grade 3 of rice leaf blast was classified. Through calculating the elongation of a set of 30 spindle disease spots, the elongation of 0.3 was selected as the threshold to distinguish grade 3 and 4. If elongation〉~0.3, the disease level was grade 3. If 0〈elongation〈0.3, it was above grade 3. A disease level more than grade 4 was classified according to the area occupied by the infestation area on a whole leaf. The testing result showed that 160 samples could be accurately classified in the 166 test samples at different levels of rice leaf blast and the overall classification accuracy was 96.39%. Classification as Grades 0, 1 or 2 and grade 3 are based on discrete characteristic data, while above grade 4 is based on continuous characteristic data. Utilizing infection rate as an evaluation criteria to artificially classify into grade 4-9, the result of the classification was affected by the characteristics of the test samples and the distribution of data. The linear regression methods were developed to analyze the correction between the measured and predicted values of the infestation rates within the disease levels of 4-9. The validation results showed good performance with the determination coefficient （R2） between actual measured and predictive value in the model being 0.934 and the root mean square error （RMSE） determined to be 0.26. The result is able to satisfy the precision demand of the quantitative calculation. This research result provided a foundation to evaluate the field disease level of rice blast and a new idea for resistance identification method of rice blast.
出处
《农业工程学报》
EI
CAS
CSCD
北大核心
2013年第19期138-144,共7页
Transactions of the Chinese Society of Agricultural Engineering
基金
国家自然科学基金项目(31101087)
高等学校博士学科点专项科研基金(20104404120002)
现代农业产业技术体系建设专项资金资助(CARS-01-33)
广东省现代农业产业技术体系(粤财教[2009]356号)
关键词
主成分分析
病害
分级
水稻叶瘟病
高光谱成像
最大类间方差法
principal component analysis, diseases, grading, rice leaf blast, hyperspectral imaging, Otsu method
