Main content area

A Band Math-ROC operation for early differentiation between Sclerotinia sclerotiorum and Botrytis cinerea in oilseed rape

Xu, Xing, Zhao, Yun
Computers and electronics in agriculture 2015 v.118 pp. 116-123
Botrytis cinerea, Brassica napus, Sclerotinia sclerotiorum, hyperspectral imagery, least squares, leaves, oilseed crops, principal component analysis, support vector machines, China
Brassica napus (oilseed rape) is a major oilseed crop widely used in China. The hygrophanous lesions inflicted by Sclerotinia sclerotiorum and Botrytis cinerea to oilseed rape leaves are initially indistinguishable and wrong identification is possible, resulting in inadequate treatment. A new Band Math-ROC algorithm to identify S. sclerotiorum and B. cinerea infection on oilseed rape was developed in the research. The characteristic (ROC) curve was operated by using hyperspectral imaging technology combined with principal component analysis (PCA), least squares support vector machine (LS-SVM), Spectral Math and receiver. The algorithm is able to identify early-stage oilseed rape lesions. The oilseed rape leaves artificially infected with S. sclerotiorum and B. cinerea were collected and submitted to hyperspectral imaging, scanned at each of the initial three days after infection. To reduce the computational complexity, Band Math-ROC algorithm was applied and the characteristic bands (748.23nm, 681.95nm) were computed to extract characteristic values; finally, a ROC curve was created to select the optimal threshold, 4.5 for band ratio operation and 0.6 for band difference operation. The results show that the algorithm proposed in the research can effectively distinguish early lesions caused by S. sclerotiorum and B. cinerea in the first two days after infection. And the classification accuracy in the third day is lower.