基于SVC的苎麻褐斑病叶片高光谱识别

对叶片高光谱信息进行分析,实现苎麻褐斑病快速无损的诊断,对提高苎麻产量和品质有重要意义。利用FieldSpec3便携式地物光谱仪和手持叶片夹持器,采集了430个苎麻褐斑病叶片和健康叶片高光谱数据。提出了一种基于离散系数的子波段主成分分析PCA方法来提取特征变量。同时,为了探讨不同主成分个数对模型的影响,分别以1~10个主成分作为特征变量,采用支持向量机分类SVC方法建立苎麻叶片褐斑病识别模型。结果表明:1)波段A(511~636 nm)、波段B(690~714 nm)、波段C(1406~1511 nm)和波段D(1870~2450 nm)离散系数较大,是建立识别模型的敏感波段;2)4个子波段中,波段C建模效果最好,选择5~10个PCA主成分作为特征变量建立SVC识别模型时,在主成分个数相同的情况下,其正确率可以达到90%以上,总体高于全波段和其他子波段。基于离散系数筛选较敏感的子波段进行PCA,选择合适的主成分个数作为特征变量,建立苎麻叶片褐斑病SVC识别模型是可行的,为开创一种新的苎麻褐斑病诊断方法提供技术支持。

Hyperspectral Identification of Ramie Leaves with the Brown Spot Based on the SVC

The quick and non-destructive diagnosis of the ramie leaves with the brown spot through the analysis of the hyperspectral information on leaves is of great significance to improving the production and the quality of ramie.Total 430 hyperspectral data,which are related to the ramie leaves suffering from the brown spot,and to the healthy ramie leaves,are collected by using a portable ASD spectrometer called FieldSpec 3 and a handheld leaf clip.From this,we present a sub-band principal components analysis(PCA)approach based on the coefficient of variation to extract the feature variables.Meanwhile,using respectively 1 to 10 main factors as the feature variables and together with the support vector classification(SVC)approach to build the identification model of the brown spot of ramie leaves.We get the following results:1)The four sub-bands,i.e.the band A(511~636 nm),the band B(630~714 nm),the band C(1406~1511 nm)and the band D(1870~2450 nm),have more higher coefficients of variation than other bands and are the sensitive bands for building the identification model;2)The modeling effect of the band C is the best one of those 4 bands,and while using 5 to 10 PCA main factors as the feature variables to construct the SVC identification model,the accuracy rate is over 90%with the same number of main factors,which is apparently higher than that of the full-wave band and other sub-bands.Therefore,it is practicable to construct the SVC identification model on the brown spot of ramie leaves by performing PCA over the sub-bands that are more sensitive to the coefficient of variation and by appropriately choosing the number of main factors that act as the feature variables,which provides the technical support for developing a new diagnose of the brown spot of ramie.