基于多光谱成像选取四季豆叶片的特征波段

Selection of Feature Bands for Phaseolus vulgaris Leaves Based on Multi-Spectral Imaging

在400～720nm波段范围，基于液晶可调谐滤波器（LCTF）和CMOS相机组合的多光谱成像系统，以四季豆叶片为研究对象每隔5nm进行成像。根据图像亮度信息法和波段指数法的相关原理，首先分别计算得到各波段四季豆叶片的波段指数值和可识别度；然后对四季豆叶片的波段指数值和可识别度进行排序，综合图像的灰度离散、亮度信息丰富和波段的相关性小等特点，得出545、630、645、720、650和570nm波段有较大的波段指数值和较好的识别度；最后根据最小欧氏距离法和光谱角度匹配法分别对四季豆叶片的特征波段的分类精度予以计算，两种方法的分类精度分别为100．00％和83．33％，得出选取的特征波段对四季豆叶片具有较好的分类精度。因此，545、630、645、720、650和570nm波段可作为四季豆叶片的特征波段。

Multi-spectral images of Phaseolus vulgaris leaves at the wavelength range of 400-720 nm with an interval of 5 nm are captured by using a multi-spectral imaging system which mainly consists of liquid crystal tunable filter （LCTF） and CMOS camera. Firstly, according to the principles of image brightness and band index, the value of band index and identifiability for Phaseolus vulgaris leaves are calculated respectively among every band. Then, through sorting the value of band index and identifiability for Phaseolus vulgaris leaves, it can be concluded that bands 545, 630, 645, 720, 650 and 570 nm have preferable identification with considering the characteristics of discrete gray levels and rich brightness of images and little correlation coefficient among different bands. Finally, the classification accuracy for Phaseolus vulgaris leaves is calculated according to the principles of minimum Euclidean distance and minimum spectral angle matching. The classification accuracy of characteristic bands for Phaseolus vulgaris leaves is 100.00% and 83.33% separately through using these two methods. We can draw a conclusion that these bands have ideal classification accuracy. Therefore, bands 545, 630, 645, 720, 650 and 570 nm can be used as feature bands for Phaseolus vulgaris leaves.