Abstract: Near ground remote sensing is a common technology to obtain the information of crop canopy components. By focusing on the spectral information from canopy leaves which is often affected by soil background and shadow, the physicochemical parameters of crops can be accurately retrieved. By analyzing the spectral and texture differences of wheat canopy components (sunlit/shadowed leaves, soil and panicles), it was found that the components with large spectral differences can be distinguished well by using spectral index, but it is difficult to separate wheat panicles and leaves with similar spectral curves according to previous studies. On the contrary, mathematical morphology by using the great difference between different components can extract pure wheat leaves well, but limited by distinguishing sunlit leaves and shadowed leaves. By combining these two methods, the spectral and spatial differences of canopy components can be fully considered to achieve the goal of accurate classification. To accurately classify and extract the information of crop canopy leaf components, a canopy component classification method combining spectral index and mathematical morphology was proposed, that is, firstly, wheat panicles were extracted from the canopy by mathematical morphology, and then remaining components were classified by constructed new spectral classification index (NSCI) according to their spectral differences. In order to prove the superiority of the proposed method, it was used to compare classification accuracy of different growth stages with other conventional classification methods, such as ISODATA、Maximum Likelihood Estimation and spectral index. In addition, the correlation between normalized difference spectral index (NDSI) of wheat leaf components and leaf nitrogen content was analyzed quantitatively. The results showed that NSCI could show exceptional performance in distinguishing various canopy components before heading stage considering that there were obvious differences in specific spectral bands, but the classification accuracy at heading stage was decreased because wheat panicles spectrum curves was similar to that of leaves. The classification method combining spectral index and mathematical morphology can better eliminate the interference of wheat panicles on the extraction of sunlit/shadowed leaves and distinguish other components (overall classification accuracy= 97.80%, Kappa coefficient=0.97, running time=3.87 min), and the classification accuracy and efficiency of this method are better than the conventional classification methods (ISODATA and Maximum Likelihood Estimation). By calculating the correlation coefficients between the NDSI constructed by different canopy components and leaf nitrogen content, it was found that NDSI containing both sunlit and shadowed leaves reflectance information is the most sensitive index to leaf nitrogen content, this may be contributed to sunlit leaves or shadowed leaves alone only contain part of the information of pure leaves, and the reflectance spectrum of sunlit leaves may contain a lot of useless information from specular reflectance, which leads to the offset of sensitive bands and the decline of correlation with leaf nitrogen content. This study proved that when predicting crop physicochemical parameters, it is necessary to eliminate the influence of background information, and ignoring signals from shadowed leaves may lead to incorrect estimation. This study could provide reference for the classification of canopy components of other crops and promote the quantitative inversion of agronomic parameters in precision agriculture.