Hyperspectral images have continuous spectral information of features and have great potential for shadow detection, but high band redundancy requires band preference. Normalized Shaded Vegetation Index (NSVI) can expand the spectral difference, and the application of NSVI in hyperspectral images will identify shadows more effectively. ZY1-02D satellite is the first hyperspectral operational satellite independently developed and successfully operated in China, with a large data signal-to-noise ratio and strong coverage capability, and it is important to perform accurate shadow detection on this hyperspectral image. In this paper, ZY1-02D AHSI images were used as experimental data to extract and analyze the spectral reflectance of vegetation in bright areas, vegetation in shaded areas and water bodies, and Combining Competitive Adaptive Reweighted Sampling (CARS) and Successive Projection Algorithm (SPA) to filter the main wavebands that can effectively distinguish typical features, the characteristics of the algorithms are considered to select the characteristic wavebands further to construct NSVI. The optimal threshold value is determined by the step method to classify the images, and the best band for constructing NSVI is compared in terms of image element value distribution, classification accuracy and spectral enhancement effect. A comprehensive evaluation is made by combining different shadow indices, bands and images to verify the significance and universality of the method in this paper. The results show that band 32 and band 73 are the best bands for NSVI construction, corresponding to the Red band and NIR band, respectively; the classification accuracy of NSVI constructed by different bands is generally higher than 90%, and the classification accuracy of NSVI constructed by the best band is 94.33% with a Kappa coefficient of 0.832 8, which is the best classification effect; NSVI can enhance the spectral difference between typical features and alleviate the “easy saturation” phenomenon of Normalized Difference Vegetation Index, and the small peaks generated by the accumulation of water bodies in this image is helpful to extract water bodies; The classification of NSVI in ZY1-02D AHSI image is better than Normalized Different Umbra Index and Shadow Index, and the classification accuracy in another scene image also reaches 93.55% with a kappa coefficient of 0.816 7; Therefore, the wavebands filtered by the algorithm are representative, and the NSVI constructed by the best waveband has better shadow detection ability in ZY1-02D AHSI images, which has a certain reference and significance for hyperspectral image shadow detection and construction of vegetation index.