The quantitative estimation of feldspar content in granites is helpful for the naming and classification of granites, and it provides basic data for subsequent studies on related geological processes. However, within the visible near- infrared to short- wave infrared range (0. 35~2. 5 μm), the traditional method of inverting rock types and mineral contents based on spectral absorption characteristic parameters is not suitable for minerals such as feldspar that lack diagnostic absorption characteristics. At the same time, the complexity of calculating the radiative transfer model based on physics limits the application of this method to a large extent in the inversion of rock mineral properties. In this paper, we establish a spectral index model for estimating the proportion of plagioclase to feldspar in granite based on multi- angle hyperspectral data. By combining different spectral preprocessing methods and spectral index types, we aim to overcome these limitations. Our results show that the double differences type index model, utilizing the measured dataset with continuum removal at the 2035 nm band (CRDDn2035), demonstrates good performance across different measured datasets, achieving an estimation accuracy of 0. 81. Through the study, we establish a spectral index model for estimating the proportion of plagioclase in granite, providing a new technical method and idea for the quantitative inversion of rock and ore information with weak absorption characteristics.