The normalized difference index originates from NDVI, which has been widely used in remote sensing applications and to guide the development of normalized difference indices in other fields due to its excellent performance. However, NDVI is insensitive to changes in high vegetation cover and performs poorly in some variable inversion tasks. In fact, this insensitivity comes mainly from the scaling strategy of NDVI for feature bands, rather than in its form of normalized difference. Furthermore, the choice of the coefficients in the above indices is often determined by the evaluation metrics in the experimental results, which is difficult to generalize due to the lack of mathematical interpretation. In this work, we revisit the scaling process of the normalized difference index to feature bands and propose a general normalized difference index, GND, in which four positive scaling coefficients are added and the value range, sensitive region and saturation point of GND are allowed to be freely changed by these four coefficients. Moreover, we supplement the lost band information of GND and use it to construct a new index, RI, and the result is that the mapping of the feature bands bx and by, used by GND, to GND and RI is bijective, and the original information is preserved. Our experiments demonstrate that mapping the original bands to GND and RI can guide the classifier based on spectral information to learn more generalized features, and a higher classification accuracy can be achieved in the latest remote sensing image semantic segmentation.