Global change research consistently requires the combined use of cross-sensor images with similar spectral wavelengths. To meet the temporal resolution and coverage requirements of remote sensing applications, new requirements are proposed for remote sensing image processing technology. Such technology requirements are related to how to obtain geometric consistency between cross-sensor/multi temporal data, how to obtain radiometric normalization, and how to obtain land cover class labels that are consistent between cross-sensor/multi temporal data. They are also related to highly automated processing. For the above requirements, we propose a framework for the automatic processing of remote sensing images with "an invariant feature point set"（IFPs） as the control data set. Specifically, we combine the spatial and temporal alignment in geometry space, radiation space, and land cover class space into a unified framework and thereby provide an indirect means to achieve fast processing. The key technologies for building IFPs are also reviewed.

remote sensing image processing;invariant feature point set;control point set;image controt point;Pseudo-In variant features;active learning;deep learning