Remote sensing image registration based on fusion of spatial transformation and dense convolution

In order to achieve high-precision registration of remote sensing images, this paper proposes a remote sensing image registration method based on spatial transformation and dense convolution fusion. First, the dense convolutional network improved by the spatial transformation module is used for feature extraction to improve the robustness of feature points; then use improved Grid-based Motion Statistics (GMS) algorithm for feature matching, and use the homography matrix-based method to eliminate mismatched point pairs to improve matching accuracy; finally, the transformed model is used to solve the transformed model to achieve accurate registration of remote sensing images. Experiments show that the proposed method can effectively improve the correct matching rate of feature points, has higher registration accuracy and stronger robustness.