Abstract: This paper presents a SIFT based low gray hyperspectral image lossy fast registration method. This method on the basis of low gray hyperspectral image lossy image registration and low gray hyperspectral image lossy down sampling, to reduce sampling benchmark Gray hyperspectral image lossy and to measure the low gray hyperspectral image lossy SIFT features are extracted, by Hyperspectral on the basis of low gray effectively matching feature points of image, get a certain number of low gray hyperspectral image control point loss, the final use of triangulation of low gray a hyperspectral image in local matching, and triangulation in repeated low gray hyperspectral triangulation image matching. The experimental results show that the proposed method not only control the number of the feature points, but also reduce the matching time of image registration, to effectively improve the image matching accuracy at the same time, also on the efficiency feature matching Have greater improvement.