Sparse-expression-graph-based semi-supervised classification method for hyperspectral images

The invention discloses a sparse-expression-graph-based semi-supervised classification method for hyperspectral images. The method comprises: according to pixel data of a hyperspectral image, a category probability matrix is obtained, wherein the pixel data include pixel data with categories marked and pixel data with unmarked categories; on the basis of the category probability matrix and spatialdomain information of the pixel data, a regular term is constructed, a constrained sparse expression objective function is obtained based on the regular term, and a similarity weight matrix is obtained by using the sparse expression objective function; and on the basis of the similarity weight matrix, the category of each pixel in the hyperspectral image is obtained by mark propagation. Therefore, problems of high sensitivity to noises, the need of manual parameter setting, and insufficient discriminating of the existing graph construction method can be solved. The sparse-expression-graph-based semi-supervised classi