hyperspectral image semi-supervised dimension reduction method based on local sparse embodiment

The present invention relates to a hyperspectral image semi-supervised dimension reduction method based on local sparse embodiment. The method comprises the following steps of: the step S1. setting adata set X={x1, x2, ..., x1, x1+1, ..., x1+u} in a high-dimensional space RD, wherein 1+u=N, the former l samples X1 are samples with class tags, the class tags are c, the number of each class of samples is Ni, i=(1, 2, ..., c), and the later u samples Xu are samples without class tags; the step S2. constructing a sparse coefficient matrix S through sparse representation; the step S3. based on a semi-supervised local sparse embodiment projection algorithm, constructing a projection matrix W; and the step S4. according to the projection matrix W, obtaining a low-dimensional sub space Y=WTX={y1,y2, ..., yN}. Through semi-supervised dimension reduction of local sparse embodiment, the method can employ class tag information of data, can maintain the data local features and can reduce the image noise information so as