The invention discloses a hyperspectral image classification method based on NSCT and an SAE. The method is realized through the following steps: 1) inputting images; 2) carrying out pretreatment; 3) carrying out non-subsampled contourlet transform; 4) selecting coefficient sub-bands; 5) selecting a square neighborhood image block; 6) extracting local texture features of each pixel; 7) obtaining a three-dimensional image matrix; 8) selecting training samples; 9) constructing a stacked auto-encoder SAE; 10) carrying out nonlinear transformation operation; 11) fine-tuning parameters in a model; and 12) outputting a hyperspectral image classification result. The method can make full use of the local texture features of hyperspectral images, has very high classification accuracy, can be used for classification of hyperspectral images in the geological exploration field and provides reference for ground object image drawing.