Hyperspectral data is becoming increasingly popular and widely used. Accurate classification of the high ground objects is one of the core application of hyperspectral remote sensing technology. Extracting feature from hyperspectral data is an effective method for classification. Deep learning is the new areas of machine learning research. It has multilayer perceptron structure so that it can learn to portray a more essential characteristic, and have better results in the field of image classification and visualization. DBN is a normal model of deep learning network. A hyperspectral image classification model based on DBN is constructed by using high dimensional feature of hyperspectral data and combining the spatial structure of hyperspectral data. Experiment shows that high spectral image classification based on DBN can get better classification results.