The full characteristics cannot be obtained by single filter in spatial information extraction of hyperspectral image. Combining bilateral filter and domain transform filter of normalized convolution, an improved algorithm of classification was proposed. The method advanced an adaptive fusion of spatial information for classification optimization. Firstly, bands of hyperspectral image were sampled into two groups. Secondly, spatial information of the two group images was extracted by the bilateral filter and the normalized convolution respectively. finally, the two kinds of spatial information were combined and classified by support vector machine. The experiments show that the algorithm is better than original support vector machine with the pure spectrum information, dimensionality reduction, the spatial-spectral information, and the method of edge-preserving filtering and recursive filtering. the performance of hyperspectral image classification algorithm is greatly improved, although training samples were only 5% and 3%, the verall accuracy of Indian and Pavia can reach 96.95% and 97.89% respectively, with 2%~13% higher than other algorithms, and the effectiveness of the method is fully verified.