Because of the high data dimensionality of hyperspectral data, conventional methods are difficult to obtain satisfied results in the study of hyperspectral classification for materials on the ground. In the process of feature images extraction based on wavelet multiresolution fusion, a new method, which uses a feature vector consisting of multiple spacious salient features to determine fusion weights, wass presented. The algorithm can effectively reduce the hyperspectral data dimensionality and obtain the feature images for the successive classification. The experiments on AVIRIS data show that classification accuracy by using the new method is higher than that of using the conventional methods in determining weights.