Hyperspectral data are characterized by many bands, but the strong correlation between the adjacent bands makes it redundant and ineffective, which becomes a challenge for information extraction. To solve this problem, this paper proposes a method combining wavelet packet transform and weight SAM. First, radiometric calibration, reflectivity calculation and wavelet denoising are conducted on the raw hyperspectral data. Second, by using daubechies4 as the wavelet basis, the target and reference spectra are both processed with an 8-layer wavelet packet decomposition, and information entropy feature vector is based on the decomposition coefficient to characterize the features of the raw spectra. Third, efforts are made to find the feature range, in which the difference in information entropy feature vector between the target and reference spectra is huge, and a weight in this range is set. Finally, alteration minerals are mapped according to the SAM principle. In this study, experiments were conducted to testify the feasibility and effectiveness of the method by using the HySpex SWIR hyperspectral core data of a volcanic type uranium deposit in southern China. The results indicate that the new algorithm can efficiently characterize the raw spectral information by using the wavelet packet information entropy feature vector. Furthermore, the local characteristic information can be extruded by setting difference feature range, and different minerals can be more easily distinguished by this method with a total classification accuracy of 75.33% and a kappa coefficient of 0.706 3. The method proposed by the authors performs better than the traditional SAM algorithm and has a great application potential.