Hyperspectral remote sensing is capable of reflecting the detailed spectrum of ground objects; thereby it can be used for anomaly identification, quantitative retrieval, state diagnosis and fine classification. Wavelet transformation, which is viewed as 'mathematical microscope' with the capability of multi-resolution analysis, can be used for dimensionality reduction and feature extraction to hyperspectral remote sensing data, especially feature extraction at different scales. The data sets used in this study include: spectral data of several ground objects in USGS spectral library, and spectral data of some pixels in an image captured by the airborne image spectrometer OMIS II. Spectral absorption features of ground objects are quite important for ground object recognition. Wave troughs of spectral curve, which represent strong spectral absorption at some specific wavelengths, are extracted and analyzed quantitatively using wavelet transformation. Spectral angle (SA) is selected as similarity measure indicator because of its effectiveness to hyperspectral remote sensing data. The experiment results demonstrate that multi-resolution analysis of wavelet transformation provides excellent performance in spectral feature extraction and spectral similarity measure, so it can be used to target identification and image classification effectively.