Vibration characteristics of planetary rovers may contain features that can be effectively used for terrain classification, terrain properties estimation, and terrain traversability assessment. However, the conventional vibration features selected to represent the terrain induced wheel response vary with the wheel speed, which could cause misclassification. In order to extract speed-independent vibration features of wheels, this paper proposes vibration model of wheels traversing different types of terrains with consideration of the dynamics of lug-terrain interaction. Based on the proposed model, low-dimensional vibration features including frequency, amplitude, mean value, and mean square value are extracted from statistical features of vibration acceleration. Nondimensionalization is performed to make them independent on wheel speed. The features can represent the terrain characteristics decoupled from the wheel speed so that they can be used to reduce terrain misclassification in the future. Experimental results show that the vibration features extracted are independent on wheel speed and effective for terrain classification.