It is very important to study the terrain classification for the performance of autonomous mobile robots, especially in the situation of planetary surface exploration. A wheeled mobile robot that installed accelerometers in x, y, z directions and microphone in z direction in the left front wheel arm is used to get the vibration signals of wheel-terrain interaction to calculate the terrain classification by traversing on sand, gravel, grass, soil and asphalt terrain with six different velocity respectively. This method can avoid the defect of visual classification method, i.e. the impact of illumination changes and the covering. Based on time amplitude domain analysis method, a new voting decisions classification algorithm has been proposed to deal with the situation of same number of votes cases via κ-Nearest Neighbors(kNN) algorithm, and the proposed algorithm has been validate d by corresponding experiments.