Abstract: Aiming at the problem that unreliability of cognitive radio single-channel transmission in mountain area, a new type of spectrum sensor and channel access strategy is proposed. At first, in order to solve the shortcomings of low accuracy and weak versatility of traditional mountain area models, the Global Mapper software is used to perform 3D high-precision restoration of mountain area. Then considering the problems of low single-channel transmission efficiency and insufficient communication guarantee capability, a multivariate hidden Markov algorithm is obtained on the basis of unary hidden Markov. Finally, it is verified with channel quality data in a mountainous area in Kunming. The experimental results show that when the number of channels is 8, the collision probability and the secondary user’s successful transmission probability of the improved multivariate model are 24% and 65%, respectively, which has a higher access rate and network throughput.