Designing an efficient message routing mechanism can improve the operational efficiency of the Mobile Sensor Network (MSN). An efficient clustering routing mechanism based on deep belief networks towards MSN is proposed in this paper. A feature extraction method on sensor node contact information is designed to extract its main features from the complex node encounter information. Then a sensor node clustering method, based on the extracted main features, is presented. Furthermore, a clustering message routing mechanism is designed, which comprehensively consider the node contact tightness and energy state to dynamically select cluster head nodes for message routing within and outside the cluster. Simulation results show that the proposed method can improve the average delivery rate, average delivery delay, and network lifetime performances by more than 14%, 24%, and 23%, respectively.