Abstract: Aiming at problems of insufficient fault state sample data and low accuracy in fault diagnosis of belt conveyor by traditional shallow neural network, a fault diagnosis method of belt conveyor based on synthetic minority oversampling technique (SMOTE) and deep belief network (DBN) was proposed. Fault state sample data of belt conveyor is generated by SMOTE to overcome imbalance distribution of the sample data. The sample data is input into DBN, fault features in the data are extracted by means of unsupervised layer-by-layer training, and fault diagnosis ability is optimized by means of supervised fine-tuning to achieve accurate fault diagnosis of belt conveyor. The simulation results show that the method improves fault diagnosis accuracy of belt conveyor.