The binocular vision measurement of coal flow is a key technology to realize energy-saving and safe operation control of belt conveyors. However, the texture and color features of coal samples are single and repeated. The coal particles′ internal gaps are distributed uneven. These factors have seriously influence on the accuracy and real-time performance of coal flow measurement. To address these issues, a binocular vision measurement method for coal flow of belt conveyors is proposed, which is based on deep learning. Firstly, the coal image is preprocessed through correction, segmentation and enhancement. Secondly, a PSM-Net model for coal stereo matching is formulated, which is also based on deep learning. The fine-tuning learning mechanism is adopted to train the PSM-Net model to obtain the coal material volume. Then, based on the two-dimensional characteristics of coal material, a calculation method for coal packing rate based on discrete element method is proposed to achieve coal packing density. Finally, the coal flow of the belt conveyors is calculated, which is based on the obtained volume and packing density. Experiment results show the effectiveness of the proposed algorithm. The accuracy of the binocular vision measurement of coal flow reaches 98. 704 3% , and the calculation rate reaches 1 127 ms per frame.