Abstract: Aiming at the problems of belt damage and tear caused by large-scale gangues or irons entering the coal belt system, a kind of Faster-RCNN+ double-sided feature pyramid networks (DSFPN) coal-transport belt foreign object recognition model is proposed. Based on the deep learning target detection framework Faster-RCNN, the model proposes DSFPN for the improvement of FPN structure. DSFPN solves the multi-scale problem of belt foreign objects through the bottom-up and top-down multi-scale feature fusion process. The test results show that the DSFPN proposed in this paper can effectively improve the detection ability of small-sized foreign bodies such as small pieces of gangues, and improve the recognition accuracy of large-sized foreign objects such as bolts and large gangues.