Abstract: In order to solve the problems such as poor recognition rate of foreign objects in transport belt caused by insufficient light and harsh underground environment, an improved YOLOv7 model is proposed in this paper. By introducing attention module into its feature extraction network, it can pay more attention to important feature information, so as to effectively improve the feature extraction ability of the network. At the same time, the model's original foundation network is replaced with a lighter backbone feature extraction network, Ghost Bottleneck, to speed up the model detection. The experimental results show that the proposed algorithm achieves 97.3% accuracy in belt foreign body recognition application, which is 3.6% higher than the baseline model, and the recognition rate is 8.5% higher, indicating the effectiveness of the algorithm. As an important part of coal transportation system, underground transport belt plays a vital role in ensuring normal production and personnel safety in mine. However, due to the complexity and uncertainty of the underground environment, the transport belt is often faced with the threat of non-coal foreign matter, such as material accumulation, equipment failure, personnel misoperation, etc. These foreign bodies may not only lead to abnormal operation of the belt, but also may cause serious safety accidents. Therefore, it is of great theoretical significance and practical application value to study the detection method of foreign matter in underground conveyance belt.