In view of the huge scale of urban underground pipe network, the traditional manual detection method can no longer meet the needs of the current projects. In this paper, the MobileNetv3?YOLOv7 network model is proposed as the algorithm for target detection of underground pipeline defects to improve the accuracy and speed of detection. First, the pipeline image dataset is preprocessed, and the input image is grayscale and resampled to balance the number of samples. Secondly, the lightweight network MobileNetv3 and YOLOv7 network frameworks are combined to increase the BiFPN feature pyramid structure to improve accuracy. Then, in terms of data processing, Mosaic data augmentation is used to improve the robustness of the model. Finally, a comparative experiment with the YOLOv7 network model is designed to verify the feasibility of the model. In this paper, the MobileNetv3?YOLOv7 network model is verified under the framework of Pytorch experiment, and the experimental results show that the model greatly reduces the amount of parameter calculation and improves the average accuracy.