In order to solve these problems, this paper proposes an automatic identification method for pipe pile cracks based on pipe pile cleaning robots. A lightweight network detection algorithm YOLOv8-MLLA-Mobilenetv4-WIoU(MWM-YOLO) was designed. Capture low-quality defect images in a muddy water environment and augment the data to expand the dataset. For low-quality images under muddy water, in view of the suppression effect caused by the mismatch between image enhancement and object detection, MLLA is used to accurately focus on key feature areas, which can effectively suppress background interference while maintaining high-resolution output, so as to enhance the synergy between image enhancement and object detection. At the same time, the latest Mobilenetv4 backbone network is used to reduce the number of parameters and calculations of the characteristic network. On this basis, considering that low-quality image data annotation inevitably contains low-quality examples, the WIoU loss function is used to replace the loss function in the original YOLOv8 network model to improve the generalization performance of the model. The experimental results show that the weight of the MWM-YOLO model is 14.9 MB, which is 30.3% less than that of the original model. The average accuracy reached 89.1%, and the inference speed was 137.54 fps, which was better than other models. Compared with the original network, the improved network model can be lightweight deployed to edge computing devices while maintaining the accuracy of defect identification, providing technical support for underwater pipe pile cleaning robots.