The sewer system plays an import role in our life, it is applied to collect and dispose domestic wastewater and rainwater. Regular inspection is necessary for the normal operation of the pipeline. Existing pipeline defect detection technologies are mostly based on manual inspection or supervised defect detection, which are time-consuming and laborious. In this paper, we propose a new weakly supervised attention guided image enhancement network for sewer pipeline defect detection. And, we establish a sewer pipelines dataset of 15 classes for weakly supervised learning. In our network, we use resnet-50 to generate the feature maps and attention maps of the images. Then the image enhancement module is designed to supplement details of the object. Finally, the original image and enhanced image will be used as input image onto training. The trained network can locate a variety of pipeline defects with highlight. Experimental result shows that our model performs better than other weakly supervised models on the particular pipe dataset.