Tree root intrusion in sewer pipelines is a common issue in urban drainage systems, which can lead to pipeline blockages, ruptures, and even collapses, severely affecting the normal operation of the drainage system. Traditional tree root detection methods are inefficient and prone to subjective influences. In recent years, with the rapid development of deep learning technology, tree root recognition methods based on object detection algorithms have begun to attract significant attention. This paper proposes an improved YOLOv5 algorithm for tree root detection in sewer pipelines. Firstly, the accuracy of bounding box regression is optimized by replacing the improved bounding box regression loss function. Secondly, an attention mechanism based on standardized attention module is introduced in the Neck part of YOLOv5 to enhance the adaptability of the model to complex background. In the experimental process, we constructed a self-built dataset containing 1000 tree root images and expanded it to 5000 images using data augmentation techniques. The experimental results show that the improved YOLOv5 algorithm reduced the bounding box regression loss convergence value by 0.0009, increased the by 0.7%, and increased the :0.95 by 1.1%, while maintaining a high detection speed (FPS = 81). This indicates that the improved algorithm can more accurately identify tree roots in sewer pipelines, providing an efficient technical means for the maintenance and management of drainage pipelines.