SHARP-Net: A Refined Pyramid Network for Deficiency Segmentation in Culverts and Sewer Pipes

Undetected defects in culverts and sewerpipes pose significant risks to public safety, including infrastructure collapses, flooding, and transportation disruptions. Manual inspections are time-consuming, costly, and prone to human error, while existing automated methods often struggle with real-world challenges, such as occlusions, irregular defect shapes, class imbalances, and high resource demands. This article introduces the semantic Haar-adaptive refined pyramid network (SHARP-Net), a novel architecture for semantic segmentation that delivers precise and efficient defect detection. SHARP-Net combines multiscale feature fusion, depthwise separable convolutions, and fine-tuned Haar-like features to enhance performance while reducing computational complexity. Evaluated on the culvert–sewer defects dataset and the DeepGlobe land cover dataset, SHARP-Net demonstrated superior performance. The base SHARP-Net (excluding Haar-like features) outperformed state-of-the-art methods, including U-Net, CBAM U-Net, ASCU-Net, FPN, and BiFPN, achieving an average improvement of 16.19% and 11.74% in IoU scores on the respective datasets, with IoU scores of 77.2% and 70.6% . The integration of Haar-like features further boosted performance by 2% –4% IoU, showcasing their effectiveness in capturing critical structural details across diverse datasets. By automating defect detection, SHARP-Net addresses critical infrastructure challenges, offering a scalable solution that improves safety, reduces inspection costs, and accelerates maintenance processes.