DAU-Net: A Dual-Attentive U-Net for Enhanced Semantic Segmentation in Underground Infrastructure Inspection

Overview

DAU-Net is a pioneering deep learning model tailored to meet the demands of industries involved in infrastructure inspection and public safety. Designed to detect defects in sewer and culvert pipes, DAU-Net analyzes CCTV footage, offering unparalleled performance in environments where manual inspection is not only challenging but also costly and risky.

Real-World Applications and Value

Infrastructure failures can have severe consequences on businesses and communities. DAU-Net:

** Reduces Costs: Accelerates the inspection process for technicians and engineers, saving time and lowering operational expenses.
** Enhances Safety: Automates inspections in hazardous environments, mitigating risks for human inspectors.
** Increases Accuracy: Provides consistent and reliable detection of structural defects, supporting timely repairs and preventative maintenance.

Performance Highlights

DAU-Net has demonstrated state-of-the-art performance:
** Culvert-Sewer Dataset: Achieves a 75.9% mean Intersection over Union (IoU), outperforming previous models by over 30%.
** Cell Nuclei Benchmark: Records an 83.6% mean IoU, showcasing broad applicability across datasets with complex structures.

(DAU_FI_2 (2) (1).png)