Automated spalling characterization with a parallel laser line-camera and U-Net++ with Kolmogorov-Arnold networks (UPPKAN)

The spalling of reinforced concrete (RC) structures threatens structural safety, and requires accurate quantification for condition assessment and repair planning. While point cloud-based methods have gained traction, they often lack efficiency and robustness in field applications. This study presents a compact, handheld parallel laser line-camera (PLLC) system capable of rapidly and automatically localizing and quantifying concrete spalling in terms of both surface area and depth using a single image. A segmentation model, known as UPPKAN, integrates U-Net++ with Kolmogorov–Arnold Networks (KANs), to extract fine-grained spalling boundaries. Projected laser lines are further exploited to determine relative spalling location and reconstruct depth profiles via line distortion analysis, ensuring robust measurements even under non-perpendicular photography. Laboratory and field experiments validate the system's accuracy, stability, and adaptability under diverse lighting and surface conditions. The proposed system offers a low-cost, efficient solution for automating routine inspection, especially for difficult-to-access locations such as ceilings. • Portable system enables fast detection and measurement of concrete spalling. • Spalling area and depth estimated from single image using laser line analysis. • UPPKAN model improves edge segmentation for irregular surface damage. • System achieves <10 % area error and < 0.5 mm depth error under various lighting conditions. • Suitable for ceiling inspection and adaptable to other irregular surface measurements.