Deep Fusion of Intensity and Depth Data for Automated Spalling Detection in Tunnel Infrastructures

This paper introduces a deep learning-based methodology for the automated identification of spalling defects in tunnel linings, emphasizing the fusion of intensity and depth information. A novel network architecture is presented, leveraging Mobile Laser Scanning (MLS) data to generate a dataset of paired intensity and depth images. The network effectively integrates these multi-modal inputs to enhance the precision of spalling segmentation. Results demonstrate the superior performance of the proposed approach in comparison to methods relying solely on intensity data, highlighting the critical role of depth information for accurate defect characterization in complex underground environments.