Multi-Focus Imaging and U-Net Segmentation for Mesoscale Asphalt Film Structure Analysis—Method for Characterizing Asphalt Film Structures in RAP

Abstract

This study presents a high-fidelity image acquisition method for asphalt film structure to address the challenge of capturing mesoscale structures, especially fine mineral filler and asphalt mastic. The method is particularly applied to the analysis of the mortar structure in reclaimed asphalt pavement (RAP) mixtures. A digital camera combined with image stacking and texture suppression techniques was used to develop a reproducible imaging protocol. The resulting sub-pixel images significantly improved clarity and structural integrity, particularly for particles smaller than 0.075 mm. U-Net-based segmentation identified 588,513 aggregate particles—34 times more than in standard images (17,428). Among them, 95% were smaller than 0.075 mm compared to just 45% in standard images. Furthermore, segmentation accuracy reached 99.3% in high-resolution images, surpassing the 98.1% in standard images. These results confirm the method’s strong capability to preserve microscale features and enhance fine particle recognition, making it more effective than conventional imaging approaches. This study bridges physical and digital workflows in asphalt material analysis, offering a scalable, reproducible pipeline for fine-structure identification. The methodology provides foundational support for data-driven pavement modeling, material optimization, and future integration into digital twin frameworks for intelligent infrastructure systems.