Application of deep learning for characterizing microstructures in SBS modified asphalt

Microstructures in asphalt, often resembling bee structures, are pivotal in influencing asphalt performance and, by extension, sustainable fuel production. This study employs deep learning techniques to investigate the impact of different Styrene–Butadiene–Styrene (SBS) modifiers on asphalt microstructures, akin to bee structures. The employed deep learning model was trained on a diverse dataset comprising 200 of images sourced from testing. The dataset was carefully curated to address specific challenges in data labeling precision. This involved individualized labeling sessions and adjustments in the number of targets per image, contributing to enhanced precision and increased dataset size. The research begins with the development of a deep learning model trained on a dataset comprising images featuring bee-like structures within asphalt. The model excels in accurately identifying and segmenting these structures. Subsequently, the deep learning approach is compared with existing methods for bee structure segmentation to establish its precision and superiority. Employing frequency distribution histograms, the distribution patterns of bee structures within various types of SBS-modified asphalt is analyzed, quantitatively assessing the influence of diverse modifier types on these microstructural attributes. The findings in this study underscore the deep learning model's efficacy in recognizing and segmenting bee structures with introduced metrics effectively capturing the distinctive characteristics of various asphalt microstructures. This study paves the way for comprehensive analyses of microstructural metrics, including parameters such as perimeter, area, quantity, and related indicators, thus contributing to the development of fundamental asphalt structural units suitable for processes like molecular simulation and finite element analysis. Moreover, it propels the application of deep learning methodologies in the realm of road materials research, opening new avenues for innovative explorations that can ultimately benefit sustainable fuel production.