Deep Learning Image Segmentation to Assess Field Ballast Angularity Index

The angularity of aggregate particles has a significant influence on the performance of aggregate layers. To characterize the shape properties of aggregate materials, Computer Vision technology has been utilized, such as the Enhanced University of Illinois Aggregate Image Analyzer (E-UIAIA). However, such an approach can only be used for individual particles in a laboratory setup due to the limitations of the segmentation algorithm. In this paper, a novel deep-learning-based approach is being proposed to evaluate the average angularity index (AI) from the high-quality segmentation results produced by the deep-learning algorithm. Experiments were conducted to compare the AI values computed by the proposed approach with results obtained with E-UIAIA. Results confirm the accuracy and robustness of the proposed method. A case study at the Transportation Technology Center’s (TTC) High Tonnage Loop (HTL) illustrates the use of the proposed technology in practice to interpret the relationship between ballast angularity and field fouling conditions, and to serve as a guidance tool for ballast inspection and maintenance.