Abstract
Quantifying the shape of granular particles poses a significant challenge in civil engineering, particularly in geotechnical, concrete, and asphalt applications, due to the complexity and irregularity of particle shapes. The wide variability in particle size complicates the application of various image processing techniques, and the resolution of the image used for analysis can significantly impact the accuracy of shape quantification. In this study, an innovative and comprehensive framework was developed for the precise detection, segmentation, and characterization of these particles in images taken by mobile phones. Even under less controlled conditions, such as variable lighting and shadows, the proposed pipeline consistently overcomes these challenges and maintains high detection and segmentation accuracy. A tiling strategy was firstly employed to preprocess large images, enhancing computational efficiency and detection rates for small objects. The YOLOv8 model was then trained and used to detect objects within segmented images, demonstrating superior detection capabilities by identifying small elements that are often overlooked by conventional methods. After detection, each object was cropped and isolated to generate individual images, eliminating errors in detecting neighboring elements as a single object. These images were further processed using “segment anything model”, achieving pixel-level accuracy in generating masks for each individual unit. Sobel edge detection was then applied to the masks, enabling precise boundary identification. Finally, the extracted boundary information was used to quantify form, angularity, sphericity, roundness and aspect ratio, providing detailed insight into the morphological and geometric properties of the detected particles. This pipeline integrates advanced techniques and offers a scalable, accurate solution for object detection and shape characterization, with broad potential for application across various engineering disciplines.