Particle Morphology Analysis Method Based on U-Net and Watershed Algorithm

The morphology of particles significantly influences their physical and mechanical properties. Therefore, the first step in particle morphology analysis is the identification of their outlines. Taking pebbles as an example, this study initially collects images of pebble particles and employs data augmentation techniques to expand the sample images. Subsequently, a lightweight U-Net model is proposed in this paper. Coupled with the watershed segmentation algorithm, efficient identification of pebble particle outlines is achieved, with an accuracy rate reaching 97%. Finally, the paper summarizes three hierarchical indicators describing particle morphology. Morphological indicators are calculated for manually annotated images and images identified using the proposed method. Comparative analysis indicates that the maximum error in particle morphology analysis using this method is 17%.