Automated SEM image processing for texture and morphology analysis

This thesis presents an automated workflow for processing scanning electron microscopy (SEM) images to analyze the texture and morphology of pharmaceutical particles. The work focuses on segmenting active pharmaceutical ingredient (API) particles and extracting quantitative shape and texture features such as area, circularity, aspect ratio, contrast, homogeneity, and Haralick features. Different segmentation approaches, including Ilastik, the Segment Anything Model (SAM), and FastSAM, are evaluated and improved through post-processing methods such as feature-based filtering and non-maximum suppression. Principal Component Analysis (PCA) and hierarchical clustering are used to analyze caffeine–mannitol powder and granule samples and to relate image-derived features to material properties such as flowability, wall friction angle, and electrostatic charge. In addition, a Dash-based web application is developed to support interactive image upload, segmentation, feature calculation, filtering, and visualization. The results show that automated SEM image analysis can support material characterization and quality assessment in pharmaceutical development.