Prompt-guided zero-shot Depth-aware transformer pipeline for democratized SEM particle characterization: A case study in battery materials

• Zero-shot transformer pipeline with interactive expert correction without domain-specific training resolves primary particles missed by laser diffraction. • Transformer-based single-image contour roughness achieves r = 0.983 against multi-image tilt-series with 96.3% concordant pairs. • Depth-derived surface-area-to-noise classifier separates coated from uncoated nanoparticles with zero overlap. Microscopy-based particle characterization tools fail on complex images, produce fixed uncorrectable outputs, and lack integrated size, shape, and surface analysis; comprehensive solutions remain commercial or instrument-dependent, with no open-source alternative. This work presents IPGSS (Interactive Prompt-Guided Segmentation System) and DRCS (Depth-based Roughness Characterization System), a training-free pipeline built on open-source vision models for comprehensive particle characterization from standard microscopy images without specialized instrumentation. IPGSS combines zero-shot segmentation, hierarchical post-processing, and interactive correction; DRCS extracts ISO-compliant roughness from a single SEM image. Certified 390 µm polystyrene particles validated accuracy at equivalent circular diameter D 50 = 388.8 µm (−0.3%). On complex aluminum oxide images, number-weighted D 50 = 49.9 µm (−0.2%) and volume-weighted D 90 matched QICPIC within 0.4–4% across >9,000 particles; IPGSS outperformed OpenCV and Cellpose. DRCS replaces the tilt-series approach of Hülagü et al. (2024) with two tracks from a single zero-tilt SEM image. The contour track achieved Spearman r = 0.983 (RMSE = 1.48 nm for near-spherical particles; 96.3% concordant pairs); the depth track provided 3D descriptors including a coating classifier that separated coated from uncoated nanoparticles with zero overlap. A ∼20 nm noise floor caused roughness overestimation, though between-type ordering was preserved. IPGSS resolved primary particles (D 50 = 0.54 µm) within solid-state battery agglomerates measured by laser diffraction at 9.47 µm, a 17.5-fold discrepancy critical for electrode design.