Neural Field-Based 3D Surface Reconstruction of Microstructures from Multi-Detector Signals in Scanning Electron Microscopy

Abstract

The 3D characterization of microstructures is crucial for understanding and designing functional materials. However, the scanning electron microscope (SEM), widely used in scientific research, captures only 2D electron intensity distributions. Existing SEM 3D reconstruction methods struggle with textureless regions, shadowing artifacts, and calibration dependencies, whereas advanced learning-based approaches fail to generalize to microscopic SEM domains due to the lack of physical priors and domain-specific data. We introduce NFH-SEM, a neural field-based hybrid framework that reconstructs high-fidelity 3D surfaces from multi-view, multi-detector SEM images. NFH-SEM integrates coarse multi-view geometry with photometric stereo cues from detector signals through a continuous neural field, incorporating a learnable forward model that embeds SEM imaging physics for self-calibrated, shadow-robust reconstruction. NFH-SEM achieves precise recovery across diverse specimens, revealing 478 nm layered features in two-photon lithography samples, 782 nm surface textures on pollen grains, and 1.559 $μ$m fracture steps on silicon carbide particles, demonstrating its accuracy and broad applicability. Our code and real-world dataset are available at https://github.com/zju3dv/NFH-SEM.

Figures (14)

• Figure 1: NFH-SEM reconstructs detailed microstructures with important functional roles across diverse materials, from pollen textures that enable attachment to pollinators to fracture patterns associated with crack propagation in solids, supporting broad scientific research.
• Figure 2: Workflow of NFH-SEM.(a) Multi-view and multi-detector SEM scanning of a sample mounted on a motorized stage. (b) 4Q-BSE images provide directional illumination with shadows caused by geometric occlusion. (c) Multi-view reconstruction initialization. (d) The posed depth maps and 4Q-BSE images jointly supervise an SDF-based neural field. A learnable BSE forward model is self-calibrated during training. (e) The reconstructed surface extracted from the neural field exhibits high geometric fidelity and rich surface details.
• Figure 3: Emission of the BSE signal from the sample surface and its detection by the 4Q-BSE detector.
• Figure 4: Qualitative comparison with conventional SEM 3D reconstruction methods on real-world dataset. Red boxes highlight enlarged normal maps. Red lines in SE images indicate the cross-sectional profiles, where the zoomed-in segments are marked between vertical gray lines. NFH-SEM faithfully reconstructs the overall geometry and fine surface details, consistent with SEM observations.
• Figure 5: Qualitative comparison with learning-based 3D reconstruction methods on real-world dataset. NFH-SEM achieves more accurate reconstructions than approaches that neglect the SEM signal generation model or lack generalization to SEM domains.