AI-driven artifact removal and image restoration in Atomic Force Microscopy via structured mask-guided inpainting

High-resolution Atomic Force Microscopy (AFM) serves as a foundational smart sensing technology for nanoscale characterization, yet raw topographical data is heavily degraded by spatial distortions and scanning artifacts. Standard image correction workflows depend on manual intervention or generic algorithms that ignore the highly directional, physics-based continuity constraints of AFM height maps. To enable smart, high-fidelity data interpretation, this work introduces an AI-driven, multi-stage image restoration framework. The system employs a lightweight convolutional neural network to automatically classify scan quality and isolate defect-contaminated images. For degraded scans, an intelligent topography-aware Smart Flatten module that combines adaptive baseline estimation with structural exclusion is developed to suppress macroscopic background distortions. Localized scan artifacts are subsequently identified using structural orientation metrics and repaired via directional restoration, enforcing strict scan-line consistency and spatial continuity. Validated across diverse AFM height scan datasets, the framework effectively eliminates instrumentation artifacts while preserving genuine nanoscale topography. This unified, AI-driven pipeline advances high-throughput microscopy by enabling automated quality classification and data enhancement for autonomous topography characterization and intelligent materials discovery.