A knowledge-guided, drift-aware computer vision pipeline for automated nanoscale atomic force microscope surface...

This study introduces a two-stage (Stage 1-Stage 2), knowledge-guided, and drift-aware computer vision (CV) pipeline that redefines the traditionally manual and observer-dependent analysis workflow of AFM measurements. The proposed approach establishes, for the first time in the literature, a nanoscale point-tracking method capable of automatically, robustly, and reliably following the same surface locations across consecutive loading conditions. Stage 1 ensures the physical consistency of raw AFM data, extracts the multilayer morphology of the surface, and generates reliable point correspondences between consecutive load steps using SIFT-ORB feature extraction combined with RANSAC-based geometric verification. Stage 2 converts these correspondences into a drift-aware chained tracking structure that preserves the identity of each point and enables pixel-level computation of tribological responses, including local friction forces and load-friction curves. The method is validated on a representative two-dimensional material system, demonstrating its capability to both characterize lateral drift emerging in consecutive scans and reliably disentangle the layer-dependent tribological responses of the surface. By processing AFM data through a holistic sensing-to-inference information flow, the pipeline introduces the first drift-aware and knowledge-guided CV-based nanoscale characterization approach that produces machine-readable, reproducible, and statistically consistent tribological outputs.