Parameter-Free Attention Super-Resolution Network for Accelerated AFM Biological Imaging

Atomic force microscopy (AFM) is a cornerstone for topographic imaging of biological specimens. However, the conventional reliance on raster scanning significantly constrains the acquisition speed of high-resolution images. To overcome this limitation, we propose the parameter-free channel-spatial attention network (PCSAN), a super-resolution (SR) model designed to accelerate AFM imaging. A paired dataset was constructed by applying controlled degradation to preacquired high-resolution (HR) virus images to generate corresponding low-resolution (LR) counterparts, which subsequently guided the supervised learning of the PCSAN model. During the inference phase, high-fidelity HR reconstructions are achieved by directly feeding newly acquired LR virus images into the trained network. Notably, the PCSAN architecture leverages a parameter-free channel-spatial attention mechanism that accentuates informative features and suppresses redundant noise without increasing the parameter count. This streamlined design significantly reduces inference latency, enabling the rapid reconstruction of HR viral topographies. Validation on SARS-CoV-2 and influenza AFM datasets demonstrates that PCSAN achieves a peak signal-to-noise ratio (PSNR) of 38.73 dB at a ×2 scaling factor, while reducing imaging time by over 50%. Our findings indicate that the proposed PCSAN model effectively preserves image integrity while substantially enhancing throughput, thereby providing a robust technical foundation for high-throughput AFM characterization.