Abstract
Lens-coupled high-resolution micro-CT is capable of achieving a resolution which is superior to that of conventional systems through the utilization of visible-light magnification. However, the spatial resolution is still fundamentally constrained by the optical diffraction limit (200 to 300nm) and mechanical precision. These limitations impede its application in nanoscale biological and material structure imaging. To overcome these limitations, a super-resolution reconstruction neural network based on implicit neural representation (INR) and plane attention (PA) is proposed. A high-precision nano-positioning stage is integrated into the rotation platform, enabling sub-pixel dithering scans to acquire multiple projection sequences containing sub-pixel information. The proposed approach implements a multi-scale hash encoding function and introduces a plane attention mechanism to aggregate spatial correlations among fan-beam sampling points. The experimental results demonstrate that the proposed method effectively surpasses the theoretical resolution boundaries of the system, thereby facilitating high-precision detail representation.