HDL-SAM: A Hybrid Deep Learning Framework for High-Resolution Imaging in Scanning Acoustic Microscopy
Keywords: super-resolution, image inpainting, acoustic microscopy
TL;DR: HDL-SAM is a hybrid deep learning model that significantly improves the resolution of scanning acoustic microscopy images by a factor of four, making it an effective tool in biomedical and materials research.
Abstract: Scanning acoustic microscopy (SAM) is a cutting-edge label-free imaging technique that allows viewing of both surface and internal structures in a variety of samples, including industrial and biological. Several factors influence the acoustic image resolution, including the signal-to-noise ratio, scanning step size, and the transducer frequency. Our proposed network involves a combination of SwinIR and the hypergraph image inpainting technique specifically adapted to improve the resolution for SAM images. The method aims to fill in missing information, significantly enhancing the resolution of the acquired images. We assessed the effectiveness of our approach against the standalone application of hypergraphs and SwinIR on the dataset, targeting a notable fourfold increase in resolution. The results indicate that the proposed method achieves superior performance, marked by an average structural similarity index measure (SSIM) of $0.92$, a peak signal-to-noise ratio (PSNR) of ${31.60}$, and a ${4 \times}$ enhancement in image resolution over the raw SAM image. This integration of SwinIR and hypergraphs modules proves indispensable for the precise interpretation of low-resolution acoustic imaging data, allowing the development of reliable tools for image restoration. This improves the fidelity and quality of SAM imaging for both research and practical use.
Submission Number: 53
Loading