Transformer-Based Super-Resolution Scanning Acoustic Imaging for Industrial Inspection

This paper presents SAMTransGAN, a novel super-resolution framework combining Generative Adversarial Networks with Transformer-based architectures for high-resolution acoustic imaging in industrial non-destructive testing.

Paper

Abstract

Despite significant advancements in super-resolution (SR) techniques, their application in industrial nondestructive testing (NDT) remains challenging due to stringent accuracy requirements. Among NDT technologies, scanning acoustic microscopy (SAM) is widely used in semiconductor industries for detecting defects. We introduce a method for generating synthetic low-resolution datasets from high-resolution images using a high-order degradation model. Our model, SAMTransGAN, incorporates Transformers as the generator network within a GAN framework, using transfer learning and a dynamic loss function balancing pixel-wise, perceptual, and adversarial losses. SAMTransGAN achieves PSNR of 28.158 dB and SSIM of 0.835 on the testing dataset, surpassing all compared models. We also present a web-based deployment framework for efficient industrial NDT applications.

Key Contributions

• SAMTransGAN: hybrid GAN + Transformer architecture for acoustic image super-resolution
• Synthetic low-resolution dataset generation via high-order degradation model
• Transfer learning and dynamic loss function for optimal SR performance
• Web-based deployment framework for practical industrial NDT use

Recommended citation: T. N. Nguyen, V. H. M. Doan, T. H. Vo, J. Choi and J. Oh, "Transformer-Based Super-Resolution Scanning Acoustic Imaging for Industrial Inspection," in *2025 IEEE International Conference on Industrial Technology (ICIT)*, Wuhan, China, 2025, pp. 1-7, doi: 10.1109/ICIT63637.2025.10965207.