Super-resolution reconstruction of industrial PET images using a prior-knowledge-based generative adversarial network

TL;DR

This paper introduces a new GAN-based method to improve the quality of industrial PET images by incorporating prior knowledge and enhancing texture details.

Contribution

The novel contribution is a prior-knowledge-based GAN with a texture enhancement network and custom loss functions for industrial PET super-resolution.

Findings

The proposed method improves visual and objective quality of super-resolution PET images.

Texture loss and super-resolution loss effectively enhance image details and reduce artifacts.

Abstract

Positron tomography technology (PET) can adapt to complex on-site environments, enabling industrial non-destructive testing without disturbance or damage. PET super-resolution reconstruction aims to reduce detection costs and improve accuracy, making it highly valuable for research. In this study, we propose a generative adversarial network (GAN)-based super-resolution model for industrial PET images that incorporates prior knowledge to address issues such as detail loss and artifact distortion in existing algorithms. We design a texture enhancement network to extract detailed features and employ a connection network to fuse texture and super-resolution features, enhancing texture details. Additionally, we introduce texture loss and super-resolution loss to further improve the model’s performance. Experimental results demonstrate that the proposed method enhances super-resolution image quality in both visual and objective evaluation metrics and has been validated in practical industrial detection.

Figures