A terahertz time-domain super-resolution imaging method using a local-pixel graph neural network for biological products

Tong Lei,Brian Tobin,Zihan Liu,Shu-Yi Yang,Da-Wen Sun

doi:10.1016/j.aca.2021.338898

Tong Lei, Brian Tobin + Show 3 more

Open Access

https://doi.org/10.1016/j.aca.2021.338898

Copy DOI

Abstract

The low image acquisition speed of terahertz (THz) time-domain imaging systems limits their application in biological products analysis. In the current study, a local pixel graph neural network was built for THz time-domain imaging super-resolution. The method could be applied to the analysis of any heterogeneous biological products as it only required a small number of sample images for training and particularly it focused on THz feature frequencies. The graph network applied the Fourier transform to graphs extracted from low-resolution (LR) images bringing an invariance of rotation and flip for local pixels, and the network then learnt the relationship between the state of graphs and the corresponding pixels to be reconstructed. With wood cores and seeds as examples, the images of these samples were captured by a THz time-domain imaging system for training and analysed by the method, achieving the root mean square error (RMSE) of pixels of 0.0957 and 0.1061 for the wood core and seed images, respectively. In addition, the reconstructed high-resolution (HR) images, LR images and true HR images at several feature frequencies were also compared in the current study. Results indicated that the method could not only reconstruct the spatial details and the useful signals from high noise signals at high feature frequencies but could also operate super-resolution in both spatial and spectral aspects.

Highlights

Of physical and chemical information about objects [1]
The research on graph embedding is closely related to graph neural network (GNN), which aims at representing nodes as vector representations, preserving both network topology structure and node content information, the analysis of graph data can be performed by neural networks [28e30]
To evaluate the performance of this super-resolution method, the reconstructed images were compared with corresponding true HR images, and the root mean square error (RMSE) of pixels was used for quantitative evaluation

Summary

Introduction

Of physical and chemical information about objects [1]. It has attracted wide attention on biomedicine [2], agri-food [3], and material recognition [4] based on both molecular and crystal structure fingerprinting. Not like visible or near-infrared hyperspectral imaging systems which have been widely used in food research [9e15], the image acquisition speed of existing THz time-domain imaging systems cannot meet the needs of large-scale food and agricultural products analysis because the THz time-domain imaging system only adopts the pixel scanning mode For this reason, developing THz time-domain imaging spatial super-resolution techniques for biological product studies has great benefits. As the THz spectral information between pixels of an image can overlap with each other due to the diffraction limitation, and each pixel on the THz image of heterogeneous biological products has a unique PSF, an approach to treat the local pixels as graphs on non-Euclidian spaces was adopted This approach embedded the graph for local pixels on a neural network and reconstructed the pixels on highresolution images through twice modelling. To evaluate the performance of this super-resolution method, the reconstructed images were compared with corresponding true HR images, and the root mean square error (RMSE) of pixels was used for quantitative evaluation

Methods

Results

Conclusion