Deep generative neural networks for spectral image processing

Puneet Mishra

doi:10.1016/j.aca.2021.339308

Abstract

An artificial intelligence approach based on deep generative neural networks for spectral imaging processing was proposed. The key idea was to treat different spectral image processing operations such as segmentation, regression, and classification as image-to-image translation tasks. For the image-to-image translation, the conditional generative adversarial networks were used. As a baseline comparison, the traditional chemometric approach based on pixels wise modelling was demonstrated. The analysis was presented with two real data sets related to fruit property prediction and kernel and shell classification of walnuts. The presented artificial intelligence approach for spectral image processing can provide benefits for any field of science where spectral imaging and processing is widely performed.

Highlights

Spectral imaging is a bi-modality analytical technique where, on one mode, the imaging captures the spatial information about the samples and, on another mode, the spectroscopy captures spectral characteristics of samples under study [1,2]
This study presented an innovative approach to model spectral imaging data using an artificial intelligence approach based on conditional generative adversarial networks
This novel approach treated the spectral images as images and used both the spatial and spectral information with convolutional operations to perform three main tasks for spectral image processing i.e., segmentation, regression, and classification

Summary

Introduction

Spectral imaging is a bi-modality analytical technique where, on one mode, the imaging captures the spatial information about the samples and, on another mode, the spectroscopy captures spectral characteristics of samples under study [1,2]. Applications of spectral imaging can be found ranging from agriculture [6,7] and foods [1] to high-end pharmaceuticals [8]. Based on the experimental needs, spectral imaging can be explored for a range of the electromagnetic spectrum such as ultraviolet [9], visible and near-infrared [10,11], mid-infrared [12], terahertz [13] etc. Several applications of Raman spectral imaging [14] can be identified

Methods

Results

Discussion

Conclusion