HyperGUI: A Web Application for Hyperspectral Image Analysis and Data Extraction

Introduction

There are numerous platforms available for hyperspectral image processing, but these platforms are either proprietary and closed source such as ENVI [1] or when open-source and free, are only accessible within programming languages, such as Spectral Python [2] a package for handling hyperspectral data in Python and, hyperSpec [3], terra [4] and hsdar [5], three hyperspectral image analysis packages for R programming language. This can often present as a barrier to new users of hyperspectral data.

Web apps are quickly gaining popularity with scientists for their ability to create data analysis software that is robust to write and run, open-sourced, high-throughput and with strong visualization abilities [6, 7, 8]. ‘Shiny’ [9] is an object oriented and reactive web application development package within R programming language, that enables scientists to develop interactive graphical user interface (GUI) based applications for data analysis purposes.

While there are similar GUI apps such as MIC-Surgery-Heidelberg/HyperGui for medial hyperspectral imaging, these are highly specialized for a specific task within a certain project, not well documented or published and are not friendly to general users. Here we introduce a webapp for general hyperspectral imaging preprocessing and data conversion which is accessible and friendly to a wide variety of users.

Here a web app named HyperGUI was developed based on Shiny, to handle loading, processing, and visualizing of hyperspectral images with ease. The objectives were to develop a software with functions to (1) load benchtop or aerial hyperspectral images, (2) calculate the surface reflectance based on dark and white current in the case of benchtop samples and based on ground spectral reference samples in the case of aerial images, (3) calculate PCA for a desired spectrum and save the reduced data based on percentage of variance explained and (4) visualize the hyperspectral image and the spectrum of ROIs for informative and sample selection purposes.

Implementation and architecture

HyperGUI is based on R’s web application development framework package, Shiny [9]. The overall architecture of the web app is divided into a server and a user interface (UI), with the server handling the file loading/saving and calculations (Figure 1), and the UI presenting interactive graphical input/output. The source code is available at https://github.com/shaikabiri/hyperGUI. The application can be run either on a localhost via RStudio [10] or online via shinyapps.io at https://urye8b-shayan-kabiri.shinyapps.io/hyperGUI/.

Figure 1

To run the application on the localhost the following procedure should be followed. The below codes can be run in either R or RStudio. The video tutorials for the installation and the application usage can be found at https://youtube.com/@hyperGUI.

#install the devtools package first
install.packages(‘devtools’)
 
#install the package using devtools:
devtools::install_github(‘https://github.com/shaikabiri/hyperGUI’)
 
#load the package
library(hyperGUI)
 
#run the app by the following command
hyperGUI::run_app()

The software’s GUI is divided into a sidebar and a display area. The display area presents the hyperspectral image in form of a monochromatic image of a wavelength of a choice, pseudo-RGB image of three wavelengths of choice or a pseudo-RGB image of three most important components of a PCA. Moreover, the displayed image can be interacted with to crop the data or display spectra for regions of interest or spectra of interest in a plot below the displayed image. The sidebar controls the type of hyperspectral image, either benchtop or aerial, file load, data save and changes to parameters of the displayed image including wavelengths of interest, and mode of display (Figure 2).

Figure 2

Principal component analysis

Principal Component Analysis (PCA) can be calculated and visualized as a pseudo-RGB image for selected spectrum, and importance of each component can be displayed separately. Reduced data can be saved by selecting a proportion of variance that can be explained by Principal Components (Figure 3). PCA can be utilised here for data compression [13] or informing ROI for reference sampling.

Figure 3

Quality control

Two hyperspectral image samples from Eartha Science applications were used to test the software, an aerial image from a published study [14], and a benchtop image from the Geological Survey Ireland (GSI) hyperspectral imaging data project [15] hereafter called sample A and sample B, respectively. Sample A is an aerially acquired visible to short wave infrared (SWIR) image of Gudalur, India, with a spectral range from 376 to 2500 nm, acquired by the AVIRIS-NG sensor. Ground references were taken using a field spectrometer (Spectra Vista Corporation, HR-1024i, USA) with a spectral range of 350 to 2500 nm. Sample B is a single core box, consisting of SWIR image of 105.98 to 110.35 m section of a sediment core from drillhole GSI-17-007 of Geological Survey Ireland’s hyperspectral project, with spectral range of 980 to 2580 nm. Software was tested on a laptop PC with AMD Ryzen 7 4800H CPU and 12 GB of RAM on Windows 11 and web browser Chrome Version 124.0.6367.202 (Official Build) (64-bit).

Overall, the web app shows acceptable performance, with no process taking more than a few seconds except for linear calibration and PCA which took approximately 3 minutes to finish processing. The heaviest task was taking a PCA on uncropped sample B, which consumed approximately 4GB of memory and 27% of CPU power, other tasks were light in memory and CPU consumption. Functionality of all features were tested successfully. Figure 4 and Figure 5 show the results of dark and white current correction and linear calibration for Sample A and B. The linear calibration for sample A closely matches with calibrated spectra published by Jha and Nidamanuri [16], even though a more sophisticated sensor-specific method was used in that study. The results of dark and white current for sample B matches the results published by GSI [17] (URL: https://gsi.geodata.gov.ie/downloads/Bedrock/Data/Hyperspectral_Sample_data.zip). Figure 6 shows PCA results for sample A and B. Saved data frames were verified in MATLAB 2021b, Python 3.10, R 4.2.2 and QGIS 3.22.1.

Figure 4

Figure 5

Figure 6

Operating system

Windows, Linux and Mac

Programming language

R version 4.3.3

Additional system requirements

Minimum ram 4GB

Dependencies

RStudio 2023.12.1

Software location

Archive (e.g. institutional repository, general repository) (required – please see instructions on journal website for depositing archive copy of software in a suitable repository)

Code repository (e.g. SourceForge, GitHub etc.) (required)

Language

English