Introduction to Hyperspectral/Multivariate Image Analysis

Course Description

Hyperspectral images are becoming increasingly common in analytical chemistry and remote sensing applications. They are based on several types of spectroscopy and spectrometry including Raman, Infrared, and Secondary Ion Mass Spectrometry (SIMS), Excitation Emission Fluorescence (EEM), etc. Introduction to Hyperspectral/Multivariate Image Analysis (MIA) shows how to apply multivariate and machine learning methods to these data cubes to extract maximum information. This hands-on course considers methods for visualization, pattern recognition, classification, curve resolution, chemical mapping, regression and analysis of particles and textures.

Introduction to Hyperspectral/Multivariate Image Analysis (MIA) starts with a brief review of sources of multivariate images and tools for viewing and investigating them. Practical image analysis with Principal Components Analysis (PCA) demonstrates how information from hyperspectral images can be compressed and displayed, and how classification tools can be used to identify chemically similar areas. Multivariate Curve Resolution (MCR) on images is presented and it is demonstrated how it can be used to create chemical maps. The course finishes with a review of particle analysis and texture analysis. For a complete list of topics please see the Course Outline below. The course includes hands-on computer time for participants to work example problems using PLS_Toolbox and MIA_Toolbox, or Solo+MIA.

Prerequisites

Linear Algebra for Chemometricians, and Chemometrics I–PCA, or equivalent experience.

Course Outline

1. Intro to 3-way Arrays
   1. Objects and variables
   2. Sources of data
   3. Structure of multivariate images
   4. Comparison to other sources of 3-way data 
2. Simple image analysis tools
   1. Selecting channels for RGB
   2. Viewing images at specific peaks
   3. Integrating peaks for noise reduction
3. Practical Multivariate Image Analysis (MIA)
   1. Review of Principal Components Analysis (PCA)
   2. Scores, loadings and projections
   3. Unusual samples, residuals and T^2
   4. Matricizing of images
   5. Scores images, loadings
   6. Score/score plots: density
   7. Links between scores space and the image plane
   8. Contrast enhancement, color maps
4. Classification on images:
   1. SIMCA
   2. Partial Least Squares Discriminant Analysis (PLS-DA)
   1. Other methods, e.g. Maximal Autocorrelation Factors (MAF)
5. Deconvolution of hyperspectral images
   1. Multivariate Curve Resolution (MCR) on Images
   2. Clustering methods
6. Chemical maps and multivariate image regression
   1. Classical Least Squares (CLS, including from MCR)
   2. Regression on images
   3. Partial Least Squares (PLS) and Artificial Neural Networks (ANNs)
7. Volumetric data and visualization
   1. Example: SIMS depth profiling
8. Variance filtering for images
   1. Maximum Autocorrelation Factors (MAF)
   2. Maximum Difference Factors (MDF)
   3. Generalized Least Squares weighting (GLS)
   4. External Parameter Orthogonalization (EPO)
9. Particle analysis
   1. Finding particles in images
   2. Calculating particle statistics
   3. Creating multivariate models on particles
10. Texture Analysis
    1. Definition of texture
    2. Methods for creating texture spectra from images
    3. Using texture spectra in regression and classification
11. Summary