Post on 10-Jul-2020
transcript
Multivariate analysis
of hyperspectral images
Geir Rune Flåten
Introducing Unscrambler HSI:
Imaging
What is hyperspectral imaging (HSI)?
Pixels
Wavelengths
K
J
I
From: https://www.osa-opn.org/home/articles/volume_26/october_2015/features/
hyperspectral_imaging_for_safety_and_security/
• Each pixel is represented not by one gray
value, but by a vector of gray values: spectra
• A stack of gray level images, one for each
variable (wavelength): data cube
Spectroscopy
+
Unscrambler HSI
• Software for Multivariate Data Analysis (MVA) of
Hyperspectral Images
• First version: Tailored for classification applications:
− Explorative analysis of HSI data in the spectral domain
− Calibration model development
• Much more output produced from a single model
compared to generic image analysis tools
Additional output compared to
Generic Image Analysis Tools
• Interactive & informative plots: model interpretation and
mapping back to image data
• Spectral transformations: separate chemical absorbance
from physical effects
• Flexible model validation options
• Variable selection: Significance tests and
remove/downweight irrelevant spectral regions
• Statistical tests for detection of outliers
History:
Origins in remote sensing
• 1972: Launch of LANDSAT-1 satellite
• Objective: study/monitor Earth’s
landmasses
• 2 instruments:
Camera system
Multispectral scanner (MSS)
−4 spectral bands:
green, red, 2 infrared
From: https://landsat.gsfc.nasa.gov/landsat-1/
History:
Development driven by Earth remote sensing
1994
Late
1980s
1974
First portable field
reflectance
spectrometer (PFRS,
0.4-2.5 µm)
Several
commercial
hyperspectral
imagers in the
market
First spectral
cube &
Spectral Image
Processing
System (SIPS)
developed:
software
supported in all
computing
platforms
Early
1990s
SIPS evolves to ENVI
(Environment for
Visualizing Images)
(beginner-friendly)
Wider research
community emerges,
focus on terrestrial
surface mapping
Development of field
spectrometer
applicable to many
research fields MacDonald et al. Remote Sens. Environ. 113 (2009) S2–S4
History: Factors behind HSI growth
• Technological advancement in instrumentation: new sensor
types & sensor arrays
• Improved data storage capabilities
• Higher speed for software handling large data files
HSI Applications
• Precision agriculture: Robot or drone diagnosing
individual plants for stress induced by disease, insects,
water or nutrient deficiency, etc.
• Food: Characterization of salmon filets for grade sorting
Aspirin
Paracetamol
From: He et al. Innov. Food Sci.
Emrg. Technol. 18 (2013)
237-245
• Pharma: In-line monitoring of active ingredient in tablets
at production line From: J. Colling, Stellenbosch
University, South Africa
HSI Applications
• Recycling: Sorting of plastic waste for efficient waste
management
From: Grahn et al. Encyclopedia of
Analytical Chemistry. John
Wiley & Sons.
• Energy: Pipeline monitoring for condition assessment and
damage or leak detection
HSI Applications
• Manufacturing: Real time detection of defects in silicon
wafers and solar cells
From: Turek et al. Energy
Procedia. 92 (2016)
232-235
Defect 1
Defect 2
• Environmental: Remote sensing for monitoring
lake water quality
From: https://aviris.jpl.nasa.gov
/data/free_data
Demo
text
Demo Dataset
• text
From: J. Colling, Stellenbosch
University, South Africa
Thank you!
More info:
https://www.camo.com/unscramblerhsi/
Product Manager Lars Gidskehaug
lg@camo.no
Marketing Manager Morten Hansen
Morten.Hansen@camo.no
Geir Rune Flåten
grf@camo.no