RealWorld Stepwise Spectral Unmixing

1
Real-World Stepwise Spectral Unmixing

• Daniel Newland
• Dr. John Schott
• Digital Imaging and Remote Sensing Laboratory
• Center for Imaging Science
• May 7, 1999

2
Outline

• Objective
• Unmixing Background
• Data and Preparation
• Results
• Conclusions

3
Objective

• Use real data to verify a stepwise spectral
  unmixing routine tested only on synthetic data
• Compare to traditional spectral unmixing
• Compare to hierarchical spectral unmixing

4
Outline

• Objective
• Unmixing Background
• Data and Preparation
• Results
• Conclusions

5
Operating Scenario

• Remote sensing by airborne or spaceborne imagers
• Finite flux reaching sensor causes
  spatial-spectral resolution trade-off
• Hyperspectral data has hundreds of bands of
  spectral information
• Spectrum characterization allows subpixel
  analysis and material identification

6
Spectral Mixture Analysis

• Assumes reflectance from each pixel is caused by
  a linear mixture of subpixel materials

Mixed Spectrum Example
7
Mixed Pixels and Material Maps
Input Image
PURE
PURE
PURE
MIXED
8
Traditional Linear Unmixing
i 1 k

• Unconstrained
• Partially Constrained
• Fully Constrained

• Constraint Conditions

9
Hierarchical Linear Unmixing

• Unmixes broad material classes first
• Proceeds to a groups constituents only if the
  unmixed fraction is greater than a given threshold

10
Stepwise Unmixing

• Employs linear unmixing to find fractions
• Uses iterative regressions to accept only the
  endmembers that improve a statistics-based model
• Shown to be superior to classic linear method
• Has better accuracy
• Can handle more endmembers
• Quantitatively tested only on synthetic data

11
Performance Evaluation
Error Metric

• Compare squared error from traditional, stepwise
  and hierarchical methods
• Visually assess fraction maps for accuracy

12
Outline

• Objective
• Unmixing Background
• Data and Preparation
• Results
• Conclusions

13
Data and Preparation

• Used HYDICE collection over the ARM site
• 210 bands around 10nm in width
• Covers wavelengths of 0.4 - 2.5 microns
• Spatial resolution of 1.75 meters per pixel
• Processed original scene to generate unmixing
  input
• Spatial averaging to form mixed pixels
• Spectral subset to remove noise
• Constructed material library and truth map

14
HYDICE Scene
Original 320 x 320
15
Atmospheric Attenuation
16
Atmospheric Effects
Road Pixel
Vegetation Pixel
Band 108 1.4 microns
17
Endmember Selection

• Endmembers are simply material types
• Broad classification road, grass, trees
• Fine classification dry soil, moist soil...
• Used image-derived endmembers to produce spectral
  library
• Average reference spectra from pure sample
  pixels
• Chose 18 distinct endmembers

18
Endmember Listing

• Strong Road
• Weak Road
• Panel 2k
• Panel 3k
• Panel 5k
• Panel 8k
• Panel 14k
• Panel 17k
• Panel 25k

• Spectral Panel
• Parking Lot
• Trees
• Strong Vegetation
• Medium Vegetation
• Weak Vegetation
• Strong Cut Vegetation
• Medium Cut Vegetation
• Weak Cut Vegetation

False-Color IR
19
Materials Hierarchy

• Grouped similar materials into 3-level hierarchy

• Level 1

• Level 2

• Level 3

20
Truth Map Creation

• Realistic classification required automated
  procedure
• Tested classification routines available in ENVI
• Chose Minimum Distance to the Mean classifier

21
Truth Map
False-Color IR
22
Truth Detail
Test Site
Trees
Parking Lot
23
Tools for Analysis

• Data processed with ENVI and IDL
• Three unmixing routines written in IDL
• IDL support programs

24
Outline

• Objective
• Unmixing Background
• Data and Preparation
• Results
• Conclusions

25
Truth Fraction Maps
Labels
Fractions
26
Linear Unmixing
Linear
Truth
27
Hierarchical Unmixing
Hierarchical
Truth
28
Stepwise Unmixing
Stepwise
Truth
29
Fraction Maps
Material Truth Linear
Hierarch. Stepwise
Panel 3k Uncut Mid Vegetation Cut Weak
Vegetation
30
Linear Color Maps
31
Hierarchical Color Maps
32
Stepwise Color Maps
33
Histogram Comparison
Linear Hierarchical
Stepwise
34
Squared Error Results
35
Hierarchical Results
36
Outline

• Objective
• Unmixing Background
• Data and Preparation
• Results
• Conclusions

37
Conclusions

• Linear unmixing does poorly, forcing fractions
  for all materials
• Hierarchical approach performs better but
  requires extensive user involvement
• Stepwise routine succeeds using adaptive
  endmember selection without extra preparation

38
Special Thanks

• Dr. John Schott
• Daisei Konno
• Lee Sanders
• Francois Alain

39
Questions?
Stepwise Unmixed Fraction Maps
False-Color IR