Digital Imaging and Remote Sensing Laboratory Real-World Stepwise Spectral Unmixing Daniel Newland Dr. John Schott Digital Imaging and Remote Sensing Laboratory.

Digital Imaging and Remote Sensing Laboratory

Real-World Stepwise Real-World Stepwise Spectral UnmixingSpectral Unmixing

Real-World Stepwise Real-World Stepwise Spectral UnmixingSpectral Unmixing

Daniel NewlandDr. John Schott

Digital Imaging and Remote Sensing LaboratoryCenter for Imaging Science

May 7, 1999


OutlineOutlineOutlineOutline

• Objective

• Unmixing Background

• Data and Preparation

• Results

• Conclusions


ObjectiveObjectiveObjectiveObjective

• 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

Hyperspectral Data Material Fraction Maps



• Objective



• Results

• Conclusions


Operating ScenarioOperating ScenarioOperating ScenarioOperating 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


Spectral Mixture AnalysisSpectral Mixture AnalysisSpectral Mixture AnalysisSpectral Mixture Analysis

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

Mixed Spectra Example

0

2000

4000

6000

8000

10000

12000

14000

0.4 0.9 1.4 1.9 2.4

Wavelength (microns)

Dig

ital

Co

un

t

Parking Lot

Vegetation

1:1 Mixture

Mixed Spectrum Example


Mixed Pixels and Material MapsMixed Pixels and Material MapsMixed Pixels and Material MapsMixed Pixels and Material Maps

Input Image1.0

1.0

0.0

0.5

1.00.0

0.0 0.5

Red Fraction Map

Green Fraction Map

UNMIXED

UNMIXEDPUREPURE

PURE MIXED


i

N

eeeii fRR

1,

i = 1 … k

Constraint Conditions

• Unconstrained:

• Partially Constrained:

• Fully Constrained:

feendmembers 10.

0 0 10. . f e

ef

Traditional Linear UnmixingTraditional Linear UnmixingTraditional Linear UnmixingTraditional Linear Unmixing


• Unmixes broad material classes first

• Proceeds to a group’s constituents only if the unmixed fraction is greater than a given threshold

Hierarchical Linear UnmixingHierarchical Linear UnmixingHierarchical Linear UnmixingHierarchical Linear Unmixing

C on c re te M eta l

M an -M ad e W ater

D ec id u ou s C on ife rou s

Trees G rass

V eg eta tion

M ixed P ixe l

Example Materials Hierarchy

Full Library

• Concrete• Metal• Water• Deciduous Trees• Coniferous Trees• Grass


Stepwise UnmixingStepwise UnmixingStepwise UnmixingStepwise 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


Performance EvaluationPerformance EvaluationPerformance EvaluationPerformance Evaluation

• Compare squared error from traditional, stepwise and hierarchical methods

• Visually assess fraction maps for accuracy

Error Metric: 2)(N

1 = SE test

pixels materialstruth ff



• Objective



• Results

• Conclusions


Data and PreparationData and PreparationData and PreparationData 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


HYDICE SceneHYDICE SceneHYDICE SceneHYDICE Scene

Original 320 x 320

Convolved80 x 80


Atmospheric AttenuationAtmospheric AttenuationAtmospheric AttenuationAtmospheric Attenuation


Atmospheric EffectsAtmospheric EffectsAtmospheric EffectsAtmospheric Effects

Band 108 1.4 microns

Road Pixel

Vegetation Pixel


Endmember SelectionEndmember SelectionEndmember SelectionEndmember 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


Endmember ListingEndmember ListingEndmember ListingEndmember 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


Materials HierarchyMaterials HierarchyMaterials HierarchyMaterials Hierarchy

• Grouped similar materials into 3-level hierarchy

– Level 1

– Level 2

– Level 3


Truth Map CreationTruth Map CreationTruth Map CreationTruth Map Creation

• Realistic classification required automated procedure

• Tested classification routines available in ENVI

• Chose Minimum Distance to the Mean classifier


Truth MapTruth MapTruth MapTruth Map

False-Color IR Classified Scene


Truth DetailTruth DetailTruth DetailTruth Detail

Test Site

Trees

Parking Lot


Tools for AnalysisTools for AnalysisTools for AnalysisTools for Analysis

• Data processed with ENVI and IDL

• Three unmixing routines written in IDL

• IDL support programs



• Objective



• Results

• Conclusions


Truth Fraction MapsTruth Fraction MapsTruth Fraction MapsTruth Fraction Maps

Lab

els

Fra

ctio

ns


Linear UnmixingLinear UnmixingLinear UnmixingLinear Unmixing

Lin

ear

Tru

th


Hierarchical UnmixingHierarchical UnmixingHierarchical UnmixingHierarchical Unmixing

Hie

rarc

hic

alT

ruth


Stepwise UnmixingStepwise UnmixingStepwise UnmixingStepwise Unmixing

Ste

pw

ise

Tru

th


Fraction MapsFraction MapsFraction MapsFraction Maps

Material Truth Linear Hierarch. Stepwise

Panel 3k

Uncut Mid

Vegetation

Cut Weak

Vegetation


Linear Color MapsLinear Color MapsLinear Color MapsLinear Color Maps


Hierarchical Color MapsHierarchical Color MapsHierarchical Color MapsHierarchical Color Maps


Stepwise Color MapsStepwise Color MapsStepwise Color MapsStepwise Color Maps


Histogram ComparisonHistogram ComparisonHistogram ComparisonHistogram Comparison

Linear Hierarchical Stepwise


Squared Error ResultsSquared Error ResultsSquared Error ResultsSquared Error Results


Hierarchical ResultsHierarchical ResultsHierarchical ResultsHierarchical Results



• Objective



• Results

• Conclusions


ConclusionsConclusionsConclusionsConclusions

• 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


Special ThanksSpecial ThanksSpecial ThanksSpecial Thanks

Dr. John Schott

Daisei Konno

Lee Sanders

Francois Alain


Questions?Questions?Questions?Questions?

False-Color IRStepwise Unmixed Fraction Maps

Digital Imaging and Remote Sensing Laboratory Real-World Stepwise Spectral Unmixing Daniel Newland Dr. John Schott Digital Imaging and Remote Sensing Laboratory.

Documents

remote sensing laboratory

remote sensing laboratory

imaging science

john schott digital

synthetic data slide

traditional spectral

truth map slide

real data