Case Study

Document Sample
Case Study Powered By Docstoc
					                     Case Study

            Use of Hyperspectral
          Environmental Monitoring
           Waste Disposal Areas

After Jason Hamel: Chester F. Carlson Center for Imaging Science
(CIS) at Rochester Institute of Technology
           Case Study
What are Landfills?
 Verycommon waste management

 Toxicwastes are not separated from
 the environment

A water resistant clay cap is placed
 over the landfill to slow the spread of
            Case Study
Clay Caps
           Case Study
Clay Cap Technology
 Caps   are designed to last 40 years

 Need to be replaced with new
 technology that actually deals with
 the waste

 Meanwhile,  chemicals have time to
 leach into the environment
          Case Study
Why Look at Landfills?
   Currently, possible dangerous sites are
    manually sampled and processed in a lab

   This can be time and money consuming for
    larger sites or a large number of sites

   Chemicals are often dangerous even at very
    low concentrations

   Remote sensing with new hyperspectral
    detectors may provide and economic
          Case Study
Example of Expected Imagery

                   AVIRIS scene with
                   224 bands

                   DOE Savannah
                   River Site (SRS)
                   Case Study
Research Objective
   Low concentrations make it very difficult to directly
    detect a chemical’s spectral signature

   Determine if new hyperspectral sensors collect
    enough information to identify materials

   Determine the detectability of specific secondary
    spectral effects of leachates (e.g.):
        Vegetation health
        Soil water moisture

   Determine if atmospheric correction is necessary
           Case Study
Vegetation Spectra

 Two   varied inputs:
     Chlorophyllconcentration (mm/cm2)
     Equivalent water thickness (cm)

 Generated     spectra
     Healthyleaf (high chlorophyll and water)
     Stressed leaf (low chlorophyll and water)
           Case Study
Vegetation Spectra

                        Spectra of

                        Green : Healthy
                        Red : Stressed
                  Case Study
Soil Spectra
   Ground measurements were taken with
    spectrometer as soil dried

   Moisture in soil was not measured while
    spectra was taken

   Relative labels given to various soil spectra
        Wet Soil
        Moist Soil
        Dry Soil
            Case Study
Soil Spectra

                         Spectra of

                         Brown : Dry
                         Orange : Moist
                         Black : Wet
           Case Study
Reflectance Data Set
    The 5 basic vegetation and soil spectra
     are mixed by:

     Rmixed = 50%R1 + 50%R2    where R1 and R2
                               are 2 basic spectra

    This creates 10 additional mixed spectra

    15 spectra in final data set
         Case Study
Atmosphere and Detector Effects
   Light reflecting off material propagates
    through atmosphere

   Detector measures the radiance reaching
    the detector at various narrow
    wavelength regions called channels

   Detector electronics record input signal in
    digital counts (DC)
          Case Study
AVIRIS Basic DC Spectra
                 Case Study
Realistic Data Set
   All detectors measure noise as well as signal

   Standard gaussian noise with standard
    deviation of 1 added to DC spectra (not
    representative AVIRIS noise value)

   Noisy sensor radiance determined

   Noisy reflectance spectra calculated by
    removing atmosphere effects
           Case Study
Noisy Basic Reflectance Spectra
                  Case Study
   6 classification algorithms used:
         Linear Spectral Unmixing (ENVI)
         Orthogonal Subspace Projection (Coded)
         Spectral Angle Mapper (ENVI)
         Minimum Distance (ENVI)
         Binary Encoding (ENVI)
         Spectral Signature Matching (Coded)

   The 5 basic vegetation and soil spectra were
    used as endmembers

   Reflectance endmembers converted to DC
    before classifying DC spectra
                     Case Study
Classification Algorithms
   Linear Spectral Unmixing (LSU)
         Generates maps of the fraction of each endmember in
           a pixel
   Orthogonal Subspace Projection (OSP)
         Suppresses background signatures and generates
           fraction maps like the LSU algorithm
   Spectral Angle Mapper (SAM)
         Treats a spectrum like a vector; Finds angle between
   Minimum Distance (MD)
         A simple Gaussian Maximum Likelihood algorithm that
           does not use class probabilities
   Binary Encoding (BE) and Spectral Signature Matching (SSM)
         Bit compare simple binary codes calculated from
                Case Study
    Classification Results
   The SAM, MD, BE, and SSM algorithms were not designed to
    classify mixed pixels

   Accuracy is the correct identification of one of the fractions in a

       Percent Accuracy
       Classifier Ground      Sensor DC      Retrieved
                  Reflectance with Atmosphere Reflectance
         SAM       66.67%        40.00%       66.67%
          MD         66.67%          80.00%          66.67%
          BE         86.67%          66.67%          86.67%
         SSM         93.33%          80.00%          93.33%
                 Case Study
   Atmosphere degrades performance of most of the
    classification algorithms studied

   Removal of the atmosphere is recommended

   The LSU and OSP fraction maps are more useful
        Provide very accurate material
         identification without a large spectral
        Detects not just the material, but the
         amount of material in a given pixel

Shared By: