CrossCutting Data Analysis Techniques

1
Cross-Cutting Data Analysis Techniques

• Dr. Lori Mann Bruce
• Electrical Computer Engineering
• GeoResources Institute
• Mississippi State University
• bruce_at_ece.msstate.edu

2
Current Research Areas

• DSP and ATR for
• Hyperspectral, Multispectral, Lidar

3
Current Research Areas

• Spatial and Spectral Feature Extraction
• Temporal Change Detection
• Hyperspectral Dimensionality Reduction
• Multiclassifiers and Decision Fusion
• Pixel Unmixing
• Effects of Archiving on Data Analysis
  (Compression and Watermarking)

4
Spatial and Spectral Feature Extraction

• Feature Extraction for
• Supervised Classification
• Wavelet transforms, Karhunen-Loeve transforms
• Singular value decomposition
• Maximum-likelihood, nearest-neighbor classifiers
• Unsupervised Classification
• Self-organizing maps
• Fully automated target detection systems

5
Weed Detection (Spatial Features in Multispectal
Images)
6
NDVI Texture Image
False Color Image
Threshold to Plant Density
7
Hyperspectral Dimensionality Reduction

• Best Spectral Band/Range Selection
• Multivariate Statistical Approaches
• Signal Transformation
• Discrete Wavelet Transforms
• Karhunen-Loeve Transforms
• Projection Pursuits

8
Hyperspectral Dimensionality Reduction
(Histogram of optimum set of bands shown in blue)
PCA
LDA
9
Spectral Dimensionality (BU-GLDB Algorithm for
finding best bases)
C1D1
C3D3
C2D2
C4D4
C6D6
C5D5
C7D7
C9D9
C8D8
Find the best pair of spectral bands that can be
combined. The criteria used is the product of the
correlation between pair of bands (C) and
discrimination capability (D) of the combined
band pair.
C1D1
C3D3
C2D2
C4D4
C5D5
C6D6
C8D8
C7D7
C1D1
C3D3
C2D2
C4D4
C5D5
C6D6
C7D7
C3D3
C2D2
C4D4
C5D5
C6D6
C7D7
Merging of these two groups result in a poorer CD
metric than the original groups.
10
Multiclassifiers Decision Fusion
Feature Extractor
Feature Extractor
Feature Extractor
Feature Extractor
Feature Extractor
Classifier
Classifier
Classifier
Classifier
Classifier
Decision Fusion
11
Accurate Pixel Unmixing
Mixed Pixel
Endmembers Abundances
Spectral Unmixing
3 Endmembers 2 vegetations 1 soil
12
Recent Refereed Journal Publications

• C.H. Koger, L.M. Bruce, D.R. Shaw, K.N. Reddy,
  Wavelet Analysis of Hyperspectral
  Reflectance Data for Detecting Pitted
  Morningglory (Ipomoea lacunosa) in Soybean
  (Glycine max), Remote Sensing of Environment,
  vol. 86, no. 1, pp. 108-119, June 2003.
• L.M. Bruce, C.H. Koger, J. Li, "Dimensionality
  reduction of hyperspectral data using discrete
  wavelet transform feature extraction," IEEE
  Trans. Geoscience. Remote Sensing, vol. 40, no.
  10, pp. 2331-2338, Oct 2002.
• L.M. Bruce, J. Li and Y. Huang, "Automated
  detection of subpixel hyperspectral targets with
  adaptive multichannel discrete wavelet
  transform," IEEE Trans. Geoscience Remote
  Sensing, vol. 40, no. 4, pp.977-980, April 2002.
• L.M. Bruce and J. Li, "Wavelets for
  computationally efficient hyperspectral
  derivative analysis," IEEE Trans. Geoscience
  Remote Sensing, vol.39, no.7, pp.1540-1546, July
  2001.
• L.M. Bruce, C. Morgan, S. Larsen, "Automated
  detection of subpixel targets with continuous and
  discrete wavelet transforms," IEEE Trans.
  Geoscience and Remote Sensing, vol. 39, no. 10,
  pp. 2217-2226, Oct 2001.
• J.-G Cao, J.E. Fowler, N.H. Younan, An
  image-adaptive watermark based on a redundant
  wavelet transform, Proc Image Processing, vol.
  2, no. 10, pp. 277 -280, 2001.
• J.E. Fowler, H. Li, Wavelet transforms for
  vector fields using omnidirectionally balanced
  multiwavelets, IEEE Trans Signal Processing,
  vol. 50, no. 12, pp. 3018-3027, Dec 2002.
• J.E. Fowler, D.N. Fox, Embedded wavelet-based
  coding of three-dimensional oceanographic images
  with land masses, IEEE Trans. Geoscience and
  Remote Sensing, vol. 39, no. 2, pp. 284-290, Feb
  2001.

13
Invasive Species Detection and Assessment
?
14
Spectral Baselines
15
Spectral Baselines
Kudzu
Sicklepod
Soda apple
Horseweed
Dogfennel
16
0.2
0.4
a9
d9
0
0.3
0.2
-0.2
0.2
0.4
a8
d8
0
0.2
-0.2
0.05
0.4
a7
d7
0
0.2
Haar
-0.05
0.5
0.1
0.4
0.3
a6
d6
0
0.2
0.1
-0.1
0.5
0.05
0.4
0.3
a5
d5
0
0.2
-0.05
0.1
0.04
0.5
0.02
0.4
a4
d4
0.3
0
0.2
-0.02
0.1
-0.04
0.5
0.02
0.4
a3
d3
0.3
0
0.2
0.1
-0.02
0.02
0.5
0.4
a2
d2
0.3
0
0.2
0.1
-0.02
0.02
0.5
0.4
a1
d1
0.3
0
0.2
0.1
-0.02
200
400
600
800
1000
200
400
600
800
1000
17
5 Class Results (Classification Accuracies and
Confidence Intervals)
18
Invasive Species Detection Automated Algorithm
Development

• Airborne level research
• Collect hyperspectral/multispectral images
• Collect ground truth
• Combine spectral features with spatial features
• Classify vegetation as target or non-target

19
Invasive Species Detection Automated Algorithm
Development
Spectral Features
Spatial Features
20
Collection Date August 30, 2000
Collection Date February 1, 2001
Collection Date July 16, 2001
21
(This image is used as the background for the
presentation.) Location Noxious Weed Studies --
Oswalt Date June 5, 2001 Sensor ITD (Spectral
Visions) RDACS Altitude 6000 ft.
22
Cogongrass (Imperata Cylindrica)

• Invasive species introduced into southern USA in
  early 1900s
• The seventh worst weed in the world
• Listed in the United States Federal Noxious
  Weeds List

23
Example Case Study
24
Thank You