utilizing multi-resolution image data vs. pansharpened image data for change detection
DESCRIPTION
Utilizing Multi-Resolution Image data vs. Pansharpened Image data for Change Detection. V. Vijayaraj , C.G. O’ Hara & N.H. Younan GeoResources Institute , Mississippi State University. Introduction Change Detection Pansharpening Change Detection Approaches - PowerPoint PPT PresentationTRANSCRIPT
ASPRS Annual Conference 2005 , Baltimore, March 09 2005
Utilizing Multi-Resolution Image data vs. Pansharpened Image data
for Change Detection
V. Vijayaraj , C.G. O’ Hara & N.H. YounanGeoResources Institute , Mississippi State University
ASPRS Annual Conference 2005 , Baltimore, March 09 2005
Outline
• Introduction• Change Detection• Pansharpening• Change Detection Approaches• Case Study using QuickBird Imagery and
eCognition Software• Conclusions
ASPRS Annual Conference 2005 , Baltimore, March 09 2005
Introduction
The use of high resolution imagery to update and maintain
spatial databases has increased. Developing efficient automated
change detection techniques to extract map accurate change
features from coregistered multitemporal, multiresolution
imagery has been an area of growing research interest.
A change detection approach to extract changed urban
features (Ex: new roads, new buildings) using object
based processing, spatial contextual information and data
fusion technique is presented.
ASPRS Annual Conference 2005 , Baltimore, March 09 2005
Change Detection
Change detection involves the analysis of coregistered images
taken at two different times for the same geographical area.
The techniques can be grouped into
•Supervised Change Detection
Change features are extracted by analyzing images
Classified using supervised classification.
•Unsupervised Change Detection
Change features are extracted by analyzing the difference
images. There are different approaches to analyzing difference
images.
ASPRS Annual Conference 2005 , Baltimore, March 09 2005
Pansharpening
Pansharpening is a pixel level data fusion technique used to increase the spatial resolution of the multispectral image using panchromatic image while simultaneously preserving the spectral information. Also known as resolution merge, image integration and multisensor data fusion. Applications
• Sharpen multispectral data
• Enhance features using complementary information
• Enhance the performance of change detection algorithms
using multi-temporal data sets
• Improve Classification accuracy
ASPRS Annual Conference 2005 , Baltimore, March 09 2005
Pansharpening …
• IHS sharpening• Brovey sharpening• Statistical regression model sharpening• High pass filter sharpening• PCA-based sharpening• Wavelet-based sharpening
The spectral and spatial quality of the sharpened image should be analyzed before using the sharpened image for further applications. The spectral information in the pansharpened image should be more similar to the multispectral image while simultaneously an increase in the high detail information is desired.
ASPRS Annual Conference 2005 , Baltimore, March 09 2005
Change Detection Approaches
•Post Classification Change Detection approach ( Decision level change analysis)
Image T2
Image T2Preprocessed
Image T2
Preprocessed Image T2
ThematicClassification T2
ThematicClassification T2
Image T1
Image T1Preprocessed
Image T1
Preprocessed Image T1
ThematicClassification T1
ThematicClassification T1
Post Classification Thematic Change
Detection
Post Classification Thematic Change
Detection
Land Cover/ Land Use Change Maps
Land Cover/ Land Use Change Maps
Some of the preprocessing steps are Coregistration, Radiometric normalization, Color transformation, and Spectral transformation.
ASPRS Annual Conference 2005 , Baltimore, March 09 2005
Change Detection Approaches
•Pre Classification Change Detection approach (Feature level change analysis)
L.R.Image T2
L.R.Image T2Preprocessed L.R. Image T2
Preprocessed L.R. Image T2
L.R. Image T1
L.R. Image T1Preprocessed L.R. Image T1
Preprocessed L.R. Image T1
Change cues, Indicators, Deltas
Change cues, Indicators, Deltas Region Group
Analysis
Region GroupAnalysis
Polygons Indicating
Probable Change
Polygons Indicating
Probable Change
Image T2
Image T2
Image T1
Image T1Preprocessed
Image T1
Preprocessed Image T1
ThematicClassification T1
ThematicClassification T1
Classification of Changed features
Classification of Changed features
Land Cover/ Land Use Change Maps
Land Cover/ Land Use Change Maps
Mask based on change cues
Mask based on change cues
Mask based on change cues
Mask based on change cues
ASPRS Annual Conference 2005 , Baltimore, March 09 2005
Change Detection Approaches
•Object based Change Detection approach (Object level change analysis using data fusion)
Image T2
Image T2Preprocessed
Image T2
Preprocessed Image T2
Image T1
Image T1Preprocessed
Image T1
Preprocessed Image T1
MultiresolutionSegmentation
into Image objects
MultiresolutionSegmentation
into Image objects
Classification of changed objects
Based on features from T1 and T2
Classification of changed objects
Based on features from T1 and T2
Land Cover/ Land Use Change Maps
Land Cover/ Land Use Change Maps
ASPRS Annual Conference 2005 , Baltimore, March 09 2005
Case Study
A Case study was conducted using QuickBird imagery of Starkville, Mississippi.
QuickBird CharacteristicsSpatial Resolution: Pan 0.6m MS 2.4 mSpectral bands: Pan: 450nm-900nm Blue: 450nm-520nm
Green:520nm-600nm Red: 600nm-690nm NIR: 760nm-900nm
Time Step1: Feb-2002Time Step2: Mar-2004
ASPRS Annual Conference 2005 , Baltimore, March 09 2005
Multispectral image time1& time2
Multispectral Time 1 Multispectral Time 2
ASPRS Annual Conference 2005 , Baltimore, March 09 2005
Multispectral Image
An area of interest – Multispectral time2
ASPRS Annual Conference 2005 , Baltimore, March 09 2005
Pansharpened Image
An area of interest – Pansharpened time2
ASPRS Annual Conference 2005 , Baltimore, March 09 2005
Object based Approach
•eCognition an object oriented image analysis software was used for change detection.
•The multispectral and Pansharpened images at time2 were segmented into image objects based on scale, color, shape and compactness.
•Segmentation was not done on Time 1 image instead the object domain at time2 was used to drill down to images in time 1 and compare object features.
ASPRS Annual Conference 2005 , Baltimore, March 09 2005
IHS Transformation
The RGB- IHS color transform was performed and the transformed layers were also used.RGB- IHS setting :R= Green; G= Red ; B= NIR
ASPRS Annual Conference 2005 , Baltimore, March 09 2005
Features
•Hue Difference: The hue Difference was thresholded to identify the new( changed) features (used to identify new urban features and water bodies)
Hue Difference=Hue time2- Hue time1
•Water Ratio: Water ratio was used to identify new water bodies inside the new features class domain
Water Ratio= (Blue+Green) / NIR
Spatial contextual information to add objects along the edge of water bodies to the appropriate class
•Hue: The highest 10% quantile of Mean Hue of the objects were used to identify other existing urban features in time2.
ASPRS Annual Conference 2005 , Baltimore, March 09 2005
Features …
•NDVI: NDVI in time step 2 was used to classify vegetation NDVI= (NIR-Red)/(NIR+Red)
NDVI was also used to identify cleared / barren areas
Some of the urban features which were classified as cleared were reclassified based on their proximity to urban features.
•Water ratio: Water ratio was used to classify existing water bodies. Building shadows were also picked up as water were removed based on amount of relative border with other water objects
ASPRS Annual Conference 2005 , Baltimore, March 09 2005
Hue Time1
Multispectral Hue Time 1 Pansharpened Hue Time 1
ASPRS Annual Conference 2005 , Baltimore, March 09 2005
Hue Time2
Multispectral Hue Time 2 Pansharpened Hue Time 2
ASPRS Annual Conference 2005 , Baltimore, March 09 2005
Hue Difference
Multispectral Hue Difference Pansharpened Hue Difference
ASPRS Annual Conference 2005 , Baltimore, March 09 2005
Water Ratio Time1
Multispectral Water Ratio Time 1 Pansharpened Water Ratio Time 1
ASPRS Annual Conference 2005 , Baltimore, March 09 2005
Water Ratio Time2
Multispectral Water Ratio Time 2 Pansharpened Water Ratio Time 2
ASPRS Annual Conference 2005 , Baltimore, March 09 2005
Water Ratio Difference
Multispectral Water Ratio Difference Pansharpened Water Ratio Difference
ASPRS Annual Conference 2005 , Baltimore, March 09 2005
NDVI Time1
Multispectral NDVI Time 1 Pansharpened NDVI Time 1
ASPRS Annual Conference 2005 , Baltimore, March 09 2005
NDVI Time2
Multispectral NDVI Time 2 Pansharpened NDVI Time 2
ASPRS Annual Conference 2005 , Baltimore, March 09 2005
NDVI Difference
Multispectral NDVI Difference Pansharpened NDVI Difference
ASPRS Annual Conference 2005 , Baltimore, March 09 2005
Change Features
Multispectral Changed Features Pansharpened Changed Features
ASPRS Annual Conference 2005 , Baltimore, March 09 2005
Multispectral Classification
ASPRS Annual Conference 2005 , Baltimore, March 09 2005
Pansharpened Classification
ASPRS Annual Conference 2005 , Baltimore, March 09 2005
Comparison
MultispectralPansharpened
ASPRS Annual Conference 2005 , Baltimore, March 09 2005
Conclusions
•A Change detection approach using high resolution imagery, object based classification, spatial context information and data fusion techniques was illustrated.
•The Pansharpened images can be used to extract features that are not distinguishable in the multispectral image.
•The spectral and spatial quality of the sharpened image need to be analyzed before using them for classification and change detection.