Uploaded by Jhon Brayan Guerrero

Lec10a Change Detection

Change Detection
Unclassified images can be compared on a pixel-by-pixel or
patch-by-patch basis;
Classified images can be compared with the results indicating
changes in specific classes over time.
Change Detection
• Georeferencing of each image must be relatively precise.
Spatial offsets due to
 sensor positioning
 relief displacement (due to an inadequate DEM)
 the individual pixels representing a particular object or region on the
ground will not coincide in the two images
Problem is minimized if
• Same sensor collected all data (e.g., Landsat 8)
• Viewing conditions are similar
• Sun angle
• Viewing angle
• Pushbroom (rather than whiskbroom)
Presentation borrows heavily from:
Change Detection
• Atmospheric effects must be accounted for
 Atmospheric correction
 Atmospheric normalization
Atmospheric correction of the individual images is not required, but
normalization is essential.
If classified images are to be compared, the categories (and labeling)
must be the same in both images.
Change Detection
• Resolution of all types should be as similar as possible.
 Spatial: Change can only be associated reliably with the lowest
spatial resolution image.
 Spectral: Spectral bands from different sensors typically have
different characteristics even for the same spectral range (green, NIR,
SWIR, …). There may be differences between images that are
associated with the band selection even when there is no real change,
(e.g., Landsat 5, 7, and 8).
• Band center
• Band width
 Radiometric: There can be significant differences introduced by
differences in dynamic range and bit depth.
 Temporal: This refers to both time of day and time of year.
• Sun angle (control for illumination effect, shadows, etc.)
• Time of year to compare similar surface conditions
• amount of foliage (leaf-on, leaf-off)
• crop stage, etc.)
• shadowing
Change Detection
Ideally, two images being compared should meet the following criteria
(Campbell, 2011):
• Acquired from the same sensor, or two sensors that have been
rigorously inter-calibrated (such as two individual sensors from the
same system of sensors)
• Acquired at the same time of day using the same field of view and
look angle
• If of different years, acquired during the same season to minimize
differences due to normal plant life cycles
• Co-registered to within two-tenths of a pixel or less.
• Cloud-free
• Atmospherically corrected to surface reflectance
• Free of other differences that are not part of the signal of interest
(i.e., soil moisture content could be a distraction or a relevant
signal depending on the application; normal forest harvest could be
confused with trees downed in a storm).
Change Detection
Visual Inspection
Visually comparing co-registered images from two dates is always the
first place to start, even if the ultimate goal is to use an automated
algorithm for classification or change detection (Campbell, 2011).
Most image processing packages include tools to swipe one image over
the other, flicker between images, and view images side-by-side.
In some cases, heads up digitizing may be used to identify and classify
change; in other cases, visual inspection is used to help select the most
appropriate automated change detection technique.
Change Detection
The seven change detection technique categories
1. Algebra Based Approach
image differencing
image regression
image ratioing
vegetation index differencing
change vector analysis
2. Transformation
• Iterative Mutivariate Alteration
• Tasseled Cap (KT)
• Gramm-Schmidt (GS)
• Chi-Square
3. Classification Based
Post-Classification Comparison
Spectral-Temporal Combined Analysis
EM Transformation
Unsupervised Change Detection
Hybrid Change Detection
Artificial Neural Networks (ANN)
4. Advanced Models
• Li-Strahler Reflectance Model
• Spectral Mixture Model
• Biophysical Parameter Method
5. GIS
• Integrated GIS and RS Method
• GIS Approach
6. visual Analysis
• Visual Interpretation
7. Other Change Detection Techniques
Measures of spatial dependence
Knowledge-based vision system
Area production method
Combination of three indicators: vegetation
indices, land surface temperature, and spatial
Change curves
Generalized linear models
Curve-theorem-based approach
Structure-based approach
Spatial statistics-based method
Algebra based Approaches
• These algorithms have a common characteristic, i.e. selecting
thresholds to determine the changed areas.
• Simple, easy to implement and interpret, but cannot provide
complete matrices of change information.
• In this category, two aspects are critical for the change detection
– selecting suitable image bands
– selecting suitable thresholds
Image Differencing
• Pros
– Simple
– Easy to interpret
• Cons:
– Difference value is absolute, and the same value may have different meaning
(e.g., 255-254 = 1-0 = 1)
– Requires atmospheric correction or normalization
Image Regression The relationship between pixel values on two dates is based on a
regression function. The residuals are an indicator of where change
• Pros
– Reduces impact of atmospheric, sensor and environmental differences.
• Cons:
– Requires development of accurate regression functions.
– Does not provide change matrix.
Image Ratioing
• Concept
– Date 1 / Date 2
– No-change = 1
– Values less than and greater than 1 are interpretable
– Pick a threshold for change
• Pros
– Simple
– May mitigate problems with viewing conditions, esp. sun angle
• Cons
– Scales change according to a single date, so same change on the ground
may have different score depending on direction of change;
i.e. 50/100 = .5; 100/50 = 2.0