Remote Sensing Supervised Image Classification. Supervised Image Classification ► An image classification procedure that requires interaction with the.

Remote SensingRemote Sensing

Supervised Image ClassificationSupervised Image Classification

Supervised Image ClassificationSupervised Image Classification

► An image classification procedure that An image classification procedure that requires interaction with the analyst requires interaction with the analyst

1. General Procedures1. General Procedures ► Training stage Training stage

 - The analyst identifies the representative  - The analyst identifies the representative training areas (training set) and develops training areas (training set) and develops summary statistics for each categorysummary statistics for each category

► Classification stage Classification stage  - Each pixel is categorized into a land cover  - Each pixel is categorized into a land cover class class

►   Output stage Output stage  - The classified image is presented in GIS or  - The classified image is presented in GIS or other forms other forms

http://aria.arizona.edu/slg/Vandriel.ppt

TrainingTraining

ClassifiersClassifiers

► Minimum distance classifierMinimum distance classifier► Parallelepiped classifier Parallelepiped classifier ► Gaussian maximum likelihood classifierGaussian maximum likelihood classifier

2. Minimum Distance Classifier2. Minimum Distance Classifier

► Calculates mean of the spectral values for Calculates mean of the spectral values for the training set in each band and for each the training set in each band and for each category category

►   Measures the distance from a pixel of Measures the distance from a pixel of unknown identify to the mean of each unknown identify to the mean of each category category

►   Assigns the pixel to the category with the Assigns the pixel to the category with the shortest distance shortest distance

►   Assigns a pixel as "unknown" if the pixel is Assigns a pixel as "unknown" if the pixel is beyond the distances defined by the analyst beyond the distances defined by the analyst

(40,60)

0,0

Minimum Distance Classifier ..Minimum Distance Classifier ..

► Advantage Advantage  computationally simple and fast  computationally simple and fast

►   Disadvantage Disadvantage  insensitive to differences in variance among  insensitive to differences in variance among categories categories

3. Parallelepiped Classifier3. Parallelepiped Classifier

► Forms a decision region by the maximum Forms a decision region by the maximum and minimum values of the training set in and minimum values of the training set in each band and for each category each band and for each category

►   Assigns a pixel to the category where the Assigns a pixel to the category where the pixel falls in pixel falls in

►   Assigns a pixel as "unknown" if it falls Assigns a pixel as "unknown" if it falls outside of all regions outside of all regions    

Parallelepiped Classifier ..Parallelepiped Classifier ..

► Advantage Advantage  computationally simple and fast  computationally simple and fast

    takes differences in variance into account takes differences in variance into account

► Disadvantage Disadvantage  performs poorly when the regions overlap  performs poorly when the regions overlap because of high correlation between because of high correlation between categories (high covariance) categories (high covariance)    

?

4. Gaussian Maximum likelihood 4. Gaussian Maximum likelihood Classifier Classifier

► Assumes the (probability density function) Assumes the (probability density function) distribution of the training set is normal distribution of the training set is normal

► Describes the membership of a pixel in a Describes the membership of a pixel in a category by probability terms category by probability terms

► The probability is computed based on The probability is computed based on probability density function for each probability density function for each category category

Gaussian Maximum likelihood Gaussian Maximum likelihood Classifier .. Classifier ..

► A pixel may occur in several categories but A pixel may occur in several categories but with different probabilities with different probabilities

►   Assign a pixel to the category with the Assign a pixel to the category with the highest probability highest probability

Gaussian Maximum likelihood Gaussian Maximum likelihood Classifier .. Classifier ..

► Advantage Advantage  takes into account the distance, variance,  takes into account the distance, variance, and covariance and covariance

►   Disadvantage Disadvantage  computationally intensive  computationally intensive

5. Training5. Training

► Collect a set of statistics that describe the Collect a set of statistics that describe the

spectral response pattern for each land spectral response pattern for each land cover type to be classified cover type to be classified

► Select several spectral classes Select several spectral classes representative of each land cover category representative of each land cover category

► Avoid pixels between land cover types Avoid pixels between land cover types

Training ..Training ..

Training ..Training ..

►   A minimum of n+1 pixels must be selected A minimum of n+1 pixels must be selected

(n=number of bands) (n=number of bands) ►   More pixels will improve statistical More pixels will improve statistical

representation, 10n or 100n are common representation, 10n or 100n are common ►   Spatially dispersed training areas Spatially dispersed training areas

throughout the scene better represent the throughout the scene better represent the variation of the cover types variation of the cover types

6. Training Set Refinement6. Training Set Refinement

► Graphic representationGraphic representation► Quantitative expressionQuantitative expression► Self-classification Self-classification

Training Set Refinement ..Training Set Refinement ..

Graphic representationGraphic representation► It is necessary to display histograms of It is necessary to display histograms of

training sets to check for normality and training sets to check for normality and purity purity

►   Coincident spectral plot with 2 std dev from Coincident spectral plot with 2 std dev from the mean is useful to check for category the mean is useful to check for category overlap overlap

►   2-D scatter gram is also useful for 2-D scatter gram is also useful for refinementrefinement

Training Set Refinement ..Training Set Refinement ..

► Quantitative expression Quantitative expression  divergence matrix, higher values indicate  divergence matrix, higher values indicate greater separability greater separability

Training Set Refinement ..Training Set Refinement ..

► Training set self-classification Training set self-classification

- interactive preliminary classification - interactive preliminary classification

- use simple and fast classifier to classify - use simple and fast classifier to classify the entire scene the entire scene

► Representative sub-scene classificationRepresentative sub-scene classification

1. Post-Classification Smoothing1. Post-Classification Smoothing

► Majority filter: use a moving window to filter Majority filter: use a moving window to filter out the “salt and pepper” minority pixelsout the “salt and pepper” minority pixels

► By assigning the majority category of the By assigning the majority category of the window to the center pixel of the windowwindow to the center pixel of the window

ReadingsReadings

► Chapter 7Chapter 7