/*
* Copyright (C) 2011-2012 Dr. John Lindsay <jlindsay@uoguelph.ca>
*
* This program is free software: you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
package plugins;
import java.text.DecimalFormat;
import java.util.ArrayList;
import java.util.Date;
import java.util.Random;
import whitebox.geospatialfiles.WhiteboxRaster;
import whitebox.geospatialfiles.WhiteboxRasterBase.DataScale;
import whitebox.geospatialfiles.WhiteboxRasterInfo;
import whitebox.interfaces.WhiteboxPlugin;
import whitebox.interfaces.WhiteboxPluginHost;
/**
* This tool is an unsupervised classification method to be applied to multi-spectral remotely sensed imagery.
* @author Dr. John Lindsay email: jlindsay@uoguelph.ca
*/
public class ModifiedKMeans implements WhiteboxPlugin {
private WhiteboxPluginHost myHost = null;
private String[] args;
/**
* Used to retrieve the plugin tool's name. This is a short, unique name containing no spaces.
* @return String containing plugin name.
*/
@Override
public String getName() {
return "ModifiedKMeans";
}
/**
* Used to retrieve the plugin tool's descriptive name. This can be a longer name (containing spaces) and is used in the interface to list the tool.
* @return String containing the plugin descriptive name.
*/
@Override
public String getDescriptiveName() {
return "Modified k-Means Classification";
}
/**
* Used to retrieve a short description of what the plugin tool does.
* @return String containing the plugin's description.
*/
@Override
public String getToolDescription() {
return "Performs a modified k-means classification on a multi-spectral dataset.";
}
/**
* Used to identify which toolboxes this plugin tool should be listed in.
* @return Array of Strings.
*/
@Override
public String[] getToolbox() {
String[] ret = {"ImageClass"};
return ret;
}
/**
* Sets the WhiteboxPluginHost to which the plugin tool is tied. This is the class
* that the plugin will send all feedback messages, progress updates, and return objects.
* @param host The WhiteboxPluginHost that called the plugin tool.
*/
@Override
public void setPluginHost(WhiteboxPluginHost host) {
myHost = host;
}
/**
* Used to communicate feedback pop-up messages between a plugin tool and the main Whitebox user-interface.
* @param feedback String containing the text to display.
*/
private void showFeedback(String message) {
if (myHost != null) {
myHost.showFeedback(message);
} else {
System.out.println(message);
}
}
/**
* Used to communicate a return object from a plugin tool to the main Whitebox user-interface.
* @return Object, such as an output WhiteboxRaster.
*/
private void returnData(Object ret) {
if (myHost != null) {
myHost.returnData(ret);
}
}
private int previousProgress = 0;
private String previousProgressLabel = "";
/**
* Used to communicate a progress update between a plugin tool and the main Whitebox user interface.
* @param progressLabel A String to use for the progress label.
* @param progress Float containing the progress value (between 0 and 100).
*/
private void updateProgress(String progressLabel, int progress) {
if (myHost != null && ((progress != previousProgress)
|| (!progressLabel.equals(previousProgressLabel)))) {
myHost.updateProgress(progressLabel, progress);
}
previousProgress = progress;
previousProgressLabel = progressLabel;
}
/**
* Used to communicate a progress update between a plugin tool and the main Whitebox user interface.
* @param progress Float containing the progress value (between 0 and 100).
*/
private void updateProgress(int progress) {
if (myHost != null && progress != previousProgress) {
myHost.updateProgress(progress);
}
previousProgress = progress;
}
/**
* Sets the arguments (parameters) used by the plugin.
* @param args An array of string arguments.
*/
@Override
public void setArgs(String[] args) {
this.args = args.clone();
}
private boolean cancelOp = false;
/**
* Used to communicate a cancel operation from the Whitebox GUI.
* @param cancel Set to true if the plugin should be canceled.
*/
@Override
public void setCancelOp(boolean cancel) {
cancelOp = cancel;
}
private void cancelOperation() {
showFeedback("Operation cancelled.");
updateProgress("Progress: ", 0);
}
private boolean amIActive = false;
/**
* Used by the Whitebox GUI to tell if this plugin is still running.
* @return a boolean describing whether or not the plugin is actively being used.
*/
@Override
public boolean isActive() {
return amIActive;
}
/**
* Used to execute this plugin tool.
*/
@Override
public void run() {
amIActive = true;
String inputFilesString = null;
String[] imageFiles = null;
String outputHeader = null;
WhiteboxRasterInfo[] images = null;
WhiteboxRaster ouptut = null;
int nCols = 0;
int nRows = 0;
double z;
int numClasses = 0;
int numImages;
int progress = 0;
int col, row;
int a, i, j;
double[][] data;
double noData = -32768;
double[][] classCentres = null;
double[] classCentre;
ArrayList<double[]> centres = new ArrayList<double[]>();
double[][] imageMetaData;
long[] numPixelsInEachClass;
int maxIterations = 100;
double dist, minDist;
int whichClass;
//double minAdjustment = 10;
byte initializationMode = 0; // maximum dispersion along diagonal
long numPixelsChanged = 0;
long totalNumCells = 0;
boolean totalNumCellsCounted = false;
double percentChanged = 0;
double percentChangedThreshold = 1.0;
double centroidMergeDist = 30;
int minimumAllowableClassSize = 1;
int initialNumClasses = 10000;
double maxDist = Double.POSITIVE_INFINITY;
int unassignedClass = -1;
boolean isNoDataPixel;
if (args.length <= 0) {
showFeedback("Plugin parameters have not been set.");
return;
}
// read the input parameters
inputFilesString = args[0];
outputHeader = args[1];
//numClasses = Integer.parseInt(args[2]);
maxIterations = Integer.parseInt(args[2]);
percentChangedThreshold = Double.parseDouble(args[3]);
centroidMergeDist = Double.parseDouble(args[4]);
if (!args[5].toLowerCase().contains("not specified")) {
maxDist = Double.parseDouble(args[5]);
}
minimumAllowableClassSize = Integer.parseInt(args[6]);
if (args[7].toLowerCase().contains("random")) {
initializationMode = 1; //random positioning
} else {
initializationMode = 0; //maximum dispersion along multi-dimensional diagonal
}
int[] clusterHistory = new int[maxIterations];
double[] changeHistory = new double[maxIterations];
try {
// deal with the input images
imageFiles = inputFilesString.split(";");
numImages = imageFiles.length;
images = new WhiteboxRasterInfo[numImages];
imageMetaData = new double[numImages][3];
for (i = 0; i < numImages; i++) {
images[i] = new WhiteboxRasterInfo(imageFiles[i]);
if (i == 0) {
nCols = images[i].getNumberColumns();
nRows = images[i].getNumberRows();
noData = images[i].getNoDataValue();
} else {
if (images[i].getNumberColumns() != nCols
|| images[i].getNumberRows() != nRows) {
showFeedback("All input images must have the same dimensions (rows and columns).");
return;
}
}
imageMetaData[i][0] = images[i].getNoDataValue();
imageMetaData[i][1] = images[i].getMinimumValue();
imageMetaData[i][2] = images[i].getMaximumValue();
}
numClasses = initialNumClasses;
data = new double[numImages][];
numPixelsInEachClass = new long[numImages];
// now set up the output image
WhiteboxRaster output = new WhiteboxRaster(outputHeader, "rw",
imageFiles[0], WhiteboxRaster.DataType.INTEGER, noData);
output.setDataScale(DataScale.CATEGORICAL);
output.setPreferredPalette("qual.pal");
// initialize the class centres either along the diagonal or randomly
if (initializationMode == 1) {
Random generator = new Random();
double range;
for (a = 0; a < numClasses; a++) {
classCentre = new double[numImages];
for (i = 0; i < numImages; i++) {
range = imageMetaData[i][2] - imageMetaData[i][1];
classCentre[i] = imageMetaData[i][1] + generator.nextDouble() * range;
}
centres.add(classCentre);
}
} else {
double range, spacing;
for (a = 0; a < numClasses; a++) {
classCentre = new double[numImages];
for (i = 0; i < numImages; i++) {
range = imageMetaData[i][2] - imageMetaData[i][1];
spacing = range / numClasses;
classCentre[i] = imageMetaData[i][1] + spacing * a;
}
centres.add(classCentre);
}
}
j = 0;
whichClass = 0;
do {
if (j > 0) {
numClasses = classCentres.length; //centres.size();
centres.clear();
for (a = 0; a < classCentres.length; a++) {
centres.add(classCentres[a]);
}
ArrayList<Long> numPixels = new ArrayList<Long>();
for (i = 0; i < numPixelsInEachClass.length; i++) {
numPixels.add(numPixelsInEachClass[i]);
}
// Remove any empty classes or classes smaller than the minimumAllowableClassSize
boolean flag = true;
a = 0;
do {
if (numPixels.get(a) == 0) {
centres.remove(a);
numPixels.remove(a);
flag = true;
a = -1;
}
a++;
if (a >= numPixels.size()) {
flag = false;
}
} while (flag);
// See if any of the class centroids are close enough to be merged.
long numPixels1, numPixels2;
do {
flag = false;
for (a = 0; a < centres.size(); a++) {
if (flag) {
break;
}
classCentre = centres.get(a);
numPixels1 = numPixels.get(a);
for (int b = a; b < centres.size(); b++) {
numPixels2 = numPixels.get(b);
if (b > a && numPixels1 > 0 && numPixels2 > 0) {
double[] classCentre2 = centres.get(b);
dist = 0;
for (i = 0; i < numImages; i++) {
dist += (classCentre[i] - classCentre2[i]) * (classCentre[i] - classCentre2[i]);
}
dist = Math.sqrt(dist);
if (dist < centroidMergeDist) {
// these two clusters should be merged
double[] classCentre3 = new double[numImages];
long totalPix = numPixels1 + numPixels2;
double weight1 = (double)numPixels1 / totalPix;
double weight2 = (double)numPixels2 / totalPix;
for (int k = 0; k < numImages; k++) {
classCentre3[k] = classCentre[k] * weight1 + classCentre2[k] * weight2;
}
centres.remove(Math.max(a, b));
centres.remove(Math.min(a, b));
centres.add(classCentre3);
numPixels.remove(Math.max(a, b));
numPixels.remove(Math.min(a, b));
numPixels.add(totalPix);
flag = true;
}
if (flag) {
break; // once two have been merged, stop looking and start over.
}
}
}
}
numClasses = centres.size();
} while (flag);
// Remove any classes smaller than the minimumAllowableClassSize
flag = true;
a = 0;
do {
if (numPixels.get(a) < minimumAllowableClassSize) {
centres.remove(a);
numPixels.remove(a);
flag = true;
a = -1;
}
a++;
if (a >= numPixels.size()) {
flag = false;
}
} while (flag);
}
numClasses = centres.size();
classCentres = new double[numClasses][numImages];
for (a = 0; a < numClasses; a++) {
classCentre = centres.get(a);
classCentres[a] = classCentre.clone();
}
j++;
// assign each pixel to a class
updateProgress("Loop " + j, 1);
double[][] classCentreData = new double[numClasses][numImages];
numPixelsInEachClass = new long[numClasses];
numPixelsChanged = 0;
for (row = 0; row < nRows; row++) {
for (i = 0; i < numImages; i++) {
data[i] = images[i].getRowValues(row);
}
for (col = 0; col < nCols; col++) {
// check to see if the cell is a nodata value in any of the input images
isNoDataPixel = false;
for (i = 0; i < numImages; i++) {
if (data[i][col] == imageMetaData[i][0]) {
isNoDataPixel = true;
break;
}
}
if (!isNoDataPixel) {
if (!totalNumCellsCounted) {
totalNumCells++;
}
// calculate the squared distance to each of the centroids
// and assign the pixel the value of the nearest centroid.
minDist = Double.POSITIVE_INFINITY;
whichClass = unassignedClass;
for (a = 0; a < numClasses; a++) {
dist = 0;
for (i = 0; i < numImages; i++) {
dist += (data[i][col] - classCentres[a][i]) * (data[i][col] - classCentres[a][i]);
}
if (dist < minDist && dist <= maxDist) {
minDist = dist;
whichClass = a;
}
}
// See if the assigned class has changed and if it has add it to the total changed cells.
// This is a criterion for stopping.
z = output.getValue(row, col);
if ((int)z != whichClass) {
numPixelsChanged++;
// Assign the output pixel the class value
output.setValue(row, col, whichClass);
}
if (whichClass != unassignedClass) {
numPixelsInEachClass[whichClass]++;
for (i = 0; i < numImages; i++) {
classCentreData[whichClass][i] += (data[i][col] - imageMetaData[i][1]);
}
}
} else {
output.setValue(row, col, noData);
}
}
if (cancelOp) {
cancelOperation();
return;
}
progress = (int) (100f * row / (nRows - 1));
updateProgress("Loop " + j, progress);
}
totalNumCellsCounted = true;
// Update the class centroids
for (a = 0; a < numClasses; a++) {
if (numPixelsInEachClass[a] > 0) {
double[] newClassCentre = new double[numImages];
for (i = 0; i < numImages; i++) {
newClassCentre[i] = classCentreData[a][i] / numPixelsInEachClass[a] + imageMetaData[i][1];
}
for (i = 0; i < numImages; i++) {
classCentres[a][i] = newClassCentre[i];
}
}
}
percentChanged = (double)numPixelsChanged / totalNumCells * 100;
clusterHistory[j - 1] = numClasses;
changeHistory[j - 1] = percentChanged;
} while ((percentChanged > percentChangedThreshold) && (j < maxIterations));
// prepare the report
double[] totalDeviations = new double[numClasses];
int numberOfUnassignedPixels = 0;
for (row = 0; row < nRows; row++) {
for (i = 0; i < numImages; i++) {
data[i] = images[i].getRowValues(row);
}
for (col = 0; col < nCols; col++) {
isNoDataPixel = false;
for (i = 0; i < numImages; i++) {
if (data[i][col] == imageMetaData[i][0]) {
isNoDataPixel = true;
break;
}
}
if (!isNoDataPixel) {
whichClass = (int) (output.getValue(row, col));
if (whichClass != unassignedClass) {
dist = 0;
for (i = 0; i < numImages; i++) {
dist += (data[i][col] - classCentres[whichClass][i]) * (data[i][col] - classCentres[whichClass][i]);
}
totalDeviations[whichClass] += dist;
} else {
numberOfUnassignedPixels++;
}
} else {
output.setValue(row, col, noData);
}
}
if (cancelOp) {
cancelOperation();
return;
}
progress = (int) (100f * row / (nRows - 1));
updateProgress("Loop " + j, progress);
}
double[] standardDeviations = new double[numClasses];
for (a = 0; a < numClasses; a++) {
standardDeviations[a] = Math.sqrt(totalDeviations[a] / (numPixelsInEachClass[a] - 1));
}
DecimalFormat df;
df = new DecimalFormat("0.00");
String retStr = "Modified k-Means Classification Report\n\n";
retStr += " \tCentroid Vector\n";
retStr += " \t";
for (i = 0; i < numImages; i++) {
retStr += "Image" + (i + 1) + "\t";
}
retStr += "SD\tPixels\t% Area\n";
for (a = 0; a < numClasses; a++) {
String str = "";
for (i = 0; i < numImages; i++) {
str += df.format(classCentres[a][i]) + "\t";
}
retStr += "Cluster " + a + "\t" + str + df.format(standardDeviations[a]) + "\t" + numPixelsInEachClass[a] + "\t" + df.format((double)numPixelsInEachClass[a] / totalNumCells * 100) + "\n";
}
retStr += "\n";
retStr += "Number of unassigned pixels (class = -1): " + numberOfUnassignedPixels + "\n\n";
for (i = 0; i < numImages; i++) {
retStr += "Image" + (i + 1) + " = " + images[i].getShortHeaderFile() + "\n";
}
retStr += "\nCluster Centroid Distance Analysis:\n";
for (a = 0; a < numClasses; a++) {
retStr += "\tClus. " + a;
}
retStr += "\n";
//double[][] centroidDistances = new double[numClasses][numClasses];
for (a = 0; a < numClasses; a++) {
retStr += "Cluster " + a;
for (int b = 0; b < numClasses; b++) {
if (b >= a) {
dist = 0;
for (i = 0; i < numImages; i++) {
dist += (classCentres[a][i] - classCentres[b][i]) * (classCentres[a][i] - classCentres[b][i]);
}
retStr += "\t" + df.format(Math.sqrt(dist));
} else {
retStr += "\t";
}
}
retStr += "\n";
}
retStr += "\nCluster Merger History:\n";
retStr += "Iteration\tNumber of Clusters\tPercent Changed\n";
for (i = 0; i < maxIterations; i++) {
if (clusterHistory[i] > 0) {
retStr += (i + 1) + "\t" + clusterHistory[i] + "\t" + changeHistory[i] + "\n";
} else {
break;
}
}
returnData(retStr);
Dendrogram plot = new Dendrogram(classCentres, numPixelsInEachClass);
returnData(plot);
for (i = 0; i < numImages; i++) {
images[i].close();
}
output.addMetadataEntry("Created by the "
+ getDescriptiveName() + " tool.");
output.addMetadataEntry("Created on " + new Date());
output.close();
// returning a header file string displays the image.
returnData(outputHeader);
} catch (OutOfMemoryError oe) {
myHost.showFeedback("An out-of-memory error has occurred during operation.");
} catch (Exception e) {
myHost.showFeedback("An error has occurred during operation. See log file for details.");
myHost.logException("Error in " + getDescriptiveName(), e);
} finally {
updateProgress("Progress: ", 0);
// tells the main application that this process is completed.
amIActive = false;
myHost.pluginComplete();
}
}
// // this is only used for debugging the tool
// public static void main(String[] args) {
// ModifiedKMeans mkm = new ModifiedKMeans();
// args = new String[6];
//// args[0] = "/Users/johnlindsay/Documents/Teaching/GEOG3420/Winter 2012/Labs/Lab1/Data/LE70180302002142EDC00/band1 clipped.dep;/Users/johnlindsay/Documents/Teaching/GEOG3420/Winter 2012/Labs/Lab1/Data/LE70180302002142EDC00/band2 clipped.dep;/Users/johnlindsay/Documents/Teaching/GEOG3420/Winter 2012/Labs/Lab1/Data/LE70180302002142EDC00/band3 clipped.dep;/Users/johnlindsay/Documents/Teaching/GEOG3420/Winter 2012/Labs/Lab1/Data/LE70180302002142EDC00/band4 clipped.dep;/Users/johnlindsay/Documents/Teaching/GEOG3420/Winter 2012/Labs/Lab1/Data/LE70180302002142EDC00/band5 clipped.dep;";
//// args[1] = "/Users/johnlindsay/Documents/Teaching/GEOG3420/Winter 2012/Labs/Lab1/Data/LE70180302002142EDC00/tmp1.dep";
// args[0] = "/Users/johnlindsay/Documents/Data/LandsatData/band1.dep;/Users/johnlindsay/Documents/Data/LandsatData/band2_cropped.dep;/Users/johnlindsay/Documents/Data/LandsatData/band3_cropped.dep;/Users/johnlindsay/Documents/Data/LandsatData/band4_cropped.dep;/Users/johnlindsay/Documents/Data/LandsatData/band5_cropped.dep";
// args[1] = "/Users/johnlindsay/Documents/Data/LandsatData/tmp2.dep";
// args[2] = "5"; // max iterations
// args[3] = "2"; // changed pixels
// args[4] = "30"; // centroid merge dist
// args[5] = "diagonal"; // centroid initiation process
//
// mkm.setArgs(args);
// mkm.run();
//
// }
}