3. Pixel-based classification

/*  Author: Afag Rizayeva
    Date: 01-17-2022
    Purpose: Ten fold pixel-based classification of Corona data with incorporation of DEM
    Inputs: Corona image geometry (study area polygon);
            2 Corona image layers manually imported in GEE;
            2.5m elevation layer;
            training data directory;
            testing data directory;

    Adjust the input files' directories and palette variable
    Run the code
    Export the accuracies as csv files and the final classification map
*/

var table = ee.FeatureCollection('users/rizayeva/DS10111041');

//     Set the palette based on number and names of classes
var palette = ['075b0e'/* forest*/,  '075b0e'/* forest*/, '075b0e'/* forest*/, 'ebebeb'/* barren*/, '9de64e'/* grass*/, 'e2d544'/* crop*/, 'ff0000'/* urban*/, '3f553c'/* wetland*/, '0033ff'/* lake*/, 'ffffff'/* snow*/, '88cdf6'/* river*/];

//     Import two Corona images
//     Rename the band names for afterward- ('da') and forward-facing ('df') cameras
var image2 = ee.Image('users/rizayeva/corona_2_5m/DS1011-1040DA_2_5m_JPEG_1band_EPSG32638').rename(['ds10111040da']);
var image3 = ee.Image('users/rizayeva/corona_2_5m/DS1011-1040DF_2_5m_JPEG_1band_EPSG32638').rename(['ds10111040df']);
Map.addLayer(image2, {min: 0, max: 200}, 'DS1011-1040DA', false);
Map.addLayer(image3, {min: 0, max: 200}, 'DS1011-1040DF', false);

//     Define and visualize the elevation layers
var elevation = ee.Image('users/rizayeva/elevation').rename('elevation');
var slope = ee.Terrain.slope(elevation).clip(table);
var aspect = ee.Terrain.aspect(elevation).clip(table);
var aspect = aspect.subtract(30).multiply(Math.PI/180).cos().subtract(1).multiply(-1/2);

var imageVisParams = {bands: ['elevation'], min: 1, max: 3445,  gamma: [1.85], opacity: 1};
Map.addLayer(elevation, imageVisParams, 'Elevation', false);
Map.addLayer(slope, {}, 'slope', false);
Map.addLayer(aspect, {}, 'aspect', false);

var MN_DA = image2.reduceNeighborhood({
  reducer: ee.Reducer.mean(),
  kernel: ee.Kernel.square(3.5),
});
var MN_DF = image3.reduceNeighborhood({
  reducer: ee.Reducer.mean(),
  kernel: ee.Kernel.square(3.5),
});
var stdDevDA = image2.reduceNeighborhood({
  reducer: ee.Reducer.stdDev(),
  kernel: ee.Kernel.circle(3.5),
});
var stdDevDF = image3.reduceNeighborhood({
  reducer: ee.Reducer.stdDev(),
  kernel: ee.Kernel.circle(3.5),
});

// var texture = image2.glcmTexture({size: 4});
// print(texture);
var ENT_DA = image2.glcmTexture({size: 3/*4*/}).select('ds10111040da_ent');//.resample('bilinear').reproject({crs: image2.projection().getInfo()['crs'], scale: 16.5});
var ENT_DF = image3.glcmTexture({size: 3/*4*/}).select('ds10111040df_ent');//.resample('bilinear').reproject({crs: image3.projection().getInfo()['crs'], scale: 16.5});
var COR_DA = image2.glcmTexture({size: 3/*4*/}).select('ds10111040da_corr');//.resample('bilinear').reproject({crs: image2.projection().getInfo()['crs'], scale: 16.5});
var COR_DF = image3.glcmTexture({size: 3/*4*/}).select('ds10111040df_corr');//.resample('bilinear').reproject({crs: image3.projection().getInfo()['crs'], scale: 16.5});
var HOM_DA = image2.glcmTexture({size: 3/*4*/}).select('ds10111040da_idm');//.resample('bilinear').reproject({crs: image2.projection().getInfo()['crs'], scale: 16.5});
var HOM_DF = image3.glcmTexture({size: 3/*4*/}).select('ds10111040df_idm');//.resample('bilinear').reproject({crs: image3.projection().getInfo()['crs'], scale: 16.5});
var SM_DA = image2.glcmTexture({size: 3/*5*/}).select('ds10111040da_asm');//.resample('bilinear').reproject({crs: image2.projection().getInfo()['crs'], scale: 20.12});
var SM_DF = image3.glcmTexture({size: 3/*5*/}).select('ds10111040df_asm');//.resample('bilinear').reproject({crs: image3.projection().getInfo()['crs'], scale: 20.12});
var VAR_DA = image2.glcmTexture({size: 3/*5*/}).select('ds10111040da_var');//.resample('bilinear').reproject({crs: image2.projection().getInfo()['crs'], scale: 20.12});
var VAR_DF = image3.glcmTexture({size: 3/*5*/}).select('ds10111040df_var');//.resample('bilinear').reproject({crs: image3.projection().getInfo()['crs'], scale: 20.12});
var DIS_DA = image2.glcmTexture({size: 3/*5*/}).select('ds10111040da_diss');//.resample('bilinear').reproject({crs: image2.projection().getInfo()['crs'], scale: 20.12});
var DIS_DF = image3.glcmTexture({size: 3/*5*/}).select('ds10111040df_diss');//.resample('bilinear').reproject({crs: image3.projection().getInfo()['crs'], scale: 20.12});
Map.addLayer(MN_DA, {}, 'MN_DA', false);
Map.addLayer(MN_DF, {}, 'MN_DF', false);
Map.addLayer(ENT_DA, {}, 'ENT_DA', false);
Map.addLayer(ENT_DF, {}, 'ENT_DF', false);
Map.addLayer(SM_DA, {}, 'SM_DA', false);
Map.addLayer(SM_DF, {}, 'SM_DF', false);
Map.addLayer(COR_DA, {}, 'COR_DA', false);
Map.addLayer(COR_DF, {}, 'COR_DF', false);
Map.addLayer(HOM_DA, {}, 'HOM_DA', false);
Map.addLayer(HOM_DF, {}, 'HOM_DF', false);
Map.addLayer(VAR_DA, {}, 'VAR_DA', false);
Map.addLayer(VAR_DF, {}, 'VAR_DF', false);
Map.addLayer(DIS_DA, {}, 'DIS_DA', false);
Map.addLayer(DIS_DF, {}, 'DIS_DF', false);

//     Map asset directory that contains all Training and Testing data
var assetIdTR='users/rizayeva/Training_Testing/Training10111040';
var assetIdTE='users/rizayeva/Training_Testing/Testing10111040';

//     Make a list that contains all the point data in the asset directory
var assetListTR = ee.List(ee.data.getList({'id':assetIdTR}));
var assetListTE = ee.List(ee.data.getList({'id':assetIdTE}));
print('Training data', assetListTR);
print('Testing data', assetListTE);
//     Define the number of datasets
var n = assetListTR.size().getInfo();

///////////// Classification
///////////// ///////////// ///////////// ///////////// ///////////// 5 TEX no DEM
//     Concatenate all layers into one 
var pixelPropertiesImage = ee.Image.cat([
  MN_DA,
  MN_DF,
  stdDevDA,
  stdDevDF,
  ENT_DA,
  ENT_DF,
  HOM_DA,
  HOM_DF,
  SM_DA,
  SM_DF
  ]).float();
// print (pixelPropertiesImage);
var bandNames = pixelPropertiesImage.bandNames();
print('pixelPropertiesImage bandNames with 5 textures no DEM', bandNames);
var bandss = ['ds10111040da_mean', 'ds10111040df_mean', 'ds10111040da_stdDev', 'ds10111040df_stdDev', 'ds10111040da_asm', 'ds10111040df_asm', 'ds10111040da_ent', 'ds10111040df_ent', 'ds10111040da_idm', 'ds10111040df_idm'];

//     Generate an empty Image Collection to later add each of the 10 classification results
  var classification = ee.ImageCollection([]);
//     Generate an empty list to later add each of the overall accuracies to be averaged
  var overallAccuracy = ee.List([]);
//     Build a for loop to run a Random Forest classifier using one set of the training and testing data each time
  for (var i=0; i<n; i++){
//     Select the subsequent set of training data each time
    var valueTR=ee.Dictionary(assetListTR.get(i));
    var imgIDTR=ee.String(valueTR.get('id')).getInfo();
    var train=ee.FeatureCollection(imgIDTR);
    // print('train', train);
    // trainingTR=training.merge(train)

//     Overlay the points on the imagery to get training.
var training = pixelPropertiesImage.sampleRegions({
  collection: train,
  properties: ['Classes'],
  scale: 2.5
});

//     Train a CART classifier with default parameters
var classifier = ee.Classifier.smileRandomForest(100).train(training, 'Classes', bandss);

//      Classify the image with the same bands used for training
var classified = pixelPropertiesImage.classify(classifier);
// print('classified', classified); 
// Map.addLayer(classified, {min: 1, max: 11, palette: palette}, 'Classification map with 5 textures no DEM' + (i+1), false);

// //     Export the Image - uncomment if want to export each
// Export.image.toAsset({
//   image: classified.toByte(),//.toFloat(),
//   description: "classification_pixel_5TEXnoDEM" + (i+1),
//   assetId: "classification_pixel_5TEXnoDEM" + (i+1),
//   scale: 2.5,
//   region: geometry10111040,
//   maxPixels:1e13,
// });

//     Overall map
classification=classification.merge(classified);

//     Validation
//     Select the subsequent set of testing data each time
  var valueTE=ee.Dictionary(assetListTE.get(i));
  var imgIDTE=ee.String(valueTE.get('id')).getInfo();
  var test=ee.FeatureCollection(imgIDTE);
  // print('test', test);
  // testing=testing.merge(test)

var validation = pixelPropertiesImage.sampleRegions({
  collection: test,
  properties: ['Classes'],
  scale: 2.5
});
var validated = validation.classify(classifier);

//     Get a confusion matrix representing expected accuracy and an overall accuracy
//     Uncomment the "print" commands if want to visualize the accuracies and matrices for each set of training and testing data in GEE console
var testAccuracy = validated.errorMatrix('Classes', 'classification');
var Matrix = ee.Feature(null, {matrix: testAccuracy.array()});
// print('Validation error matrix with 5 textures no DEM' + (i+1), testAccuracy);
var overall = testAccuracy.accuracy();
// print('Overall' + (i+1), overall);

// //    Export the matrix - uncomment if want to export each
// Export.table.toDrive({
//   collection: ee.FeatureCollection(Matrix),
//   description:'testAccuracy_pixel08122021_wDEM' + (i+1),
//   folder: 'testAccuracy',
//   fileFormat: 'csv'
// });

//     Add subsequent overall accuracy with each loop, to initially empty "overallAccuracy" list
overallAccuracy = overallAccuracy.add(overall);

  }

//     Image collection of classification results from each set of training data
var classifiedX = classification;
// print('Classification maps collection', classifiedX);

//     Best map representing the most common pixel value based on all sets of training data
var bestClassificationPxl = classifiedX.mode();

//     The following map may not load due to memory limits, but should succesfully export to Assets or Drive
Map.addLayer(bestClassificationPxl.toByte(), {min: 1, max: 11, palette: palette}, "Best pixel-based classification with 5 textures no DEM", false);

//     Export the best map to Drive
Export.image.toDrive({
  image: bestClassificationPxl.toByte(),
  description: "Pxl_5TEXnoDEM",
  folder: "pixelBasedClassifications",
  scale: 2.5,
  region: table,
  maxPixels:1e13,
});

//     Calculate the average overall accuracy
var overallAccuracyPxl = ee.FeatureCollection(overallAccuracy.reduce(ee.Reducer.mean()));
print('Overall accuracy with 5 textures no DEM', overallAccuracyPxl);

//     Calculate the Standard Deviation based on overall accuracies from all training data
var standardDeviation = ee.FeatureCollection(overallAccuracy.reduce(ee.Reducer.stdDev()));
print('SD with 5 textures no DEM', standardDeviation);

//     The "print" function may not always work due to the
//     To make sure it works, we can reduce the number of training and testing datasets used
//     For example, use just two datasets (2 instead of n in the beginning of the for loop)
//     Otherwise, it should always be possible to export them to Drive
Export.table.toDrive({
  collection: overallAccuracyPxl,
  description:'overallAccuracyPxl5TEXnoDEM10111040',
  folder: 'classification_accuracies',
  fileFormat: 'csv'
});

Export.table.toDrive({
  collection: standardDeviation,
  description:'standardDeviationPxl5TEXnoDEM10111040',
  folder: 'classification_accuracies',
  fileFormat: 'csv'
});

///////////// ///////////// ///////////// ///////////// ///////////// 8 TEX no DEM
//     Concatenate all layers into one 
var pixelPropertiesImage = ee.Image.cat([
  MN_DA,
  MN_DF,
  stdDevDA,
  stdDevDF,
  ENT_DA,
  ENT_DF,
  HOM_DA,
  HOM_DF,
  SM_DA,
  SM_DF,
  VAR_DA,
  VAR_DF,
  DIS_DA,
  DIS_DF,
  COR_DA,
  COR_DF
  ]).float();
// print (pixelPropertiesImage);
var bandNames = pixelPropertiesImage.bandNames();
print('pixelPropertiesImage bandNames with 8 textures no DEM', bandNames);
var bandss = ['ds10111040da_mean', 'ds10111040df_mean', 'ds10111040da_stdDev', 'ds10111040df_stdDev', 'ds10111040da_asm', 'ds10111040df_asm', 'ds10111040da_var', 'ds10111040df_var', 'ds10111040da_diss', 'ds10111040df_diss', 'ds10111040da_ent', 'ds10111040df_ent', 'ds10111040da_corr', 'ds10111040df_corr', 'ds10111040da_idm', 'ds10111040df_idm'];

//     Generate an empty Image Collection to later add each of the 10 classification results
  var classification = ee.ImageCollection([]);
//     Generate an empty list to later add each of the overall accuracies to be averaged
  var overallAccuracy = ee.List([]);
//     Build a for loop to run a Random Forest classifier using one set of the training and testing data each time
  for (var i=0; i<n; i++){
//     Select the subsequent set of training data each time
    var valueTR=ee.Dictionary(assetListTR.get(i));
    var imgIDTR=ee.String(valueTR.get('id')).getInfo();
    var train=ee.FeatureCollection(imgIDTR);
    // print('train', train);
    // trainingTR=training.merge(train)

//     Overlay the points on the imagery to get training.
var training = pixelPropertiesImage.sampleRegions({
  collection: train,
  properties: ['Classes'],
  scale: 2.5
});

//     Train a CART classifier with default parameters
var classifier = ee.Classifier.smileRandomForest(100).train(training, 'Classes', bandss);

//      Classify the image with the same bands used for training
var classified = pixelPropertiesImage.classify(classifier);
// print('classified', classified); 
// Map.addLayer(classified, {min: 1, max: 11, palette: palette}, 'Classification map with 8 textures no DEM' + (i+1), false);

// //     Export the Image - uncomment if want to export each
// Export.image.toAsset({
//   image: classified.toByte(),//.toFloat(),
//   description: "classification_pixel_8TEXnoDEM" + (i+1),
//   assetId: "classification_pixel_8TEXnoDEM" + (i+1),
//   scale: 2.5,
//   region: geometry10111040,
//   maxPixels:1e13,
// });

//     Overall map
classification=classification.merge(classified);

//     Validation
//     Select the subsequent set of testing data each time
  var valueTE=ee.Dictionary(assetListTE.get(i));
  var imgIDTE=ee.String(valueTE.get('id')).getInfo();
  var test=ee.FeatureCollection(imgIDTE);
  // print('test', test);
  // testing=testing.merge(test)

var validation = pixelPropertiesImage.sampleRegions({
  collection: test,
  properties: ['Classes'],
  scale: 2.5
});
var validated = validation.classify(classifier);

//     Get a confusion matrix representing expected accuracy and an overall accuracy
//     Uncomment the "print" commands if want to visualize the accuracies and matrices for each set of training and testing data in GEE console
var testAccuracy = validated.errorMatrix('Classes', 'classification');
var Matrix = ee.Feature(null, {matrix: testAccuracy.array()});
// print('Validation error matrix with 8 textures no DEM' + (i+1), testAccuracy);
var overall = testAccuracy.accuracy();
// print('Overall' + (i+1), overall);

// //    Export the matrix - uncomment if want to export each
// Export.table.toDrive({
//   collection: ee.FeatureCollection(Matrix),
//   description:'testAccuracy_pixel08122021_wDEM' + (i+1),
//   folder: 'testAccuracy',
//   fileFormat: 'csv'
// });

//     Add subsequent overall accuracy with each loop, to initially empty "overallAccuracy" list
overallAccuracy = overallAccuracy.add(overall);

  }

//     Image collection of classification results from each set of training data
var classifiedX = classification;
// print('Classification maps collection', classifiedX);

//     Best map representing the most common pixel value based on all sets of training data
var bestClassificationPxl = classifiedX.mode();

//     The following map may not load due to memory limits, but should succesfully export to Assets or Drive
Map.addLayer(bestClassificationPxl.toByte(), {min: 1, max: 11, palette: palette}, "Best pixel-based classification with 8 textures no DEM", false);

//     Export the best map to Drive
Export.image.toDrive({
  image: bestClassificationPxl.toByte(),
  description: "Pxl_8TEXnoDEM",
  folder: "pixelBasedClassifications",
  scale: 2.5,
  region: table,
  maxPixels:1e13,
});

//     Calculate the average overall accuracy
var overallAccuracyPxl = ee.FeatureCollection(overallAccuracy.reduce(ee.Reducer.mean()));
print('Overall accuracy with 8 textures no DEM', overallAccuracyPxl);

//     Calculate the Standard Deviation based on overall accuracies from all training data
var standardDeviation = ee.FeatureCollection(overallAccuracy.reduce(ee.Reducer.stdDev()));
print('SD with 8 textures no DEM', standardDeviation);

//     The "print" function may not always work due to the
//     To make sure it works, we can reduce the number of training and testing datasets used
//     For example, use just two datasets (2 instead of n in the beginning of the for loop)
//     Otherwise, it should always be possible to export them to Drive
Export.table.toDrive({
  collection: overallAccuracyPxl,
  description:'overallAccuracyPxl8TEXnoDEM10111040',
  folder: 'classification_accuracies',
  fileFormat: 'csv'
});

Export.table.toDrive({
  collection: standardDeviation,
  description:'standardDeviationPxl8TEXnoDEM10111040',
  folder: 'classification_accuracies',
  fileFormat: 'csv'
});

///////////// ///////////// ///////////// ///////////// ///////////// 5 TEX w DEM
//     Concatenate all layers into one 
var pixelPropertiesImage = ee.Image.cat([
  MN_DA,
  MN_DF,
  stdDevDA,
  stdDevDF,
  ENT_DA,
  ENT_DF,
  HOM_DA,
  HOM_DF,
  SM_DA,
  SM_DF,
  elevation,
  slope,
  aspect
  ]).float();
// print (pixelPropertiesImage);
var bandNames = pixelPropertiesImage.bandNames();
print('pixelPropertiesImage bandNames with 5 textures with DEM', bandNames);
var bandss = ['ds10111040da_mean', 'ds10111040df_mean', 'ds10111040da_stdDev', 'ds10111040df_stdDev', 'ds10111040da_asm', 'ds10111040df_asm', 'ds10111040da_ent', 'ds10111040df_ent', 'ds10111040da_idm', 'ds10111040df_idm', 'elevation', 'slope', 'aspect'];

//     Generate an empty Image Collection to later add each of the 10 classification results
  var classification = ee.ImageCollection([]);
//     Generate an empty list to later add each of the overall accuracies to be averaged
  var overallAccuracy = ee.List([]);
//     Build a for loop to run a Random Forest classifier using one set of the training and testing data each time
  for (var i=0; i<n; i++){
//     Select the subsequent set of training data each time
    var valueTR=ee.Dictionary(assetListTR.get(i));
    var imgIDTR=ee.String(valueTR.get('id')).getInfo();
    var train=ee.FeatureCollection(imgIDTR);
    // print('train', train);
    // trainingTR=training.merge(train)

//     Overlay the points on the imagery to get training.
var training = pixelPropertiesImage.sampleRegions({
  collection: train,
  properties: ['Classes'],
  scale: 2.5
});

//     Train a CART classifier with default parameters
var classifier = ee.Classifier.smileRandomForest(100).train(training, 'Classes', bandss);

//      Classify the image with the same bands used for training
var classified = pixelPropertiesImage.classify(classifier);
// print('classified', classified); 
// Map.addLayer(classified, {min: 1, max: 11, palette: palette}, 'Classification map with 5 textures with DEM' + (i+1), false);

// //     Export the Image - uncomment if want to export each
// Export.image.toAsset({
//   image: classified.toByte(),//.toFloat(),
//   description: "classification_pixel_5TEXwDEM" + (i+1),
//   assetId: "classification_pixel_5TEXwDEM" + (i+1),
//   scale: 2.5,
//   region: geometry10111040,
//   maxPixels:1e13,
// });

//     Overall map
classification=classification.merge(classified);

//     Validation
//     Select the subsequent set of testing data each time
  var valueTE=ee.Dictionary(assetListTE.get(i));
  var imgIDTE=ee.String(valueTE.get('id')).getInfo();
  var test=ee.FeatureCollection(imgIDTE);
  // print('test', test);
  // testing=testing.merge(test)

var validation = pixelPropertiesImage.sampleRegions({
  collection: test,
  properties: ['Classes'],
  scale: 2.5
});
var validated = validation.classify(classifier);

//     Get a confusion matrix representing expected accuracy and an overall accuracy
//     Uncomment the "print" commands if want to visualize the accuracies and matrices for each set of training and testing data in GEE console
var testAccuracy = validated.errorMatrix('Classes', 'classification');
var Matrix = ee.Feature(null, {matrix: testAccuracy.array()});
// print('Validation error matrix with 5 textures with DEM' + (i+1), testAccuracy);
var overall = testAccuracy.accuracy();
// print('Overall' + (i+1), overall);

// //    Export the matrix - uncomment if want to export each
// Export.table.toDrive({
//   collection: ee.FeatureCollection(Matrix),
//   description:'testAccuracy_pixel08122021_wDEM' + (i+1),
//   folder: 'testAccuracy',
//   fileFormat: 'csv'
// });

//     Add subsequent overall accuracy with each loop, to initially empty "overallAccuracy" list
overallAccuracy = overallAccuracy.add(overall);

  }

//     Image collection of classification results from each set of training data
var classifiedX = classification;
// print('Classification maps collection', classifiedX);

//     Best map representing the most common pixel value based on all sets of training data
var bestClassificationPxl = classifiedX.mode();

//     The following map may not load due to memory limits, but should succesfully export to Assets or Drive
Map.addLayer(bestClassificationPxl.toByte(), {min: 1, max: 11, palette: palette}, "Best pixel-based classification with 5 textures with DEM", false);

//     Export the best map to Drive
Export.image.toDrive({
  image: bestClassificationPxl.toByte(),
  description: "Pxl_5TEXwDEM",
  folder: "pixelBasedClassifications",
  scale: 2.5,
  region: table,
  maxPixels:1e13,
});

//     Calculate the average overall accuracy
var overallAccuracyPxl = ee.FeatureCollection(overallAccuracy.reduce(ee.Reducer.mean()));
print('Overall accuracy with 5 textures with DEM', overallAccuracyPxl);

//     Calculate the Standard Deviation based on overall accuracies from all training data
var standardDeviation = ee.FeatureCollection(overallAccuracy.reduce(ee.Reducer.stdDev()));
print('SD with 5 textures with DEM', standardDeviation);

//     The "print" function may not always work due to the
//     To make sure it works, we can reduce the number of training and testing datasets used
//     For example, use just two datasets (2 instead of n in the beginning of the for loop)
//     Otherwise, it should always be possible to export them to Drive
Export.table.toDrive({
  collection: overallAccuracyPxl,
  description:'overallAccuracyPxl5TEXwDEM10111040',
  folder: 'classification_accuracies',
  fileFormat: 'csv'
});

Export.table.toDrive({
  collection: standardDeviation,
  description:'standardDeviationPxl5TEXwDEM10111040',
  folder: 'classification_accuracies',
  fileFormat: 'csv'
});

///////////// ///////////// ///////////// ///////////// ///////////// 8 TEX w DEM
//     Concatenate all layers into one 
var pixelPropertiesImage = ee.Image.cat([
  MN_DA,
  MN_DF,
  stdDevDA,
  stdDevDF,
  ENT_DA,
  ENT_DF,
  HOM_DA,
  HOM_DF,
  SM_DA,
  SM_DF,
  VAR_DA,
  VAR_DF,
  DIS_DA,
  DIS_DF,
  COR_DA,
  COR_DF,
  elevation,
  slope,
  aspect
  ]).float();
// print (pixelPropertiesImage);
var bandNames = pixelPropertiesImage.bandNames();
print('pixelPropertiesImage bandNames with 8 textures with DEM', bandNames);
var bandss = ['ds10111040da_mean', 'ds10111040df_mean', 'ds10111040da_stdDev', 'ds10111040df_stdDev', 'ds10111040da_asm', 'ds10111040df_asm', 'ds10111040da_var', 'ds10111040df_var', 'ds10111040da_diss', 'ds10111040df_diss', 'ds10111040da_ent', 'ds10111040df_ent', 'ds10111040da_corr', 'ds10111040df_corr', 'ds10111040da_idm', 'ds10111040df_idm', 'elevation', 'slope', 'aspect'];

//     Generate an empty Image Collection to later add each of the 10 classification results
  var classification = ee.ImageCollection([]);
//     Generate an empty list to later add each of the overall accuracies to be averaged
  var overallAccuracy = ee.List([]);
//     Build a for loop to run a Random Forest classifier using one set of the training and testing data each time
  for (var i=0; i<n; i++){
//     Select the subsequent set of training data each time
    var valueTR=ee.Dictionary(assetListTR.get(i));
    var imgIDTR=ee.String(valueTR.get('id')).getInfo();
    var train=ee.FeatureCollection(imgIDTR);
    // print('train', train);
    // trainingTR=training.merge(train)

//     Overlay the points on the imagery to get training.
var training = pixelPropertiesImage.sampleRegions({
  collection: train,
  properties: ['Classes'],
  scale: 2.5
});

//     Train a CART classifier with default parameters
var classifier = ee.Classifier.smileRandomForest(100).train(training, 'Classes', bandss);

//      Classify the image with the same bands used for training
var classified = pixelPropertiesImage.classify(classifier);
// print('classified', classified); 
// Map.addLayer(classified, {min: 1, max: 11, palette: palette}, 'Classification map with 8 textures with DEM' + (i+1), false);

// //     Export the Image - uncomment if want to export each
// Export.image.toAsset({
//   image: classified.toByte(),//.toFloat(),
//   description: "classification_pixel_8TEXwDEM" + (i+1),
//   assetId: "classification_pixel_8TEXwDEM" + (i+1),
//   scale: 2.5,
//   region: geometry10111040,
//   maxPixels:1e13,
// });

//     Overall map
classification=classification.merge(classified);

//     Validation
//     Select the subsequent set of testing data each time
  var valueTE=ee.Dictionary(assetListTE.get(i));
  var imgIDTE=ee.String(valueTE.get('id')).getInfo();
  var test=ee.FeatureCollection(imgIDTE);
  // print('test', test);
  // testing=testing.merge(test)

var validation = pixelPropertiesImage.sampleRegions({
  collection: test,
  properties: ['Classes'],
  scale: 2.5
});
var validated = validation.classify(classifier);

//     Get a confusion matrix representing expected accuracy and an overall accuracy
//     Uncomment the "print" commands if want to visualize the accuracies and matrices for each set of training and testing data in GEE console
var testAccuracy = validated.errorMatrix('Classes', 'classification');
var Matrix = ee.Feature(null, {matrix: testAccuracy.array()});
// print('Validation error matrix with 8 textures with DEM' + (i+1), testAccuracy);
var overall = testAccuracy.accuracy();
// print('Overall' + (i+1), overall);

// //    Export the matrix - uncomment if want to export each
// Export.table.toDrive({
//   collection: ee.FeatureCollection(Matrix),
//   description:'testAccuracy_pixel08122021_wDEM' + (i+1),
//   folder: 'testAccuracy',
//   fileFormat: 'csv'
// });

//     Add subsequent overall accuracy with each loop, to initially empty "overallAccuracy" list
overallAccuracy = overallAccuracy.add(overall);

  }

//     Image collection of classification results from each set of training data
var classifiedX = classification;
// print('Classification maps collection', classifiedX);

//     Best map representing the most common pixel value based on all sets of training data
var bestClassificationPxl = classifiedX.mode();

//     The following map may not load due to memory limits, but should succesfully export to Assets or Drive
Map.addLayer(bestClassificationPxl.toByte(), {min: 1, max: 11, palette: palette}, "Best pixel-based classification with 8 textures with DEM", false);

//     Export the best map to Drive
Export.image.toDrive({
  image: bestClassificationPxl.toByte(),
  description: "Pxl_8TEXwDEM",
  folder: "pixelBasedClassifications",
  scale: 2.5,
  region: table,
  maxPixels:1e13,
});

//     Calculate the average overall accuracy
var overallAccuracyPxl = ee.FeatureCollection(overallAccuracy.reduce(ee.Reducer.mean()));
print('Overall accuracy with 8 textures with DEM', overallAccuracyPxl);

//     Calculate the Standard Deviation based on overall accuracies from all training data
var standardDeviation = ee.FeatureCollection(overallAccuracy.reduce(ee.Reducer.stdDev()));
print('SD with 8 textures with DEM', standardDeviation);

//     The "print" function may not always work due to the
//     To make sure it works, we can reduce the number of training and testing datasets used
//     For example, use just two datasets (2 instead of n in the beginning of the for loop)
//     Otherwise, it should always be possible to export them to Drive
Export.table.toDrive({
  collection: overallAccuracyPxl,
  description:'overallAccuracyPxl8TEXwDEM10111040',
  folder: 'classification_accuracies',
  fileFormat: 'csv'
});

Export.table.toDrive({
  collection: standardDeviation,
  description:'standardDeviationPxl8TEXwDEM10111040',
  folder: 'classification_accuracies',
  fileFormat: 'csv'
});