script.md

Machine-Learning-through-Google-Earth-Engine

By Anish K M

Land Cover Classification using Sentinel-2 Imagery

This project uses Google Earth Engine to perform land cover classification on Sentinel-2 imagery for the year 2023.

Steps:

1. Load Sentinel-2 image collection and preprocess the data.

2. Merge and sample training data for different land cover classes.

3. Train a Random Forest classifier on the sampled data.

4. Classify the image and display the results.

5. Evaluate the classifier performance using confusion matrices and accuracy.

6. Export the training data for further analysis.

Loading and Preprocessing the Sentinel-2 Image Collection

Loading Sentinel-2 Imagery

var image = imageCollection.filterDate('2023-01-01', '2023-12-31').filterBounds(ddn).median()

The script loads Sentinel-2 images from the year 2023 over a specific region (ddn).

The median() function is used to reduce the image collection to a single image,

which represents the median pixel value across the year.

Selecting Bands and Adding Spectral Indices

var bands = ['B2', 'B3', 'B4', 'B8'] var image = image.select(bands).addBands(image.normalizedDifference(['B8', 'B4']).rename('NDVI'))

The code selects specific bands (B2, B3, B4, and B8) and adds the Normalized Difference Vegetation Index (NDVI) as an additional band.

NDVI is computed using the near-infrared (B8) and red (B4) bands.

Display Parameters

var displayparameters = { min: 1000, max: 4500, bands: ['B8', 'B4', 'B3'] } Map.addLayer(image, displayparameters, "Image")

The image is displayed on the map using the specified bands (B8, B4, B3) and display parameters.

Loading Training Data

var label = "Class" var training = Water.merge(Forest).merge(Urban)

The script combines different land cover classes (Water, Forest, Urban) into a single training dataset and assigns them a label "Class".

Extracting Features for Training

Sampling Training Data

var trainingimage = image.sampleRegions({ collection: training, properties: [label], scale: 10 })

Features are extracted from the image for each training region. The sampling is done at a 10-meter scale.

Splitting Data into Training and Testing Sets

Randomly Splitting Data

var traingData = trainingimage.randomColumn() var trainSet = traingData.filter(ee.Filter.lessThan('random',0.8')) var testSet = traingData.filter(ee.Filter.greaterThanOrEquals('random',0.8'))

The data is split into training (80%) and testing (20%) sets using a random column.

Training the Random Forest Classifier

Training the Classifier

var classifier = ee.Classifier.smileRandomForest({numberOfTrees:100, variablesPerSplit: 2, minLeafPopulation: 1, bagFraction: 0.5, seed: 0}) var classifier = ee.Classifier.smileRandomForest(100).train(trainSet, label, bands)

The script trains a Random Forest classifier using the training set. The classifier uses 100 trees and the specified parameters

Classifying the Image

Applying the Classifier

var classified = image.classify(classifier)

The trained classifier is used to classify the entire image or region of interest.

Displaying the Classified Image

Map.centerObject(ddn, 10) Map.addLayer(classified, {min: 1, max: 3, palette: ['green', 'blue', 'red']}, 'LandCover')

The classified image is displayed with a color palette representing different land cover classes.

Evaluating the Classifier

Training Accuracy

var trainAccuracy = classifier.confusionMatrix() print('Training error matrix', trainAccuracy) print('Training overall accuracy', trainAccuracy.accuracy())

The confusion matrix and accuracy for the training set are printed.

Validation Accuracy

testSet = testSet.classify(classifier) var validationAccuracy = testSet.errorMatrix(label, 'classification') print('Validation error matrix', validationAccuracy) print('Validation accuracy', validationAccuracy.accuracy())

The script evaluates the classifier's performance on the test set and prints the validation confusion matrix and accuracy.

Exporting Data

Exporting to Google Earth Engine Asset

Export.table.toAsset({ collection: training, description: 'LCsample2023', assetId: 'LCsample2023' })

The training data is also exported to Google Drive in SHP format.