Evaluation of Bangalore Land Cover Using ML agorithms

The study discusses the use of classification algorithms to analyze remotely sensed data for land cover and land use information. The data can be acquired from various satellites and classified using free and open source software. The article evaluates the performance of different machine learning algorithms in classifying Landsat-8 OLI multispectral dataset of Bangalore City into four land use classes. An ensemble classifier model is developed which results in higher classification accuracy than individual classifiers. This information can be useful for various environmental applications such as urban planning, disaster management, biodiversity studies, etc.

In this work we do a comparative evaluation of the performances of different machine learning algorithms such as Maximum Likelihood Classifier, Random Forest, Multi-layer Perceptron, Support Vector Machine, XGBoost, Stacked Denoising Auto-Encoder and Energy Based Models is performed by computing user’s, producer’s, overall accuracy, kappa statistics, k-fold cross validation and ROC curve. An ensemble classifier model was developed that renders higher classification accuracy than the individual classifiers.

The detailed report on this was published on IAS and can be found here - http://reports.ias.ac.in/report/19330/evaluation-of-different-machine-learning-algorithms-for-multi-spectral-satellite-image-classification.

Below image so sample classification map results: