Land Cover Prediction from Satellite Imagery Using Machine Learning Techniques

Abstract

In this work various machine learning techniques such as nearest neighbor algorithm, decision tree, support vector machine, random forest, naïve bayes classifier has been used for land cover prediction from satellite imagery. The input features are collected from satellite image using time-series normalized difference vegetation index (NDVI). The output for six class classifications is impervious, forest, orchard, farm, grass and water. To balance the data in each class synthetic minority oversampling technique (SMOTE) has been used. All the work has been carried out using python software. The highest accuracy is obtained using k-NN.