Dataset Creation and Comparative Analysis of Machine Learning Models for Mangrove Classification in Coastal Maharashtra, India

Abstract

This research paper presents a new dataset of Landsat 8 image tiles processed for mapping mangroves in the coastal region of Maharashtra, India using bandcombinations of Band 5, 6 and 4. The dataset includes labelled image tiles which can beused for binary classification of mangroves using Convolutional Neural Network (CNN) and Random Forest algorithms.The radiometric correction of images and creation of composite images were done usingArcGIS Pro, while the image tiles were extracted and labelled using the Geotile library in Google Colab. The performance of CNN and Random Forest algorithms were comparedfor the classification of mangroves. This dataset can be used for further research on mangrove mapping and monitoring using remote sensing techniques.