Development of a Deep Neural Network (DNN) Model for Feature Selection from Satellite Images

This is a Preprint and has not been peer reviewed. This is version 1 of this Preprint.

Add a Comment

You must log in to post a comment.


Comments

There are no comments or no comments have been made public for this article.

Downloads

Download Preprint

Authors

Soma Mitra, Debkumar Chowdhury, Mauparna Nandan, Kajori Parial, Saikat Basu

Abstract

Advances in space-based observation, using remotely sensed data, has proved to be an important tool to monitor the globe, including the areas inaccessible to humans. The Sundarban deltaic region, witnessing the confluence of vast expanses of tropical mangrove forests, tidal rivers, and estuaries, is one such area. Considered as one of the richest biodiversity hotspot zones on earth, home to a large spectrum of biodiversity (flora and fauna), including endangered or threatened species, this forest plays a critical role in land reclamation, coastal habitat protection, and local socioeconomics. However, the forests have been experiencing changes due to climatic forces and anthropogenic activities. Monitoring these changes is crucial for adopting precise management practices. In this work Landsat 8 images were used to identify the land use and land cover in the Sundarbans. For classification, a new deep Neural Network (DNN) model is proposed. A comparative analysis of overall accuracy (OA) of the proposed DNN model with two popular Machine Learning (ML) models, Random Forest (RF) and XGBoost showed 98.9% 97.0% and 98.1% OA respectively. Additive exPlanations (SHAP) were used for each model to obtain important features. It was observed that NIR, SWIR1, Blue, and EVI were the most important features. The proposed DNN model outperformed the RF and XGBoost models with these four important features, achieving 98.5% accuracy. On comparison, it was concluded that deep learning techniques are more effective in feature selection from remote sensing images.

DOI

https://doi.org/10.31223/X5FH5H

Subjects

Other Earth Sciences, Other Environmental Sciences

Keywords

XGBoost, Random Forest, Explainable AI, DNN, Sundarbans, , Random Forest, Explainable AI, DNN, Sundarbans

Dates

Published: 2023-09-07 23:27

Last Updated: 2023-09-08 04:23

License

CC-By Attribution-NonCommercial-NoDerivatives 4.0 International

Additional Metadata

Conflict of interest statement:
None

Data Availability (Reason not available):
https://landsat.gsfc.nasa.gov/data/data-resources/, Google Earth Engine