Forest and Water Bodies Segmentation Through Satellite Images Using U-Net

TL;DR

This paper presents a U-Net based deep learning approach for segmenting forests and water bodies in satellite images to aid environmental monitoring, achieving over 82% accuracy in both categories.

Contribution

It introduces a U-Net model tailored for satellite image segmentation of forests and water bodies, demonstrating effective accuracy for environmental monitoring tasks.

Findings

Validation accuracy of 82.55% for forests

Validation accuracy of 82.92% for water bodies

Effective automated monitoring of environmental changes

Abstract

Global environment monitoring is a task that requires additional attention in the contemporary rapid climate change environment. This includes monitoring the rate of deforestation and areas affected by flooding. Satellite imaging has greatly helped monitor the earth, and deep learning techniques have helped to automate this monitoring process. This paper proposes a solution for observing the area covered by the forest and water. To achieve this task UNet model has been proposed, which is an image segmentation model. The model achieved a validation accuracy of 82.55% and 82.92% for the segmentation of areas covered by forest and water, respectively.