Satellite Image Classification Using Python

In this repository, we have classified a Landsat 8 image using Machine learning. This repository contains the code and data for classifying a satellite image of the Kathmandu Valley and its neighboring areas into four land covers: Forest, Urban, Water, and Open/cultivated Land, with an accuracy of 86 percent.

Introduction

The objective of this project is to classify a satellite image of the Kathmandu valley and its neighboring areas into four land covers, viz. Forest, Urban, Water, and Open/cultivated Land with an accuracy of 86 percent in Python. The image was preprocessed using ArcMap 10.8.2 and training samples were taken from the preprocessed image. During training, a callback was performed, checkpoints were saved and hyperparameter tuning was performed.

All the references were taken from the ''here''

Prerequisites

• ArcMap 10.8.2

Installing

1. Clone the repository.
2. Open the project in your preferred IDE.
3. Install the required packages using pip.

##Classification A detailed notebook is uploaded to this repository. Simply download your copy of the notebook, and follow along with the process as described. It can also be accessed at the bottom of this readME file. If you got better results than me, please share it. email: dh.prtsh@gmail.com .

Results

The satellite image was classified into four land covers with an accuracy of 86 percent using TensorFlow to train the model, gdal to read the image, and keras_tuner for hyperparameter tuning.

Authors

• Pratyush Dhungana(https://github.com/pratyush-dh)

Acknowledgments

• Pratik Dhungana (https://github.com/pratik-tan10)

Predicted_Image

Notebook

Test Confusion Matrix

Train Confusion Matrix

Accuracy Graph

Loss Graph