This repository holds the code and resources for our undergraduate thesis project, "Prediction of Regional Biocapacity Using Satellite Imagery and Deep Learning," from the University of Khartoum. Our project uses deep learning to analyze and predict the biocapacity and ecological footprint of various regions with satellite imagery, aiming to promote sustainable environmental management.
This project focuses on interpreting multi-spectral satellite images to evaluate the biocapacity and ecological footprints of regions, with a special emphasis on Tuti Island and Port Sudan. By testing three advanced convolutional neural network (CNN) architectures, we achieved over 95% accuracy in classifying various geographical areas. Our developed methodology offers valuable insights into ecological impacts and aids in sustainable management practices.
- Satellite Imagery Analysis: Utilizes multi-spectral data for biocapacity assessment.
- Deep Learning Models: Implements neural networks for accurate predictions.
- Ecological Footprint Estimation: Provides insights into regional ecological impacts.
benchmark: Contains benchmarking scripts and results.biocapacity: Includes data and scripts for biocapacity analysis.
- Clone the repository:
git clone https://github.com/asimzz/prediction-of-regional-biocapacity.git
Our approach involves using deep learning techniques to interpret multi-spectral satellite images. We benchmarked three state-of-the-art CNN architectures: Residual Networks (ResNet), Inception Networks, and Efficient Networks. The models were trained and evaluated using the EuroSAT dataset, achieving high accuracy in classifying geographical regions. The best-performing model was then applied to assess the biocapacity of Tuti Island and Port Sudan.
The models developed in this project achieved over 95% accuracy in classifying different geographical regions using satellite imagery. Detailed results, including performance metrics and biocapacity estimations, are provided in the thesis.pdf document.