🌍 GTC-Land-Type-Classification

Project 14: Land Type Classification using Sentinel-2 Satellite Images

---

🔍 Project Overview

This project develops a machine learning model to classify land cover types in Egypt using Sentinel-2 satellite imagery.
The system can detect categories such as:

• 🌾 Agriculture
• 💧 Water bodies
• 🏙️ Urban areas
• 🏜️ Deserts
• 🛣️ Roads
• 🌳 Trees

The workflow includes data preparation, exploratory analysis, model training & validation, and deployment via a lightweight web app.
Final deliverables include a trained model, performance report, and an interactive application for testing land classification.

---

👥 Collaborators & Responsibilities

Task	Collaborator
Data Pipeline, EDA & Feature Engineering	Roaa Raafat
Model Architecture & Evaluation	Mariam Mohamed Sayed Mohamed
Deployment & Dashboard	Mariam Badr
Video & GitHub README	Eman Elnaggar & Mariam Elnemrawy
Presentation	Mariam Ahmed

---

📦 Dataset

EuroSAT Dataset (Kaggle)

---

🛠️ Workflow

1- Data Pipeline

• Notebook: GTC_01_data_pipeline.ipynb
• Script: src/data_pipeline.py
• Tasks:
  • Downloads & organizes the EuroSAT dataset
  • Cleans CSVs and ensures consistent labels
  • Generates dataset statistics & class maps
  • Outputs train/validation/test splits

2- Exploratory Data Analysis

• Notebook: GTC_02_eda_visualization.ipynb
• Tasks:
  • Class distribution plots
  • Sample images per class (grid view)
  • Checks for corrupted/missing/duplicate images
  • RGB intensity histograms
  • Data augmentation visualizations
  • Class-level mean RGB statistics
  • Feature extraction (color histograms, texture descriptors)

3- Feature Engineering

• Script: src/features.py
• Tasks:
  • Geospatial-safe augmentations (rotation, flip, jitter, crop/zoom)
  • Normalization & preprocessing utilities

4- Modeling & Evaluation

• Script: gtc_03_modeling-6.ipynb
• Tasks:
• Load and preprocess the EuroSAT dataset (RGB satellite images for 10-class land use classification) with custom transforms including resizing to 224x224, data augmentation (random flips, rotations, jitter, affine), and normalization (mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]).
• Split dataset into train/validation/test (60%/20%/20%) and create DataLoaders with batch size 64.
• Define and train multiple models: a custom CNN (VGG-inspired) and transfer learning models (ResNet-50, EfficientNet-B0, Vision Transformer B-16), all fine-tuned for 10 classes.
• Train with Cross-Entropy loss, Adam optimizer (lr=0.001, weight decay=1e-4), mixed precision (AMP), LR scheduler (ReduceLROnPlateau), and early stopping (patience=10).
• Evaluate models on test set with accuracy, precision/recall/F1, confusion matrix, and training history plots.
• Save best model checkpoints based on validation accuracy.

---

📤 Outputs Generated

• train.csv, validation.csv, test.csv (cleaned datasets)
• dataset_statistics.json & plots (dataset_statistics.png, sample_images.png)
• augmentation_examples.png
• Engineered features in data/features/
• Trained model in models/best_efficientnet_b0_model.pth

---

⚡ Quick Start (Colab)

### 1- Clone the repository
!git clone https://github.com/Mariam-Badr-MB/GTC-Land-Type-Classification.git
%cd GTC-Land-Type-Classification

### 2- Install dependencies
!pip install -r requirements.txt

### 3- Configure Kaggle API
from google.colab import files
uploaded = files.upload()  # upload kaggle.json

!mkdir -p ~/.kaggle
!cp kaggle.json ~/.kaggle/
!chmod 600 ~/.kaggle/kaggle.json

4- Run Notebooks

• GTC_01_data_pipeline.ipynb
• GTC_02_eda_visualization.ipynb
• gtc-03-modeling-6

---

View Presentation

🎨 Project Presentation on Canva

---

Video

Video

📬 Contact

• Mariam Badr Yehia:

• Mariam Mohamed Sayed Mohamed:

• Mariam Mohamed Elnemrawy:

• Mariam Ahmed Mohamed:

• Eman Yasser Elnaggar:

• Roaa Raafat Said:

🤝 Contributing

Contributions are welcome! Feel free to fork the repo, create a new branch, and submit a pull request.

📜 License

This project is licensed under the GNU v3 License.