🛰️ Satellite-Image-Segmentation

This repository contains an end-to-end pipeline for satellite image segmentation and object extraction from noisy images. The project evaluates multiple segmentation techniques, including K-Means, Mean Shift, and Graph-Based segmentation, followed by region-growing algorithms and connected component analysis (CCA) for refinement.

---

🚀 Project Overview

Satellite images often contain noise, making object extraction challenging. This project implements:

• Noise Reduction using Gaussian & Median Filtering.
• Segmentation Methods: K-Means, Mean Shift, and Graph-Based segmentation.
• Refinement: Region Growing & Connected Component Analysis (CCA).
• Evaluation Metrics: IoU, Dice Score, and Pixel Accuracy.
• Visualization: Comparison of segmentation results at different stages.

---

💂️🏼 Folder Structure

📺 Satellite-Image-Segmentation
│-- 📁 Dataset/                     # EuroSAT dataset (RGB)
│-- 📝 requirements.txt             # Required dependencies
│-- 📝 README.md                    # Project documentation
|--📜 segmentation_notebook.ipynb   # Jupyter Notebook with complete 
                                  implementation 

---

🛠️ Installation & Requirements

pip install -r requirements.txt

1️⃣ Clone the Repository

git clone https://github.com/chethanakantipudi
/Satellite-Image-Segmentation.git
cd Satellite-Image-Segmentation

2️⃣ Install Dependencies

pip install -r requirements.txt

---

🔬 Methods Used

1️⃣ Noise Reduction

• Gaussian Filtering
• Median Filtering

2️⃣ Segmentation Techniques

• K-Means Clustering
• Mean Shift Segmentation
• Graph-Based Segmentation (Felzenszwalb’s Algorithm)

3️⃣ Refinement Techniques

• Region Growing Algorithm
• Connected Component Analysis (CCA) to remove small noisy regions.

---

📊 Evaluation Metrics

• IoU (Intersection over Union)
• Dice Score
• Pixel Accuracy

---

🖼️ Results Visualization

We visualize segmentation results at four stages:

1. Original Image (No Noise Reduction)
2. After Noise Reduction (Gaussian Blur)
3. After K-Means Segmentation
4. Final Refined Segmentation (Region Growing + CCA)

Example visualization:

📜 License

This project is licensed under the MIT License.