AstroClusterModel: Clustering Astronomical Images

AstroClusterModel is a machine learning pipeline that clusters astronomical FITS images based on structural and compositional features like planetary blobs, ring structures, and radial intensity profiles. It combines feature engineering, dimensionality reduction (UMAP/PCA), and clustering (K-Means) to uncover meaningful groupings in space imagery.

---

Computer Vision Techniques used:

1. Radial Profile Analysis

   

2. Blob detection using Laplacian of Gauss

   

3. Canny Edge Detection

4. Hough Transform

   

Overview

• Extracts 52 handcrafted features: radial, blob, and ring features.
• Reduces dimensions using UMAP/PCA.
• Clusters using K-Means.
• Provides cluster interpretations and sample images.

📸 Sample Astronomical Images

---

📦 Pipeline Summary

🔹 1. Feature Extraction

Feature Group	Description
Radial	Mean, std, peaks, zero-crossings in radial intensity profiles
Blob	Count, average size, and intensity of blobs (planet-like regions)
Ring	Count, radius, and concentricity of rings via Hough Circle Transform

---

🔹 2. Dimensionality Reduction

Technique	Purpose
UMAP	Non-linear reduction for better visual cluster separation
PCA	Linear reduction for easier interpretation

---

🔹 3. Clustering

• Uses K-Means on reduced features.
• Automatically interprets clusters using statistical summaries.
• Displays sample images from each cluster.

---

📊 Results

📌 Cluster Distribution

🌌 2D Visualization (UMAP)

---

🧪 Cluster Summaries

🌀 Cluster 0

---

🪐 Cluster 1

---

🌗 Cluster 2

---

🔑 Project Use Cases

• Automatically categorize thousands of astronomical images without manual inspection
• Identify exoplanetary systems with similar structural patterns for comparative research
• Flag statistical outliers that may represent new astronomical phenomena or instrument errors
• Provide quantitative metrics for comparing morphological features across star systems