NovaDetect: Lunar Landslide & Boulder Detection using Chandrayaan Imagery

"Decoding the Moon, one slope at a time." NovaDetect is building a novel AI-powered system to detect lunar landslides and boulders using Chandrayaan imagery, combining vision, terrain modeling, and geospatial insights for safer space exploration.

Pipeline

---

Overview

NovaDetect is an end-to-end AI-powered pipeline that detects landslides and boulders on the Moon using images from Chandrayaan-1 and 2 missions. The solution leverages:

• Deep learning (U-Net CNN) for landslide segmentation
• Terrain-aware computer vision with LoG + DTM fusion for boulder detection

Our aim is to automate the detection process across diverse terrains on the Moon with high precision, scalability, and geospatial interpretability.

---

Core Features

Landslide Detection (U-Net CNN)

• Trained on TMC images + manually annotated masks
• Learns spatial patterns of landslides (irrespective of lighting/slope)
• Outputs:
  • Annotated landslide masks
  • CSV with bounding box coordinates & timestamps

Boulder Detection (LoG + DTM Fusion)

• Uses blob detection + DTM-based elevation filtering
• Extracts:
  • Location (X, Y)
  • Estimated length, diameter
  • Elevation and slope features (if DTM is used)
• Outputs:
  • Annotated images with overlays
  • CSV of boulder statistics

---

Key Innovations

• AI-Terrain Fusion: First to combine U-Net segmentation with DTM geometry for lunar surface analysis
• Smart Data Alignment: Regex-based timestamp matching ensures accurate image-mask pairing
• DTM-Enhanced Filtering: Uses slope, elevation, and local variance to eliminate false positives and prioritize geologically significant boulders
• Universal Scalability: Location-independent system works across all Chandrayaan-covered lunar regions

---

Technologies Used

Category	Tools & Libraries
Language	Python
Deep Learning	TensorFlow, Keras
Image Processing	OpenCV, scikit-image, Matplotlib
Data Handling	NumPy, Pandas
DTM Handling	Rasterio (optional, for elevation data)
GIS Tools	QGIS (for visualization & validation - optional)

---

Output Files

Type	Description
.png	Annotated predictions (landslides/boulders)
.csv	Geometric & statistical info of detected features
.h5	Trained U-Net model for landslide detection

---

Sample Outputs

Landslide Detection:

• Mask overlay showing segmented landslide regions
• CSV: timestamp, bounding box coordinates

Boulder Detection:

• Image with red/cyan circular overlays for boulders
• CSV: X, Y, radius, estimated length/diameter, elevation info (if DTM used)

---

How to Run

1. Clone the repository

git clone https://github.com/shreya-13-04/NovaDetect.git
cd NovaDetect

2. Install Dependencies

pip install -r requirements.txt

3. Landslide Detection

python train_landslide_model.py     # Train U-Net model
python predict_landslides.py image.png   # Predict on new image

4. Boulder Detection (with/without DTM)

# Run in Google Colab (recommended)
# Upload image and optionally mount Google Drive for DTM

---

Dataset Sources

• TMC (Terrain Mapping Camera) – Chandrayaan-1 & 2
• OHRC (Optical High-Resolution Camera) – Chandrayaan-2
• DTM – Derived from Chandrayaan stereo imagery

---

Team Members

• Harish Suresh
• Pooja Shree S
• Arjun K
• Shreya B

---

Research Impact

This pipeline advances lunar geological research by providing automated, high-precision detection of critical surface features with unprecedented accuracy and scope. The fusion of AI with planetary data opens new possibilities for understanding lunar surface dynamics and geological processes.

---

Built with ❤️ for lunar exploration and planetary science