Left: Raw OHRC telemetry of a Permanently Shadowed Region (pitch black). Right: AETHER-enhanced output revealing hidden crater topography.
AETHER is a deep learning framework designed to solve one of the most significant challenges in planetary science: visualizing the Permanently Shadowed Regions (PSRs) of the Moon. These areas, which haven't seen sunlight in billions of years, are prime locations for water ice but are notoriously difficult to image. Using a rich suite of data from ISRO's Chandrayaan-2 mission, AETHER employs Zero-Reference Deep Curve Estimation (Zero-DCE) to enhance faint signals, transforming noisy, low-light patches into scientifically valuable maps without synthesizing fake details.
- Zero-Reference Enhancement: Utilizes a highly efficient, lightweight CNN to dynamically estimate pixel-wise exposure curves, completely eliminating the need for paired ground-truth lunar imagery.
- Multi-Modal Data Fusion: Intelligently integrates optical (OHRC), radar (DFSAR), elemental (CLASS), and spectral (IIRS) data to create a holistic and information-rich view of the lunar surface.
- Scientific Integrity: Rejects hallucination-prone generative models in favor of physically grounded exposure scaling combined with Non-Local Means (NLM) spatial denoising, maximizing the Signal-to-Noise Ratio (SNR).
- Mission-Ready Outputs: Produces high-resolution, georeferenced polar maps critical for landing site selection, resource prospecting, and geomorphological studies.
AETHER is built with a modern, research-focused tech stack designed for high-performance computing and reproducibility.
- Core Framework: PyTorch for its flexibility and dynamic computation graphs.
- Data Handling: GDAL & Rasterio for geospatial data processing, Pandas for metadata management.
- Image Processing: OpenCV and Scikit-Image for preprocessing and classical enhancement algorithms.
- Models & Training: Hugging Face Accelerate for distributed training and Weights & Biases for experiment tracking.
- Hardware: Designed for NVIDIA GPUs using CUDA for accelerated model training.
Our pipeline transforms raw, heterogeneous data into clear, scientifically-valuable maps. The cornerstone of AETHER is its two-phase training curriculum designed specifically for the extreme noise floor of lunar PSRs.
We treat the lunar surface as a multi-layered dataset. Each instrument provides a unique layer of information, and by combining them, we create a "super-image" that offers far more context than a simple photograph.
- 🛰️ Primary Optical Data (OHRC): This is the high-resolution visual canvas we aim to enhance.
- 🪨 Structural & Texture Data (DFSAR): The radar data acts as our structural blueprint, revealing physical topography unaffected by shadows.
- 🧪 Elemental Composition Data (CLASS): Provides the chemical fingerprint, mapping elements like Silicon, Iron, and Magnesium.
- 💧 Mineralogical & Volatile Data (IIRS): The infrared spectrometer identifies specific minerals and the spectral signature of water ice.
- 🗺️ Geometric & Navigational Data (SPICE): SPICE kernels align all layers perfectly into a single, cohesive coordinate system.
Because true "bright" ground-truth images of PSRs do not exist, we simulate them. We take sunlit OHRC patches, synthetically darken them, and inject them with mixed Poisson-Gaussian noise to match the sensor's noise floor. The model is supervised on these synthetic pairs to learn basic feature recovery.
The pre-trained model is fine-tuned on actual pitch-black PSR patches. Using mathematical loss functions (Spatial Consistency, Exposure Control, and Illumination Smoothness), the model learns to enhance the true secondary scattered light without relying on a reference image.
The massive OHRC images are processed in overlapping tiles to manage memory. Because enhancing a faint signal amplifies existing sensor noise, the raw Zero-DCE output is passed through a Non-Local Means (NLM) spatial denoiser and CLAHE to clean the image and reveal the physical topography before final map synthesis.
AETHER directly contributes to the next era of lunar exploration by:
- Enabling Safer Landings: Providing mission planners with clear, detailed views of potential landing sites for robotic and crewed missions.
- Accelerating Scientific Discovery: Unlocking vast, unexplored regions for geological analysis and the search for water ice.
- Advancing Planetary AI: Pioneering the use of zero-reference techniques for low-data, extreme-environment scenarios in space exploration.
aether/
│
├── data/ # Utilities for ISRO PRADAN data scraping and bulk downloading
├── notebooks/ # Jupyter notebooks for Exploratory Data Analysis (EDA)
├── models/ # Pre-trained model weights (Zero-DCE)
│
├── src/ # 🔒 Core Research Modules (Restricted for Open Source PRs)
│ ├── data_io/ # PDS4 XML metadata parsing and raw .img ingestion
│ ├── models/ # Deep learning architectures (Zero-DCE, NLM Denoiser)
│ └── projection/ # GDAL/Rasterio geospatial mapping and map synthesis
│
├── scripts/ # Executable pipeline scripts
│ ├── extract_patches.py # Sliding-window patch extraction and categorization
│ └── infer.py # Model inference, NLM spatial denoising, and blending
│
├── tests/ # Unit and integration tests (pytest)
├── CONTRIBUTING.md # Vital guidelines, IP rules, and GSSoC roadmap
├── requirements.txt # Project dependencies
└── README.md # Project documentation
- Python 3.10+
- Anaconda or Miniconda
- NVIDIA GPU with CUDA 11.8+
- Clone the repository:
git clone [https://github.com/TechieSamosa/AETHER.git](https://github.com/TechieSamosa/AETHER.git)
cd AETHER
- Create and activate a conda environment:
conda env create -f environment.yml
conda activate aether
- Run inference on a sample image:
python scripts/enhance.py --input /path/to/sample.img --output /path/to/enhanced.png --model models/aether_v1.pth
We are thrilled to be a part of GirlScript Summer of Code 2026! We welcome contributions from the global community, specifically focusing on data engineering, memory optimization, and pipeline automation.
The AETHER framework is the foundation of an upcoming scientific research paper. The core mathematical architecture and deep learning models in the src/ directory are strictly locked. Before making any pull requests, you must read our CONTRIBUTING.md to understand the strict boundaries between open-source engineering contributions and academic co-authorship. By interacting with this repository, you automatically agree to these terms.
- Read the Docs: Review our Contributor Guide & Technical Roadmap to understand how we source Chandrayaan-2 data.
- Find an Issue: Look for issues tagged
good first issue,gssoc-26, orhelp wanted. - Focus Areas: We are actively looking for PRs related to:
- Automated web-scraping for ISRO PRADAN data.
- PyTorch DataLoader memory optimization.
- GIS map stitching using GDAL/Rasterio.
- Unit testing and code documentation.
This work builds upon the state-of-the-art in deep learning and planetary science. We gratefully acknowledge the data provided by the ISRO Science Data Archive (ISDA).
- "Zero-Reference Deep Curve Estimation for Low-Light Image Enhancement" (Guo et al., 2020)
- "Peering into lunar permanently shadowed regions with deep learning" (Bickel et al., 2021)
- Chandrayaan-2 Mission Data Handbook, ISRO
- "Analysis of Permanently Shadowed Regions of the Moon using LRO and Chandrayaan-2 Data"
AETHER’s core architecture, deep learning methodology, and academic research are developed and maintained by:
- Aditya Sachin Khamitkar (Lead Architect & Researcher)
- Tushar (Core Researcher)
- Nitin (Core Researcher)
A huge thank you to the global developers and GSSoC '26 mentees who have optimized our pipelines, improved our data ingestion, and helped scale AETHER! Impactful engineering contributions are automatically recognized here.
| Contributor (GitHub Handle) | Tier | Specific Contribution |
|---|---|---|
@example-contributor |
🥇 Gold | Implemented automated PRADAN data scraper, reducing manual download time by 95% |
How to get listed: Make a meaningful, merged PR. Tiers are assigned by the core team based on impact: 🥇 Gold (architectural/high-impact), 🥈 Silver (significant feature/optimization), 🥉 Bronze (bug fixes, docs, tests).
This project is licensed under the MIT License. See the LICENSE file for details. This permissive license allows for wide use and collaboration.