https://github.com/who-else-but-arjun/isro_xrf_sr
Source Codes for super resolution of the lunar elemental abundance map using a semi-supervised deep spatial interpolation model. This hybrid approach combined ResNet50 for spatial feature extraction with Graph Neural Network (GATv2Conv) layers and Convolutional Neural Networks (CNNs), followed by fusion layers.
https://github.com/who-else-but-arjun/isro_xrf_sr
cnn data-analysis graph-neural-networks pytorch semi-supervised-learning spatial-interpolation super-resolution
Last synced: about 2 months ago
JSON representation
Source Codes for super resolution of the lunar elemental abundance map using a semi-supervised deep spatial interpolation model. This hybrid approach combined ResNet50 for spatial feature extraction with Graph Neural Network (GATv2Conv) layers and Convolutional Neural Networks (CNNs), followed by fusion layers.
- Host: GitHub
- URL: https://github.com/who-else-but-arjun/isro_xrf_sr
- Owner: who-else-but-arjun
- Created: 2024-12-20T20:31:24.000Z (5 months ago)
- Default Branch: main
- Last Pushed: 2024-12-20T21:03:07.000Z (5 months ago)
- Last Synced: 2025-02-14T14:57:40.865Z (3 months ago)
- Topics: cnn, data-analysis, graph-neural-networks, pytorch, semi-supervised-learning, spatial-interpolation, super-resolution
- Language: Python
- Homepage: https://drive.google.com/file/d/12J4m65JyD-cekRiMmBzjrCBWQKoxNWYq/view?usp=drive_link
- Size: 50.8 KB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Super-Resolution of Lunar XRF Elemental Abundances
This repository contains code developed for super-resolution by deep spatial interpolation model of the mapping of lunar surface elemental abundances obtained from Chandrayaan-2 CLASS instrument XRF data. The detailed description of the project is avaiable in this report (https://drive.google.com/file/d/12J4m65JyD-cekRiMmBzjrCBWQKoxNWYq/view?usp=drive_link).
## Project Structure
### Main Components:
1. **file_initialisation.py**:
- Initializes the 64 subregion CSV files for the Moon's surface.
- For each subregion, computes the latitudes and longitudes of 2 km x 2 km pixels.
- Extracts features from image tiles using a pre-trained ResNet50 model.
- Populates the CSV files with selected features, mare/highland classifications, and metadata.
2. **top_k_features.ipynb**:
- Analyzes feature importance from ResNet50 extractions.
- Identifies the top 300 features for further processing based on variance thresholding.
3. **Final.py**:
- Combines all parts of the project.
- Constructs subgraphs for abundance maps.
- Train the GNN-CNN combined model on these subgraphs.
- Returns the final high resolution abundance maps.
### Prerequisites :
- Python 3.10+
- GPU with CUDA support (optional but recommended).
- Clone the repository:
```bash
git clone https://github.com/who-else-but-arjun/ISRO_XRF_SR.git
cd ISRO_XRF_SR
```
- Install dependencies:
```bash
pip install -r requirements.txt
```
- **Shapefile for Mare Regions**: `LROC_GLOBAL_MARE_180.SHP` ([https://drive.google.com/file/d/12J4m65JyD-cekRiMmBzjrCBWQKoxNWYq/view?usp=drive_link](https://drive.google.com/drive/folders/1Gpget_fLbG4ElaxwJFbuMO90GMVLFbqo?usp=sharing)).
- **Lunar Surface Image**: `Lunar_LRO_LROC-WAC_Mosaic_global_100m_June2013.tif` (https://planetarymaps.usgs.gov/mosaic/Lunar_LRO_LROC-WAC_Mosaic_global_100m_June2013.tif).
- **CSV Files**: Intermediate outputs from `file_initialisation.py`.
Place the downloaded files in their respective directories as outlined below:
- `LROC_GLOBAL_MARE_180.SHP`: `./Mare Classification/`
- `Lunar_LRO_LROC-WAC_Mosaic_global_100m_June2013.tif`: Root directory.
```
.
├── file_initialisation.py # Initialization and feature extraction.
├── top_k_features.ipynb # Feature analysis and selection.
├── Final.py # Full pipeline integration.
├── Lunar_LRO_LROC-WAC_Mosaic_global_100m_June2013.tif
├── requirements.txt # Dependency file.
├── Mare Classification/ # Directory for shapefiles.
| - LROC_GLOBAL_MARE_180.DBF
| - LROC_GLOBAL_MARE_180.PRJ
| - LROC_GLOBAL_MARE_180.SHP
| - LROC_GLOBAL_MARE_180.SHP.XML
| - LROC_GLOBAL_MARE_180.SHX
| - LROC_GLOBAL_MARE_README.TXT
├── PART2Output.npz # Stores updated subregions
├── graphs/
├── masks/
└── regions/ # Outputs of file_initialisation.py.
```
### Outputs:
- **regions CSVs**:
- Files named `subregion__.csv`.
- Columns include:
- `lat_center` and `lon_center`: Geographic coordinates of the pixel centers.
- Elemental abundances: Features like `Fe`, `Ti`, `Ca`, etc.
- `mareOrHighland`: Binary classification indicating whether the region is Mare or Highland.
- Top 300 spatial features: Feature columns extracted from lunar rgb images and top 300 selected features based on variance thresholding.
- **Graphs and their corresponding masks**:
- Structured inputs for graph neural network models.
- Encodes spatial and feature-based relationships within subregions.
- Graphs stored in `graphs\graphs_subregion__\subgraph_{i}_{j}_{row_idx}_{col_idx}.pt`.
- Masks for the graphs stored in `masks\masks_subregion__\mask_{i}_{j}_{row_idx}_{col_idx}.pt`.
- **Final High-Resolution Maps**:
- Enhanced elemental abundance data for each subregion.
- Saved as updated `subregion__.csv` files for mapping.
---
## Running the Project
### Step 1: Run file_initialisation.py
This script initializes and populates the subregion CSV files.
```bash
python PART1.py
```
### Step 2: Run the Final Code
The `final.py` script combines all components and generates the high-resolution abundance maps.
```bash
python Final.py --mode 1
```
mode = 1 for population of initialised csv files with the input elemental abundances data.
```bash
python Final.py --mode 2
```
mode = 2 for create the graphs for each subregion and training and interpolation for final high resolution abundnaces mapping.
---