An engine for analyzing and mapping long-term glacier retreat patterns using satellite imagery. This project provides a standalone application for analyzing glacier disintegration using classic computer vision and machine learning techniques.
This project is a computer vision tool built with PyQt6 for analyzing glacier satellite imagery. It allows users to load their own pre-trained machine learning models (.pkl files) to perform two primary tasks:
- Glacier Retreat Classification: Predicts if a glacier in an image is Stable or Retreating based on a comprehensive set of 89 extracted image features.
- Image Segmentation: Classifies each pixel in an image as Ice, Water, or Other using image features to generate a detailed segmentation map and statistics.
The application is designed to be user-friendly, allowing researchers to apply their own models to new imagery without complex scripting. It runs analysis in the background using QThread to keep the UI responsive.
- PyQt6 GUI Application
- Standalone desktop tool for Windows, macOS, and Linux
- Tabbed interface for Retreat Classification and Image Segmentation
- Allows loading of custom joblib (
.pkl) models, scalers, and feature lists - Asynchronous model execution using
QThread - Interactive visualization panel using Matplotlib
- Predicts if a glacier is Stable or Retreating
- Extracts 89 features (spectral, texture, edge, morphological, statistical, and gradient)
- Displays prediction results with probabilities and feature values
- Exports results as PNG or CSV
- Classifies each pixel as Ice, Water, or Other
- Uses a 9-feature extraction pipeline per pixel (RGB, local means, brightness, color ratios)
- Provides visual overlays and class statistics
- Exports visualization as PNG or segmentation map as .npy
Requires Python 3.8+ and the following libraries:
PyQt6
matplotlib
numpy
pandas
rasterio
joblib
scikit-image
scipy
Pillow# Clone repository
git clone https://github.com/Manjushwarofficial/Glacier-Probe-Model.git
cd Glacier-Probe-Model
Ensure you have the following pre-trained .pkl models:
glacier_retreat_model_svm_(rbf).pklglacier_retreat_scaler.pklglacier_retreat_features.pklglacier_segmentation_model.pklglacier_segmentation_scaler.pkl
python main.py- Click Get Started on the welcome screen
- In Model Configuration, use Browse... to select model files
- Click Load Models
- Go to Single Image Prediction or Image Segmentation tab
- Click Load Image and choose
.tif,.jpg, or.png - Click Predict Retreat Status or Run Segmentation
Glacier-Probe-Model/
├── glacier_probe/
│ ├── __init__.py
│ ├── preprocessing/
│ │ ├── atmospheric_correction.py
│ │ ├── cloud_masking.py
│ │ └── coregistration.py
│ ├── features/
│ │ ├── spectral_indices.py # NDSI, NDWI, NDVI calculation
│ │ ├── texture_features.py # GLCM implementation
│ │ └── terrain_features.py # Slope, aspect from DEM
│ ├── models/
│ │ ├── random_forest.py # RF classifier
│ │ ├── unet.py # U-Net architecture
│ │ └── train.py # Training scripts
│ ├── detection/
│ │ ├── threshold_detection.py # NDSI thresholding
│ │ ├── ml_detection.py # ML-based detection
│ │ └── dl_detection.py # DL-based detection
│ └── visualization/
│ ├── temporal_plots.py
│ └── interactive_maps.py
├── data/
│ ├── annotatios/
│ ├── processed/
│ ├── raw/
│ ├── reference/
│ └── README.MD
├── models/
│ ├── .ipynb_checkpoints/
│ ├── glacier_classifications.pkl
│ ├── glacier_retreat_features.pkl
│ ├── glacier_retreat_model_svm_(rbf).pkl
│ ├── glacier_retreat_scaler.pkl
│ ├── glacier_segmentation_model.pkl
│ └── glacier_segmentation_scaler.pkl
├── notebooks/
│ ├── 01_data_exploration.ipynb
│ ├── 02_ndsi_baseline.ipynb
│ ├── 03_ml_training_classification.ipynb
│ └── 04_ml_training_semantic_segmentation.ipynb
│ ├── 04_dl_refinement.ipynb
│ └── 05_temporal_analysis.ipynb
├── scripts/
│ ├── download_data.py # GEE/Sentinelsat data download
│ ├── ml_detection.py
│ ├── dl_segmentation.py
│ └── generate_report.py
├── tests/
│ ├── test_preprocessing.py
│ ├── test_features.py
│ └── test_models.py
├── configs/
│ ├── ml_config.yaml # ML hyperparameters
│ └── dl_config.yaml # DL training config
├── models/ # Saved model checkpoints
├── results/ # Detection outputs
├── reports/ # Generated analysis reports
├── requirements.txt
├── .gitignore
├── LICENSE.md
└── README.md
- Spectral Features: Mean/std of bands, color ratios, brightness, saturation
- Edge Features: Sobel & Canny edge stats
- Texture Features: GLCM, LBP, entropy
- Morphological Features: Area, perimeter, eccentricity, solidity
- Statistical Features: Skewness, kurtosis
- Gradient Features: Gradient mean, std, direction
- RGB values
- Local mean RGB (7x7 window)
- Brightness
- Color ratios (B/R, G/R)
- Classification: Accuracy, Precision, Recall, F1 Score
- Segmentation: Pixel Accuracy, IoU, Dice Coefficient
Contributions are welcome!
-
Fork the repository
-
Create a new branch:
git checkout -b feature/new-feature
-
Commit your changes and submit a PR
@software{glacier_probe_model_app_2025,
author = {Manjushwar},
title = {Glacier Probe Model: Application for Glacier Analysis},
year = {2025},
url = {https://github.com/Manjushwarofficial/Glacier-Probe-Model}
}Licensed under the MIT License. See LICENSE.md for details.
- ESA Copernicus Programme for Sentinel-2 data
- USGS for Landsat Collection 2 data
- NSIDC for Randolph Glacier Inventory
- GLIMS community for glacier outline validation
- Core Application Framework (PyQt6)
- Classification Pipeline (89 features)
- Segmentation Pipeline (9 features)
- Asynchronous Execution (QThread)
- Result Visualization (Matplotlib)
- Export (PNG, CSV, NPY)
Status: Core application with classification & segmentation is complete. Batch processing UI is present but not yet implemented.