ML_geo_production

ML_geo_production is a geospatial ML pipeline that runs one or more PyTorch semantic-segmentation models over GeoTIFF imagery, ensembles their outputs, and produces combined, normalized probability arrays for areas of interest. It processes only the area of the raster that intersect a defined bounding box, and can optionally compare predictions with polygon data (GeoPackage) for change detection.

For questions about the repo, email rajoh@kds.dk

---

Features

• Model ensembling for multiple PyTorch semantic-segmentation models\
• Processes only GeoTIFF regions that intersect the area of interest\
• Outputs a single combined & normalized probability array for the AOI\
• Optional comparison against a GeoPackage of polygons for change detection\
• Example configs work with [https://github.com/SDFIdk/multi_channel_dataset_creation]dataset and pretrained models

---

Quick Start (step-by-step)

1. Install this repository

mamba env create -f environment.yml
mamba activate ML_geo_production
pip install -e .

---

2. Clone the example dataset repo side-by-side

Clone the dataset repository so it sits next to this repository in the same parent directory:

git clone https://github.com/SDFIdk/multi_channel_dataset_creation

The example config files in config_files/ work out of the box with the dataset included in that repository.

---

3. Download example models

python src/ML_geo_production/download_example_models.py

Note:
The example models were trained using the training code from:
https://github.com/SDFIdk/ML_model_training

---

Basic Usage

Main processing script:

python src/ML_geo_production/process_images.py --json config_files/save_probs_preds_and_change_detection.json

Example above use an ensamble of three models, save both probs, preds and change detection.

Example workflows:

python src/ML_geo_production/do_change_detection.py --json config_files/change_detection.json
python src/ML_geo_production/create_prediction_raster.py --json config_files/raster_production.json
python src/ML_geo_production/process_many_areas.py --json config_files/process_many_areas.json  --shapefile ../multi_channel_dataset_creation/example_dataset/shape_files/many_areas.shp

The process_many_areas.py example above shows an example of how to process many areas after each other.

---

Config Files

See examples in config_files/ --- these are ready to run using the dataset from
https://github.com/SDFIdk/multi_channel_dataset_creation

---

Inputs & Requirements

• GeoTIFFs must share a compatible CRS\
• Models must output probability tensors\
• Optional polygon layers must match the CRS (or will be reprojected automatically)

---

Outputs

• Combined, normalized probability arrays\
• Prediction raster outputs\
• Change-detection results when polygon comparison is enabled

---

License (MIT)

MIT License

Copyright (c) 2025

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.