July 2025 – Present | Brandenburgische Technische Universität Cottbus-Senftenberg
Ongoing research project within the MotionCB initiative at BTU Cottbus-Senftenberg. I am building deep learning pipelines to automatically detect and map large-scale photovoltaic (solar) farms across the whole of Germany using high-resolution Sentinel-2 satellite imagery and ancillary geospatial data.
Key contributions:
- Designed end-to-end workflows for downloading, preprocessing, and tiling multi-spectral satellite data
- Experimenting with state-of-the-art segmentation models (U-Net, DeepLabV3+, SegFormer) and self-supervised pre-training on remote sensing data
- Implementing multi-temporal analysis to reduce false positives caused by seasonal changes and cloud cover
- Creating vectorized geo-referenced masks of detected solar installations for downstream renewable energy monitoring
The project directly supports Germany’s Energiewende by providing up-to-date, nationwide maps of solar infrastructure growth.
Tech: Python, PyTorch, Rasterio, GDAL, GeoPandas, Jupyter, Git
