Advanced Earth observation & image processing
Our research lab pushes how neural networks interpret planetary data. We re-engineer transformers (SegFormer), zero-shot foundation models (SAM) and 3D point-cloud architectures (PointNet++) to extract geometric and material insight from space, aerial and land.
- SAR processing: speckle-resilient backbones that read radar phase data in all weather.
- Hyperspectral cubes: 3D-CNNs and spectral-spatial transformers that isolate chemical and biological signatures across hundreds of bands.
- Multispectral fusion: high-speed edge delineation for ultra-high-resolution optical data.
