Deep Learning Engineer — Satellite Image Classification

We're building a land-use change detection system using multispectral Sentinel-2 imagery at global scale. We need to classify land use across 6 categories (forest, cropland, urban, water, grassland, bare soil) and detect changes between time periods. Current approach is a U-Net variant — working but underperforming on cloud-contaminated images and class boundaries. We need someone to improve the architecture and training pipeline, handle missing-data imputation for cloudy pixels, and scale inference to continental coverage. Expected output: improved model, training pipeline, and an inference service that can process a full country in under 24 hours on our cloud setup.