A remote-sensing workflow for extracting reliable signals from multi-temporal satellite imagery under cloud cover, alignment, and scene-quality constraints.

What I built

  • Image ingestion, best-scene selection, and cloud-masking workflows for Sentinel-2 data
  • Preprocessing for band alignment and repeatable multi-temporal analysis
  • Detection and change-detection experiments using MMDetection and Open-CD model families
  • Scalable inference workflows suitable for larger remote-sensing batches

Outcome

  • Reduced noisy scene selection overhead before model training and inference
  • Improved the repeatability of multi-temporal analysis workflows
  • Created a practical path from geospatial preprocessing to model-based detection tasks

Stack

Sentinel-2, remote sensing, object detection, change detection, MMDetection, Open-CD, and preprocessing pipelines