Auto Satellite Detection Prototype

Automated Satellite Image Object Detection & Labelling — Prototype

Step 1 — AOI Selection

Draw a polygon/rectangle on the map to define the Area of Interest.

AOI GeoJSON (editable):

Step 2 — Load Satellite Image

You can upload a local image (JPEG/PNG) or provide an image URL (WMS/tiles require server-side CORS tokens).

Upload image Or provide image URL (HTTP/S)

Optional: Provide georeference (Top-left lat,lng and Bottom-right lat,lng) — used for GeoJSON export

Step 3 — Preprocess & Inspect

Preview and tile the image (tiling helps with large images).

Max dimension for inference (px) Tile size (optional, px)

Step 4 — Object Detection (Browser)

This prototype uses coco-ssd via TensorFlow.js for demonstration. For production, run server-side Detectron2 / YOLOv8 and custom-trained models.

Model not loaded.

Step 5 — Review / Edit Labels

Accept / reject detections, add manual boxes by dragging on the image.

Step 6 — Export Labels & Dataset Export format COCO (JSON) GeoJSON

Notes & Next steps

• Production deployments should use server-side models (Detectron2 / YOLOv8) and GPU instances.
• To fetch Sentinel/Planet/Maxar images automatically, configure API keys and a backend proxy to handle CORS and licensing.
• This prototype demonstrates the human-in-the-loop annotation + browser inference flow and label export for training.