118 lines
2.8 KiB
Markdown
118 lines
2.8 KiB
Markdown
# Demo Use Cases
|
|
|
|
These are the workflows GeoIntel must be built around. Codex should prioritize these over generic feature expansion.
|
|
|
|
## Demo 1 — Building Detection + GRB QA
|
|
|
|
### Purpose
|
|
Show the exact portfolio fit for GeoAI engineering: raster input, computer vision, georeferenced detections, reference-data comparison, QA metrics and GIS export.
|
|
|
|
### Area
|
|
Default: Geel or another Kempen area with mixed buildings and roads.
|
|
|
|
### Input
|
|
- Georeferenced aerial image or GeoTIFF.
|
|
- Reference building layer from GRB or local fixture.
|
|
|
|
### Pipeline
|
|
1. Create project.
|
|
2. Define area.
|
|
3. Import raster.
|
|
4. Fetch/cache GRB building reference layer.
|
|
5. Tile raster.
|
|
6. Run YOLO/object-detection provider.
|
|
7. Convert detections back to geospatial polygons or boxes.
|
|
8. Store detections in PostGIS.
|
|
9. Run QA/QC against GRB building footprints.
|
|
10. Show results on map.
|
|
11. Export detections and QA report.
|
|
|
|
### Required outputs
|
|
- Detection layer.
|
|
- GRB reference layer.
|
|
- Matched features layer.
|
|
- False positives layer.
|
|
- False negatives layer.
|
|
- Precision/recall/F1/IoU dashboard.
|
|
- GeoJSON export.
|
|
|
|
### Portfolio message
|
|
“I built a GeoAI pipeline that detects buildings on imagery and validates the results against official Flemish reference geometry.”
|
|
|
|
## Demo 2 — Vegetation Segmentation
|
|
|
|
### Purpose
|
|
Show segmentation and raster/vector conversion.
|
|
|
|
### Input
|
|
- Raster image or Sentinel-derived raster later.
|
|
- Optional area polygon.
|
|
|
|
### Pipeline
|
|
1. Import raster.
|
|
2. Run segmentation provider or thresholded index workflow.
|
|
3. Produce mask.
|
|
4. Polygonize mask.
|
|
5. Calculate area and coverage percentage.
|
|
6. Visualize mask and polygons.
|
|
7. Export GeoJSON/mask.
|
|
|
|
### Required outputs
|
|
- Segmentation mask.
|
|
- Polygonized vegetation layer.
|
|
- Area statistics.
|
|
- Export package.
|
|
|
|
## Demo 3 — Vector Change Detection
|
|
|
|
### Purpose
|
|
Show geospatial comparison before full remote-sensing change detection is implemented.
|
|
|
|
### Input
|
|
- Two reference/detection vector layers for different dates or runs.
|
|
|
|
### Pipeline
|
|
1. Select baseline layer.
|
|
2. Select comparison layer.
|
|
3. Normalize CRS.
|
|
4. Match overlapping objects.
|
|
5. Identify added/removed/changed geometries.
|
|
6. Calculate area deltas.
|
|
7. Show change layer.
|
|
|
|
### Required outputs
|
|
- Added features.
|
|
- Removed features.
|
|
- Changed features.
|
|
- Change statistics.
|
|
- GeoJSON export.
|
|
|
|
## Demo 4 — Raster Indices Preview
|
|
|
|
### Purpose
|
|
Prepare the Sentinel/remote-sensing part.
|
|
|
|
### Input
|
|
- Multiband raster with suitable bands.
|
|
|
|
### Pipeline
|
|
1. Select bands.
|
|
2. Calculate NDVI/NDWI/NDBI where bands exist.
|
|
3. Render preview.
|
|
4. Calculate min/max/mean/histogram.
|
|
5. Extract thresholded areas later.
|
|
|
|
### Required outputs
|
|
- Index raster artifact.
|
|
- Histogram/statistics.
|
|
- Preview overlay.
|
|
|
|
## Demo priority
|
|
Build in this order:
|
|
1. Demo 1 first.
|
|
2. Demo 3 second.
|
|
3. Demo 2 third.
|
|
4. Demo 4 fourth.
|
|
|
|
Demo 1 is the main recruiter-facing workflow.
|