CoastalPulse — Coastal Asset Damage Assessment Dashboard

A full-stack GIS tool: upload aerial/satellite imagery, get automatic object detection where a free pretrained model can do it, manually annotate where it can't, and see class-wise cost/area/ count analytics plus a KDE hotspot heatmap on a satellite map.

No demo/seed data. Every project starts empty. Everything on the dashboard comes from images you actually upload and objects that are actually detected or annotated.

---

---

---

---

---

What's real and working right now

• Automatic detection for "Vessel": a pretrained YOLOv8n model (ONNX, CPU-only, no CUDA/GPU needed) runs on upload. COCO's 80 classes include "boat", so vessel detection is genuinely automatic — no training required. This was verified end-to-end during development: upload → detect → geo-referenced polygon → dashboard, all live, no mocking.
• Manual annotation canvas for everything else (Dock, Seawall, Ramp, or any custom class): drag to draw a box on the uploaded image, pick a class, save. Also verified working end-to-end.
• Geo-referencing: when you upload an image you can set its corner lat/lon coordinates. Every detection/annotation on that image gets converted from pixel coordinates to a real GeoJSON polygon (centroid + area in sq ft) using that bounding box. Images without bounds can still be annotated but won't appear on the map or in dashboard totals — the UI tells you this.
• Dashboard: cost bar chart, area distribution chart, count donut chart, all reading live aggregated data from the backend, filterable by All/Residential/Commercial.
• Hotspot map: Leaflet + a real KDE heatmap (leaflet.heat) with working Blur/Radius sliders.
• Multi-project support: create, switch between, and manage multiple projects. New projects auto-create a starter class set (Dock, Vessel, Seawall, Ramp) so there's no blank-page problem.

Running it locally (no Docker)

Backend

cd backend
python3 -m venv venv
source venv/bin/activate        # Windows: venv\Scripts\activate
pip install -r requirements.txt
uvicorn app.main:app --reload --port 8000

Runs at http://localhost:8000. Interactive API docs at http://localhost:8000/docs. First run downloads nothing extra — the YOLOv8n ONNX model (app/ml_models/yolov8n.onnx, ~12MB) is already included in this repo.

Frontend

cd frontend
npm install
npm run dev

Runs at http://localhost:5173, proxying /api/* to the backend on :8000 (see vite.config.ts). Open it, create a project, and go.

Running it with Docker (one command)

./deploy.sh
# or directly:
docker compose up --build

Frontend at http://localhost:8080, backend at http://localhost:8000. Data persists in named Docker volumes (backend_uploads, backend_db) across restarts.

Deploying for real (steps, since I can't do this part for you)

Backend → Render:

1. Push this repo to GitHub.
2. In Render, "New Web Service" → connect the repo → it should pick up render.yaml automatically (Blueprint deploy), or manually set: Docker runtime, root ./backend, Dockerfile ./backend/Dockerfile.
3. Add a persistent disk if you want uploaded images/DB to survive restarts (see render.yaml — free tier may not support this; check Render's current pricing).
4. Note the resulting URL, e.g. https://coastalpulse-api.onrender.com.

Frontend → Vercel:

1. Import the repo, set root directory to frontend.
2. Build command npm run build, output directory dist (already in vercel.json).
3. Add an environment variable VITE_API_BASE = your Render backend URL from above.
4. Deploy.

Backend CORS: once you have a real frontend domain, tighten allow_origins=["*"] in backend/app/main.py to your actual Vercel domain.

Project structure

backend/
  Dockerfile
  requirements.txt
  app/
    main.py             FastAPI app + all routes (projects, classes, images, annotations, detect, dashboard)
    models.py            SQLAlchemy models
    schemas.py            Pydantic request/response schemas
    detection.py           YOLOv8n ONNX inference (real, CPU-only)
    geo_utils.py             Pixel<->geo conversion, polygon centroid/area
    database.py               DB engine setup (env-configurable DATABASE_URL)
    ml_models/yolov8n.onnx      Pretrained COCO weights (~12MB, included)

frontend/
  Dockerfile, nginx.conf, vercel.json
  src/
    api.ts                  API client (env-configurable API base)
    types.ts                 Shared TS types
    components/
      CostBarChart.tsx, AreaDistributionChart.tsx, CountDonutChart.tsx
      MapView.tsx, HeatmapLayer.tsx, HotspotWidget.tsx
      AnnotationCanvas.tsx    Real interactive drag-to-draw annotation tool
      TopNav.tsx, PropertyTypeFilter.tsx
    pages/
      DashboardPage.tsx
      SetupWizard.tsx           Classes / Images (geo bounds) / Annotation / Parameters

docker-compose.yml
render.yaml
deploy.sh