NISAR SAR Change Detection - Boilerplate code
Web application with linux infraestructure and sub-programs oriented towards NISAR SAR Nasa Space Apps challenge hackathon, focused on combining the strength of machine learning with a user friendly interface.
🌐 Live demo:
https://raspberrypi.tail0a4b52.ts.net/terra-pulse/
Tech Stack
| Layer | Technology |
|---|
| Backend | Java, Spring Boot |
| Frontend | Vanilla JavaScript, HTML, CSS |
| Machine learning | Python (TF), C |
| Infrastructure | Podman, Caddy |
| CI/CD | GitHub Actions |
| Networking | Tailscale |
| Server | Raspberry Pi |
Features (MVP)
- AI Schemes generator
- Data visualization
- Change detection metrics based on NISAR SAR:
- Change level and direction
- Detection confidence
- VV / VH backscatter time series
- Dashboard with relevant information
⚠️ Project in initial phase, functionalities are changing
Deployment
The project is deployed on a Raspberry Pi using containers, with secure public access provided without the need to open ports on the router.
- Caddy as a reverse proxy with automatic HTTPS
- Podman and Podman Compose for container management
- Tailscale Funnel to expose the application publicly
- GitHub Actions to automate the workflow
Every push to main is validated before the images are built and deployed (frontend, model lab and backend builds, backend tests and a startup check of the whole stack), every deployment is smoke tested on the Raspberry Pi and through the public URL, and a failed deployment is rolled back to the last known-good images, see tools/ci.
Every push to develop goes through the same pipeline to a separate develop stack on the Raspberry Pi, with its own database, so changes are QA'd there before the weekly merge releases them to production, see Develop stack. Every pull request into develop also gets its own preview, linked in a comment on the pull request and removed when it is closed, see PR previews.
