TECHNOLOGY
Technology Stack
A full-stack hyperspectral intelligence platform built on open-source geospatial science, a lean Python backend, and a modern Next.js frontend.
REQUEST LIFECYCLE
End-to-End Flow
Select an EnMAP scene
Choose from three pre-loaded hyperspectral scenes captured over agricultural zones in the UAE
Live analysis engine activates
Two-pass spectral anomaly detection runs across all 224 bands of the selected scene
Risk raster computed
Per-pixel RMS spectral deviation scores identify crop stress signatures across the full scene
Priority zones extracted
High-risk connected regions are ranked by severity, area, and spectral confidence
Georeferenced outputs generated
Zone boundaries, spectral evidence, and inspection budget are exported with full coordinate metadata
Interactive dashboard rendered
Zones, risk maps, spectral charts, and budget recommendations are displayed in real time
FULL TECHNOLOGY STACK
Dependencies by Layer
Frontend
Backend API
Processing Engine
Scientific Benchmark
Data
Deployment
SCIENTIFIC DESIGN
Design Principles
No Label Leakage
The live engine uses only unsupervised spectral anomaly statistics. No disease labels, no test-set data, no ground truth.
Frozen Benchmark
The six-model scientific validation is completely separate from the live engine. Benchmark results are pre-computed and immutable.
Scene-Relative Thresholds
Priority thresholds are percentile-based within each scene. They are not absolute disease severity levels.
Reproducible Runs
Every run writes a manifest.json with the run ID, source path, CRS, and leakage-control statement.
Windowed Streaming
Full GeoTIFF scenes are processed block-by-block with constant memory footprint regardless of scene size.
Honest Uncertainty
The system explicitly reports limitations: wavelength metadata unavailable, proxy labels only, thresholds are relative.