HyperFVC Prototype

Inputs

ENVI header ENVI data cube Optional ONNX model

The page now auto-loads the first bundled `.hdr` and matching `.dat` pair on open. Upload a different ENVI pair at any time to replace the bundled scene.

Bundled data folder

Checking data folder...

Inference

Prediction opacity

Prototype inference uses vegetation indices and SWIR/red-edge rules. Replace with ONNX Runtime after exporting your trained PyTorch model.

Classes

PV-

NPV-

BE-

Uncertain-

Spectral Preview

Default bands: 83, 50, 19

FVC Prediction

Hyperspectral mode

TileLoading bundled tile...

ShapeAwaiting ENVI header...

SensorFX10 + FX17

DeploymentONNX-ready

Band Relevance

Run inference to score wavelengths

Interpretation Targets

Red-edge: chlorophyll decline and PV-NPV transition.
NIR: canopy structure and live vegetation contrast.
SWIR: water loss, dry matter, lignocellulose, and senescence.

Train In PyTorch

Use labelled semi-arid UAS tiles to train U-Net with a spectral adapter and channel attention.

model = SpectralUNet(num_bands=290, classes=3)

Export To ONNX

Freeze the trained model into a portable inference graph for desktop, cloud, edge, or service deployment.

torch.onnx.export(model, sample, "hyperfvc.onnx")

Accelerate Runtime

Run with ONNX Runtime first, then TensorRT when NVIDIA GPU inference is needed for near-real-time mapping.

onnxruntime.InferenceSession("hyperfvc.onnx")

Prototype Scope

The first build is intentionally semi-arid: load a UAS hyperspectral tile, map PV/NPV/BE, inspect wavelength relevance, and prepare the interface that will later call the ONNX model.

Phase 1Semi-arid prototype with local ENVI tile inference.

Phase 2Viewer integration with ONNX model upload or backend inference.

Phase 3Train Site A, test Site B, fine-tune and measure transfer robustness.

FutureAdapt PV/NPV proxies to polar moss and lichen hydration state.