Skip to content

Quality Assurance (QA) panels

The spectralbridge-qa command now emits both a PNG panel and a machine-readable *_qa.json file for every flightline. The PNG highlights spectral checks while the JSON records the underlying metrics so you can track drift over time or feed it into dashboards.

# Deterministic quick pass (≤25k sampled pixels per flightline)
spectralbridge-qa --base-folder output_demo --quick

# Exhaustive sampling with custom RGB mapping
spectralbridge-qa --base-folder output_demo --full --n-sample 150000 --rgb-bands 650,550,480

Key sections on the panel:

  • RGB quicklook – automatically picks 660/560/490 nm (override with --rgb-bands). Red callouts overlay any flagged issues from the metrics.
  • Histograms – pre vs post correction distributions with shared bins so you can judge how BRDF/topo shifts the scene.
  • Δ median vs wavelength – bandwise medians with IQR ribbon; uses header wavelengths or sensor defaults.
  • Convolved scatter – compares corrected data to any *_convolved_envi or *_resampled_<sensor>_envi outputs with a 1:1 reference line.

Each PNG lives alongside <prefix>_qa.json, which mirrors the QAMetrics dataclass (provenance, header, mask, correction, convolution, negatives_pct, overbright_pct, issues). When the brightness correction stage runs, the JSON also lists per-band gain/offsets so the QA team can trace changes back to illumination harmonisation.

Use --out-dir if you want to collect all PNG/JSON pairs into a single folder for review. Re-running the command overwrites previous outputs, so a second pass after fixing headers or rerunning corrections always reflects the current state.

QA Dashboard

The legacy QA dashboard command still expects _qa_metrics.parquet files. Until the dashboard is updated to read the new JSON schema, keep legacy parquet artifacts if you rely on that summary view.