Small-target radiometry - 27 May 2026

The first radiometry model tests subpixel target contrast.

The project has a repeatable Python model — Planck radiance, pixel fill factor, atmosphere, and NETD — for weak hot targets in the assumed drone survey geometry. These are calculated contrasts, not measurements.

Model inputs

The first pass connects target size, temperature, bandpass, path length, and detector noise.

5.94 cm Current 120 m GSD from 18 deg HFOV across a 640 pixel detector.
0.20 Pixel fill fraction for a 3 cm round target at the current GSD.
30 mK Conservative Boson+ professional NETD used for current LWIR cases.
0.990 Provisional clear-air LWIR transmission over the 120 m nadir path.

Evidence boundary

This is radiometric contrast, not a detection claim.

The model integrates Planck radiance over the selected band, mixes target and background by pixel fill fraction, applies a low-altitude transmission term, and compares apparent temperature contrast with NETD.

It does not yet include optical blur, MTF, calibration residuals, non-uniformity, smoke, sun-warmed clutter, threshold tuning, or false positives. Those remain end-to-end simulation and field-test work.

Current scenarios

Hot targets have large first-order contrast margin; weak warm spots are the useful stress case.

Scenario Band Target Fill Apparent delta SNR Disposition
Current flame proxy 8-14 um 3.0 cm, 700 K 0.20 185.4 K 6179.7 Benchmark margin
Current ember proxy 8-14 um 1.0 cm, 700 K 0.02 20.6 K 686.6 Benchmark margin
Current hot-char proxy 8-14 um 3.0 cm, 430 K 0.20 36.6 K 1220.6 Benchmark margin
Weak warm spot 8-14 um 1.0 cm, 320 K 0.02 0.3 K 9.9 Advisory margin
Lower-altitude flame proxy 8-14 um 3.0 cm, 700 K 0.45 418.5 K 13948.9 Benchmark margin
MWIR provisional value flame proxy 3-5 um 3.0 cm, 700 K 0.20 3164.8 K 63296.1 Bandpass study only

Interpretation

The next hard problems are dilution, blur, calibration, and background rejection.

Fill factor dominates early

At 120 m, the 3 cm target fills about one fifth of a pixel, while a 1 cm target fills about two percent.

Hot targets are not the stress case

Flame-like targets have large first-order radiometric margin before optical blur and field evidence are included.

Weak warm spots matter

The 1 cm, 320 K case sits near advisory margin before calibration residuals or false-positive constraints are applied.

Altitude is a real lever

Dropping the 3 cm flame case from 120 m to 80 m increases fill fraction from 0.20 to 0.45.

Bandpass bounded

The follow-on LWIR/MWIR trade keeps the first payload on the LWIR path unless later evidence justifies two channels.

End-to-end model is next

Spot size, MTF, calibration quality, sensor non-uniformity, thresholds, and clutter must be added before performance claims.

Next use

This model becomes an input to bandpass selection and the end-to-end simulator.

The bandpass trade keeps the first payload on the LWIR path while treating MWIR or dual-band sensing as a later classification or retrieval upgrade.

The simulation work should combine this contrast model with optical ensquared energy, detector calibration state, route geometry, thresholding, and held-out false-positive scenes.