Tests the AMG8833 (same Grid-EYE family as this project's
part), comparing two-point and multi-point NUC. Reports achievable uncertainty
below the datasheet typical spec.
Thermal sensing · reference list
Literature
References for the Grid-EYE thermal channel. Links point to the original publisher or a freely available copy; no documents are re-hosted here.
Calibrating low-cost thermopile arrays beyond factory spec
Relevant to reducing the AMG8854's ±3.0 °C typical accuracy: two-point and multi-point non-uniformity correction, and die-temperature regression.
Ambient / die-temperature-dependent regression calibration;
RMSE reduced from 6.2 °C to 0.8 °C on an uncooled sensor.
Applicable to the Grid-EYE's onboard-thermistor-referenced pixels.
Ten ML regressors reduced bias but not the underlying random
error. Bounds what post-hoc correction achieves; consistent with the flat-field
finding that correction reaches the temporal noise floor.
Water skin vs. bulk temperature (cool-skin effect)
An IR radiometer measures the water's radiating surface, not the bulk. A still, heated bowl on the bench showed a 10–20 °C skin-to-bulk difference.
The Temperature at the Ocean-Air Interface
Original theoretical treatment of the cool-skin effect; the
basis for later models.
Basis of the COARE bulk-flux correction model.
Current revision of the same model.
In-situ radiometric skin measurement against floating bulk
thermometers; freshwater, nearer this project's scale than open ocean.
Validating IR against flowing water
Radiometric surface reading as a proxy for the temperature of moving water: calibration and validation methodology.
A controlled flowing water surface as a radiometric
calibration source, with thermistor ground truth below the surface.
Radiant (IR) and kinetic (bulk, in-situ) stream temperature
agreed within ±0.5 °C across flowing conditions. Reference target
for the field-validation stage.
Review of calibration and validation pitfalls specific to
flowing-water IR.
Low-cost uncooled sensor vs. river temperature comparison.
Field-deployment and validation overview.
Windows and external optics in front of an IR sensor
A window bonded in the optical path both attenuates the scene radiance by its transmittance τ and adds its own emission at the window temperature. These references give the radiometric model, the measurement methods, the standardized procedures, and the physical limits. They are the basis for the Window Calibration work.
The six-component radiance balance for a window in the path,
its reduction to
R = ε·τw·Robj +
(1 − ε·τw)·Renv under a
single-temperature assumption, and a two-temperature contrasted-scene ratio method
for τw that does not need a calibrated camera. States the limits used
in this project: transmittance varies with window temperature, with target
temperature through the spectral shift, and with pixel position for oblique rays;
the environments on the two sides need not be equal.Shows the band-averaged transmittance of a real (non-grey)
window depends on both window and target temperature, and gives the two-target
differencing equation for measuring it. Notes that the correction becomes
indeterminate (0/0) when the window temperature equals the target temperature —
the degeneracy encountered in the first warm-up run.
Three procedures for measuring and compensating for the
transmittance of a window, filter, or atmosphere. A normative reference within
ISO 18434-1.
Places window-transmittance and reflected-temperature
compensation inside a documented measurement workflow.
Defines a figure of merit that trades transmitted signal against
window self-emission, and shows how anti-reflection coatings raise it and how that
advantage falls off as the window heats. Sets the ceiling on what window treatment
can buy.
Companion treatment of recovering a scene behind a window whose
own emission is significant.
Practical window-material selection: transmittance versus
thickness, matching the passband to the detector's spectral response, and the
blackbody-reference calibration route.
The window in this work is a UQG ZSW-132 —
Ø13 mm × 2 mm, multispectral ZnS, uncoated. This sheet
gives the transmission curve (~70–72% across 3–10 µm,
rolling off through the long wave), thermal conductivity 27 W/m·K,
and the mechanical and optical constants. The spectral roll-off across the
sensor's 8–14 µm band is the mechanism behind the
target-temperature dependence of band-averaged τ noted above.
Still open: a measured band-averaged τ for this exact
part, once a window-temperature probe closes the calibration and separates
transmission from the window's self-emission.
Compiled by Eggert Gudmundsson. Subset of coap_GridEye/LITERATURE.md. Local copies not re-hosted. Updated 2026-09-01.