Every object above absolute zero emits infrared radiation. This isn’t a special property of hot objects — your desk, the floor, and the walls of the room you’re in are all emitting infrared right now. What differs between objects at different temperatures is the spectrum and intensity of that emission. Thermal imaging detects and maps these differences to create an image.
Blackbody Radiation: The Physics
A perfect blackbody radiates energy according to Planck’s Law. Two key results from this:
Wien’s Displacement Law tells you the peak emission wavelength:
λ_peak (μm) = 2898 / T (K)
At human body temperature (37°C = 310 K): λ_peak = 9.35 μm — solidly in the LWIR band.
At vehicle engine temperature (500°C = 773 K): λ_peak = 3.75 μm — in the MWIR band.
At sun surface temperature (5778 K): λ_peak = 0.5 μm — visible green light.
This is why LWIR cameras excel at human body detection, while MWIR cameras are better for detecting aircraft engines and hot vehicle exhausts. It’s physics, not just engineering preference.
Stefan-Boltzmann Law tells you the total radiated power: radiated power scales as T⁴. This means a small temperature increase produces a large change in emission — which is why thermal cameras can detect subtle temperature differences.
How the Detector Works
Uncooled Microbolometer (LWIR)
Each pixel in an uncooled microbolometer is a thermally isolated membrane suspended above a readout circuit. When infrared radiation is absorbed by the membrane, its temperature rises by a tiny amount — typically microkelvin to millikelvin. This temperature change alters the membrane’s electrical resistance (or voltage, in the case of pyroelectric detectors).
The readout integrated circuit (ROIC) measures each pixel’s resistance change at video rates, producing a raw array of values proportional to the absorbed infrared power. This is the “raw detector output” before any image processing.
Why “uncooled” is misleading: the detector itself isn’t cooled, but the detector chip must be temperature-stabilized. Built-in thermoelectric coolers (TECs) in the detector package maintain the chip at a stable temperature (typically 20–30°C) to prevent detector drift. “Uncooled” means there’s no cryogenic cooler — not that temperature control is absent.
Cooled Photon Detector (MWIR)
Cooled detectors — InSb, HgCdTe (MCT) — work by photon counting rather than thermal response. A photon with sufficient energy causes an electron transition from the valence band to the conduction band, generating a measurable current. This process is much faster and more sensitive than the bolometric effect, but requires cooling to cryogenic temperatures (77–200 K) to suppress thermal noise.
The cooling is done by a Stirling cycle cooler — a miniaturized engine that requires 5–25 minutes of cooldown before the detector reaches operating temperature. This cooldown requirement is a significant operational consideration for tactical systems.
The Imaging Pipeline
Raw detector output is not a viewable image. The pipeline between raw data and displayed image involves:
- Dark current subtraction: Remove the pedestal signal that exists even with no illumination
- Non-Uniformity Correction (NUC): Apply per-pixel gain and offset corrections to flatten the response across the array
- Bad pixel replacement: Substitute values for known-defective pixels using neighboring pixel interpolation
- Dynamic range compression: Map the scene’s radiance range to the display’s output range
- Contrast enhancement: Histogram equalization or manual contrast/brightness adjustment
- Colorization (optional): Apply false-color palette (white-hot, black-hot, rainbow, etc.)
The quality of this pipeline — not just the detector — determines real-world image quality. Two modules with identical detector arrays can produce very different image quality depending on the processing implementation.
What Separates Premium from Budget Modules
| Parameter | Budget Module | Premium Module (e.g., SPECTRA L12) |
|---|---|---|
| NETD | 80–120 mK | 30–50 mK |
| Fixed-pattern noise | Visible banding | Indistinguishable from noise floor |
| NUC algorithm | Basic two-point | Advanced multi-point with temperature tracking |
| Athermalization | None (image quality varies with temperature) | Passive or active, stable across -40 to +70°C |
| Interface | USB only | MIPI + CML, RS422 control |
| MTTF | 10,000 hrs | > 50,000 hrs |
For a commercial product that will be deployed in the field for years, the processing quality, NUC implementation, and thermal stability of a professional module justify the cost difference many times over.
Understanding the physics and pipeline of thermal imaging helps you evaluate datasheets critically and ask the right questions when selecting a module for your application.
