12 µm × 640 = 7.68 mm. That is the horizontal silicon a 640-pixel uncooled array actually presents. A thermal imaging lens that was drawn for a 17 µm VGA chip will not cover a 12 µm 640-class detector the same way, and a visible C-mount FA barrel will not transmit the band the bolometer sees. Selection starts at pixel pitch and array format, then at the material that can pass that light.
Key Takeaways
The five inputs in the machine vision lens selection guide still apply. On thermal, “format” is array × pitch, and “environment” is often outdoor or furnace-adjacent. The Lensmind family is listed as a thermal imaging line, including near-infrared thermal optics in the Product Bible — write the waveband on the RFQ so the coating is not a guess.
Thermal arrays are sold as 160, 320, 640, 1024 and similar counts. The millimetres are count × pitch. A 640×480 detector at 17 µm is 10.88 × 8.16 mm. At 12 µm the same count is 7.68 × 5.76 mm. The image circle and the focal length both move. “VGA thermal” without pitch is as empty as “5 MP C-mount.”
A microbolometer is an uncooled thermal sensor. Infrared between about 7.5 and 14 µm heats each pixel and changes its resistance. Ordinary optical glass does not transmit that band. Germanium, chalcogenide glass, and a few other IR materials do. Lensmind LMXHG barrels show a germanium front element in catalog photography. Putting a 25 mm FA lens on a bolometer camera is a mechanical experiment, not an optical one.
The Product Bible also names a near-infrared thermal family. NIR (0.7–1.4 µm class) and LWIR (8–14 µm class) are not interchangeable coatings. Write the detector waveband on the RFQ. Wrong material is the “no contrast” failure in the field.
Pixel pitch on uncooled arrays has stepped from 25 µm-class down through 17 µm and 12 µm, with 8 µm appearing on newer chips. Smaller pitch shrinks the die for the same resolution, which shrinks the lens. It also collects less energy per pixel, which is why fast f-numbers stay in the conversation. Match the lens to the pitch you actually bought, not to last year’s 17 µm demo camera.
| Pitch | 640×480 size (mm) | What changes |
|---|---|---|
| 17 µm | 10.88 × 8.16 | Larger die, longer EFL for same FOV |
| 12 µm | 7.68 × 5.76 | Smaller optics, less energy per pixel |
| 8 µm | 5.12 × 3.84 | Tightest circle; confirm MTF at that pitch |
Uncooled cameras are energy-starved compared with a 5 MP visible global shutter under LED bars. Fast thermal lenses (near f/1) are common because the bolometer needs photons. Do not stop a thermal lens down the way you stop an FA lens for depth of field — you will watch NETD collapse on the camera’s own metric. Quote the f-number the detector was characterized at, then keep it.
Same thin-lens estimate: f ≈ (sensor width × WD) / (field width + sensor width). For a 12 µm 640-wide chip (7.68 mm) at 5 m looking at a 2 m-wide busbar: f ≈ (7.68 × 5000) / (2000 + 7.68) ≈ 19.1 mm. A 17 µm 640-wide chip at the same FOV wants ~27 mm. Pitch changed the lens. Confirm on the drawing; germanium cells have their own principal planes.
The live thermal imaging lens family is the money page for this RFQ. Compact, mid, tele, zoom, and hooded LMXHG bodies cover different fields. Do not pick by barrel length alone. Pick by detector width, EFL, and f-number. Coatings that pass 8–14 µm will not automatically pass a NIR illuminator, and the reverse is also true.
Illustrative path: 640×480, 12 µm, 7.68 mm width, 8–14 µm bolometer, 8 m to a 3 m-wide kiln wall. f ≈ (7.68 × 8000) / (3000 + 7.68) ≈ 20.4 mm. Fast f-number near the camera’s characterization. Germanium cell, not FA glass. If the next camera is 17 µm 640, recompute — the EFL jumps. Contrast that dies after a coating swap to VIS is a waveband error, not a “bad lens.”
Thermal glass is sized by pitch × pixels and by the band the detector actually sees. 12 µm × 640 is 7.68 mm. Uncooled microbolometers live around 7.5–14 µm. Visible FA barrels do not. Write pitch, array, waveband, and field before anyone quotes an EFL.
Key takeaways
If you already have the detector datasheet and the field, send those two. Hangzhou will map them onto the thermal family or tell you the waveband is NIR rather than LWIR. First-pass feasibility is 48 hours.
Written by the Lensmind Optics team — Hangzhou factory (Lin’an) and Zhongshan office, company registered 2019, ISO 9001 certified 2024. Founder on this site: Baker Zhao.
Not if the detector is an uncooled microbolometer in the 7.5–14 µm band. Visible glass does not transmit that light. Use germanium or another IR material matched to the pitch.
The die size changes. 640×480 at 17 µm is 10.88 × 8.16 mm; at 12 µm it is 7.68 × 5.76 mm. Focal length and circle both move.
Typically 7.5–14 µm long-wave infrared. That is not the same as a near-infrared industrial camera. Write the band on the RFQ.
Uncooled pixels need energy. Cameras are often characterized near f/1. Stopping down to gain depth of field will starve the detector.
LMXHG catalog photography shows germanium elements for 8, 12 and 17 µm uncooled detectors. Send the detector drawing for a named match.
No. NIR illuminators around 850 or 940 nm use glass with NIR coatings. LWIR bolometers need IR materials. Mixing them is the no-contrast failure.
A short-working-distance large-FOV feasibility input pack is the maker-ready file that states part size, stand-off, pixel budget, and which image costs the buyer will...
read more…
A drawing-share pack is the minimum disclosure set and agreement that must exist before a lens maker receives device drawings that are the heart...
read more…
Turning a 50 micrometre defect at 200 millimetres into a lens commitment means locking pixel pitch, image height, and a verification target before anyone...
read more…
