Uncooled SWIR camera stability for inspection systems

An uncooled SWIR camera can suit a repeatable inspection when its signal and noise remain within the decision margin throughout operation. A stable room helps, but it does not prove that the sensor, illumination and optics have reached stable conditions.

This guide is for inspections that use SWIR image intensity to separate acceptable and defective parts. You will define a temperature test, distinguish offset drift from random noise, and decide whether cooling is necessary. The acceptance criterion is reliable classification, not simply an image that looks unchanged.

Decide from inspection margin, not room temperature

Choose an uncooled camera when representative testing shows adequate separation between good and defective parts across startup, normal operation and expected temperature changes. Consider active sensor temperature control when that separation becomes insufficient, particularly with weak signals or long exposures.

SWIR inspection generally measures reflected, transmitted or emitted light within the camera's sensitivity band. It does not automatically measure object temperature. If the spectral choice remains uncertain, first review SWIR versus NIR and visible imaging.

The cooling decision has several interacting causes:

These are selection principles, not cooling specifications for any catalog camera. Confirm the actual cooling architecture before choosing a candidate.

Understand what temperature changes inside the image

An uncooled sensor has no active cooling system maintaining a regulated sensor temperature. Its temperature can still differ from ambient because of internal power dissipation and heat transfer through the mount.

Dark current is charge generated without incident light. It generally increases with sensor temperature, although the relationship depends on detector design. Do not assume a universal doubling interval or apply another camera's temperature curve.

For a simplified photon-counting model:

The charge terms are expressed in electrons, and dark-current rate is expressed in electrons per second. The variance addition assumes independent noise contributions. Converting digital image values into electrons requires a documented conversion gain; do not assume one.

A dark correction can remove an estimated average offset. It cannot remove the random shot noise associated with accumulated dark charge. Temperature-dependent pixel differences, response changes and defective-pixel behavior can also affect classification.

Image scaling matters too. Automatic contrast can make drifting data look stable, or make stable data look different. Evaluate numerical pixels and inspection scores rather than judging the display palette.

Build a test that separates camera, light and optics

Use representative good parts, borderline defects and stable reference targets. Mount everything as it will operate, including camera attachment, lens support and working distance. Keep part position repeatable so geometry does not masquerade as thermal drift.

Record the following with every image batch:

• Ambient temperature near the camera, without touching its body.
• Camera temperature telemetry, if documented, with the reported location identified.
• Exposure, gain, pixel format, frame rate and acquisition duty cycle.
• Illumination setting and any independently measured light output.
• Correction state, correction events and time since startup.
• The inspection score, threshold and final decision.

Internal telemetry may describe the electronics rather than the detector. Do not label it sensor temperature unless the documentation does.

Use fixed exposure and gain for the diagnostic test. Disable display normalization in the analysis path, and preserve the numerical image format. The high-bit-depth grayscale export checks explain why a rendered image is not equivalent to preserved pixel data.

Test illumination separately using the lighting stability test plan. A reference patch can reveal combined source-and-camera drift, but cannot identify which component changed without an independent measurement.

Run a baseline, temperature sweep and return test

A useful sequence begins with startup and ends by returning to the original condition. That return distinguishes reversible temperature response from changes caused by elapsed time or mechanical movement.

1. Capture startup behavior. Run the intended acquisition duty cycle immediately after power-up. Record scores and temperature until they settle.
2. Establish a baseline. At the nominal condition, collect dark frames, illuminated reference frames and representative part images.
3. Visit the operating extremes. Repeat the same acquisitions at the minimum and maximum expected ambient conditions, staying within every component's documented limits.
4. Return to nominal. Repeat the baseline acquisition and compare it with the first baseline.
5. Repeat the production correction sequence. Evaluate the documented correction procedure exactly as production will use it, including its timing and any acquisition interruption.

For an illustrative controlled test, ambient plateaus might be 18 °C, 25 °C and 32 °C. Those values are examples, not equipment ratings or a recommended operating range.

Define a settling criterion before testing. One possible screen is less than 0.2 °C change in the relevant temperature reading over 10 minutes, together with a stable reference score. Extend the dwell if either quantity is still trending. A stable ambient reading alone is insufficient.

At each plateau, begin with 100 dark frames and 100 illuminated reference frames. These batches help estimate drift and noise; they do not establish a very low production failure rate.

Measure offset, temporal noise and decision margin separately

For each pixel, calculate its mean across the dark-frame sequence. Changes in that mean reveal offset drift and spatial patterns. Also calculate each pixel's standard deviation across time to estimate temporal noise.

Do not use the standard deviation across one image as a substitute. That measurement mixes spatial nonuniformity, target texture and illumination variation with noise.

For illuminated images, track the reference-region mean, spatial uniformity and actual inspection score. Use identical regions at every plateau. Keep saturated pixels out of quantitative comparisons, and inspect persistent outliers rather than hiding them inside a large average.

For a fixed threshold, a useful screening metric is:

These expressions assume good scores are higher than defective scores. Reverse the signs when the inspection works in the opposite direction.

Worked example: unchanged contrast can still fail

Suppose good parts have a mean score of 1,000 DN and defective parts average 880 DN. A fixed threshold of 940 DN gives equal initial margins:

At the warmer plateau, suppose both measured means rise by 50 DN. Their separation remains 120 DN, but the defective mean becomes 930 DN:

Here, 3 DN is the measured standard deviation of the defect score, not individual pixel noise. The inspection fails this particular screening criterion despite unchanged class separation.

A six-standard-deviation allowance is an illustrative engineering screen, not a guaranteed failure probability. Real score distributions may be non-Gaussian, correlated or affected by rare defects. Final acceptance needs representative production trials and a risk-appropriate sample plan.

Keep pixel sampling and focus in the stability budget

A temperature test can expose an optical problem rather than an electronic one. Focus changes can alter the score of a narrow scratch, edge or inclusion while leaving a broad reference patch nearly unchanged.

Check the smallest relevant feature at every plateau. For an illustrative horizontal field of 160 mm sampled by 640 pixels:

Four pixels across a feature is a starting assumption for this example, not a universal detection requirement. Contrast, blur, orientation and the algorithm determine whether that sampling is sufficient.

If temperature changes edge sharpness, consult infrared lens focus stability. Do not compensate for a focus problem by continually moving the intensity threshold.

Establish the spectral and mechanical requirements using the SWIR lens selection basics. Recheck focus, distortion and score stability with the final lens, not only a convenient evaluation lens.

Choose candidate hardware without assuming cooling details

The infrared cameras category includes different infrared imaging technologies. Use the camera finder to identify candidates, then request confirmation of detector type, cooling architecture and accessible correction controls.

The LS-IRC-0029 is a SWIR area scan camera with a GigE interface. The supplied catalog context does not establish whether it uses active sensor cooling, so treat it as an evaluation candidate rather than a confirmed uncooled option.

Ask for documented dark-current or noise behavior versus temperature and exposure, where available. Also confirm what temperature telemetry represents, whether correction changes numerical output, and what startup procedure the manufacturer specifies.

The LS-SWL-0229 is a 35 mm SWIR lens specified for 900-1700 nm with a C-mount. Its listing does not establish a validated pairing with the camera above.

Check image-circle coverage, back-focus requirements, spectral transmission, aperture and achievable field of view before combining components. A matching mount alone does not establish optical compatibility.

Compare alternatives in the thermal, SWIR and UV lenses category and the lens finder. Keep the same lens and aperture when comparing cooling options unless the test explicitly evaluates complete systems.

Know when the usual uncooled answer is wrong

A stable ambient environment does not make an uncooled camera automatically suitable. Long exposures may accumulate enough dark charge to limit weak-signal inspection even after startup has settled.

Conversely, active cooling is not an automatic cure. It does not stabilize illumination, correct part-position errors or guarantee adequate focus. A cooled detector can still produce unreliable decisions if the optical signal is insufficient or the threshold is poorly chosen.

Watch for these misleading passes:

• Only bright samples pass. Include the weakest acceptable signal and lowest-contrast defects.
• Only the average stays stable. Inspect individual pixels, small regions and actual defect scores.
• Automatic settings hide drift. Log every adaptive change, then repeat with the intended production settings.
• One warm run passes. Repeat startup, temperature cycling and return-to-baseline checks.
• Correction restores the mean. Verify temporal noise and defective-pixel behavior after correction too.

Use the broader camera, lens and lighting selection test plan when several parts of the imaging chain change together.

Complete the ordering and acceptance checklist

Before committing to a cooling choice, document the following:

• Operating envelope: expected ambient range, startup conditions and acquisition duty cycle.
• Verified camera architecture: active cooling present or absent, supported temperature controls and documented limits.
• Fixed test configuration: exposure, gain, aperture, illumination, pixel format and correction recipe.
• Representative samples: good parts, borderline defects and expected appearance variation.
• Measured behavior: dark offset, temporal noise, reference drift, focus and inspection-score distributions.
• Acceptance limits: minimum decision margin and allowed false-accept and false-reject performance.
• Recovery behavior: response after restart, correction and return to nominal conditions.
• Production monitoring: reference checks that detect drift before it invalidates inspection decisions.

Choose the uncooled approach only when the complete test meets those limits with useful margin. If it does not, determine whether more optical signal, a shorter exposure, improved correction or active temperature control addresses the measured cause.

Start the lens configuration process once sampling and spectral requirements are defined. For a camera cooling assessment, send your operating envelope, exposure, representative images and acceptance limits through applications support.