A DJI thermal camera's accuracy specification is not a guarantee that every field temperature is correct. A useful measurement also needs a suitable surface, controlled reflections, enough native target pixels, correct focus and range, and a record of the target's operating conditions.
For the Zenmuse H30T and Matrice 4T, DJI publishes High Gain accuracy of ±2°C or ±2%, whichever is greater. That describes the camera under the applicable manufacturer conditions. It does not automatically cover uncertainty introduced by polished metal, an undersized component, the atmosphere or a changing electrical load.
This reference helps inspection teams decide what to capture, when to hold a quantitative result and what a reviewer needs before accepting it. It is not a calibration certificate, an application standard or permission to approach energized equipment.
Technical sources rechecked October 7, 2026. Examples are planning illustrations, not SpeedyDrone field tests.
In this reference
1. Understand what the camera is measuring
A thermal camera estimates surface temperature from the infrared radiation arriving at its detector. It does not touch the object, and it does not directly reveal the temperature inside a cable, wall or machine. The useful question is therefore not simply “What number is on screen?” but “What surface and radiation path produced this number?”
For many opaque inspection targets, the camera receives a combination of emission from the surface and radiation reflected from its surroundings, modified by the path through the air. If the surface or material transmits relevant infrared radiation, transmission adds another complication. A method for one surface should not be transferred automatically to another.
Material, finish, coating, moisture, actual surface temperature and equipment load.
Thermal surroundings, viewing geometry, distance, atmospheric temperature and humidity.
Native target pixels, lens, focus, gain, compensation settings and the selected measurement region.
Define the measurement purpose before the flight. Detection asks whether a thermal feature is present. Comparison asks how a component differs from a relevant peer under comparable conditions. Quantitative measurement asks for a surface temperature with a defensible method and stated limitations. A capture adequate for detection may not be adequate for either of the other tasks.
Record the specific component and surface, not only an asset name. “Transformer inspection” does not identify whether the measurement concerns a painted tank, a polished connector, a bushing or a reflected sky patch. Those targets can require different assumptions even within one image.
2. Separate camera accuracy from field uncertainty
DJI's H30T FAQ specifies a blackbody target observed at 13 m in a windless 25°C laboratory for its published accuracy figures. This is a controlled reference condition, not an instruction that every mission should fly at 13 m or that every target at that distance is reliable. Keep the test conditions attached to the specification.
Without the Infrared Density Filter, H30T High Gain is specified at ±2°C or ±2%; Low Gain is ±5°C or ±3%, whichever is greater. Matrice 4T publishes the same two accuracy classes in its specifications. Do not attach H30T's stated laboratory setup to Matrice 4T unless the relevant Matrice documentation explicitly establishes it.
Manufacturer High Gain specification, whichever is greater. At a nominal 150°C, the percentage term is 3°C. This arithmetic illustrates the published tolerance; it is not a field uncertainty estimate or a 95% confidence interval.
Three ideas should stay separate. Accuracy concerns agreement with the quantity being measured. Repeatability concerns how consistent repeated observations are under a defined setup. Uncertainty describes the remaining doubt associated with a measurement result. Several stable readings can still be biased by the same wrong emissivity setting or reflected background.
For a quantitative programme, build a method-specific uncertainty assessment with the qualified reviewer. Consider the reference instrument, surface assumptions, geometry, spatial sampling, environment and observed repeatability. Do not invent “±2°C camera + ±1°C weather = ±3°C total.” Components need an appropriate measurement model, evidence and treatment of dependencies.
NIST Technical Note 1297 provides general uncertainty-evaluation and reporting principles. It is not a DJI approval, a Canadian thermography requirement or an uncertainty budget for this article. If you report a confidence or coverage statement, retain the evaluation and the basis for that statement, not just a copied percentage.
3. Control emissivity and reflected surroundings
Emissivity describes how efficiently a surface emits thermal radiation relative to a blackbody at the same temperature. It is a property of the relevant surface condition and measurement context, not one permanent number assigned to a material name. Paint, oxidation, finish, contamination and moisture can matter as much as the underlying metal.
FLIR identifies polished copper and aluminium as examples of very low-emissivity surfaces, often below 0.10. Such surfaces can behave like thermal mirrors. A striking hot or cold patch may be radiation reflected from a nearby machine, the sky, a wall or the observer rather than the component's own emission.
Radiation reaching the camera
Camera receives both contributions
Reflected temperature is not air temperature
Atmospheric temperature concerns the air along the camera-to-target path. Reflected apparent temperature concerns the thermal surroundings reflected by the target into the lens. The two values can differ substantially, especially when a shiny surface reflects a cold sky or a hot neighbouring object.
A weather-app air temperature is not evidence that you have measured the reflected background. Likewise, selecting a material from a software table does not prove that its emissivity matches the finish of the inspected component.
Investigate before changing the settings
When a reading looks implausible, capture the same feature from another safe viewpoint and identify possible reflectors. A reflection may move or change as the viewpoint changes. This is a useful diagnostic clue, not conclusive proof: component geometry and a real thermal pattern can also change in the image.
Use a documented emissivity or reflected-temperature method where quantitative results matter. A known high-emissivity reference can help when the owner-approved procedure permits it, but it must represent the relevant surface temperature and be given time to equilibrate. Do not apply tape, paint or another treatment to energized, inaccessible or safety-critical equipment without the required procedure and authority.
Never tune emissivity until a suspicious number looks reasonable. Keep the original assumptions, explain any justified correction and preserve both the source file and analysis record. If the surface remains too reflective for the method to support an absolute result, report a qualitative observation or arrange an appropriate corroborating measurement.
4. Calculate target coverage, not just flight distance
Seeing a target is not the same as measuring it. Instantaneous field of view, or IFOV, describes the angular footprint of one detector pixel. At greater distance, that footprint covers a larger area. A small hot component can then share a pixel with a cooler background, reducing the apparent peak and making spot placement unreliable.
FLIR's practical guidance is to cover the hot area where a spot value is requested with at least 3 × 3 pixels. Treat this as a planning concept, not a universal acceptance criterion or a DJI guarantee. Optical spreading, target shape, contrast, focus and the consequence of an incorrect result may require substantially more coverage.
A resolved area provides more samples of the intended surface.
A small area can share a sample with its background.
A transparent metric planning example
For a small-angle estimate, one-pixel footprint in millimetres is approximately distance in metres × IFOV in milliradians. Projected target width divided by that footprint estimates the number of native pixels across the target. Use the camera/lens IFOV, not the visible-light camera's zoom or an enlarged thermal output size.
Assume, purely for illustration, an IFOV of 1.0 mrad and a flat target facing the camera. At 20 m, one pixel covers approximately 20 mm. A 60 mm square covers about three pixels across; a 10 mm bolt covers only half a pixel. Neither calculation proves the temperature is accurate. It only screens whether the geometry is plausible before focus, reflections and the remaining controls are assessed.
Scroll within the table on a phone to read every column.
| Distance | One-pixel footprint | Three-pixel width | 60 mm target coverage |
|---|---|---|---|
| 10 m | About 10 mm | About 30 mm | About 6 pixels across |
| 20 m | About 20 mm | About 60 mm | About 3 pixels across |
| 50 m | About 50 mm | About 150 mm | About 1.2 pixels across |
The relevant size is the hot region or measurement patch, not the whole assembly. A large cabinet may fill the frame while the connector you need occupies less than a pixel. An oblique target also has a smaller projected width than its physical width. If safety prevents adequate coverage, change the equipment or method, or downgrade the claim; do not compensate by enlarging the screenshot.
Distance adds an atmospheric path as well. Air temperature, humidity and the camera-to-target range affect the received infrared signal, particularly over longer paths. H30T can read laser distance and environmental temperature automatically in DJI Pilot 2, with emissivity and humidity entered manually. Verify that the rangefinder actually refers to the measured surface. Automation does not validate a hit on a narrow component or the representativeness of environmental inputs.
5. Check viewing angle, focus and region placement
There is no single best viewing angle for every thermal inspection. An oblique view can shrink projected target area, change surface emissivity behaviour and expose a different reflected background. Define a safe geometry for the specific component and use alternate views when they help separate reflection from a feature attached to the asset.
Focus matters because a blurred boundary mixes a small hot region with its background. DJI lists the H30T infrared minimum focusing distance as 13 m and recommends working beyond it. The FAQ also says that inside 13 m the image is out of focus while temperature-measurement accuracy itself is not affected. Preserve both statements: this is not a prohibition on all close measurements, but it does not establish reliable sampling of a tiny blurred component.
Matrice 4T's thermal focus range is 5 m to infinity. Its optics are different, so do not transfer H30T's focusing guidance to it. A manufacturer's focus limit is also not an electrical clearance, an aircraft safety buffer or a complete mission-planning rule.
The measurement region needs an identity
Place a spot or area tool on the intended surface, away from mixed edges when the procedure allows. An area maximum can help find a feature, but an oversized region might include a reflected hot object or an unrelated part. A mean value can conceal a small anomaly by averaging it with surrounding pixels. Record which statistic and region you used.
Keep a matched visible image or other unambiguous context showing the component, face and measurement location. For repeat inspection, retain the viewing side, range and region definition. A point at the same image coordinate next month may refer to a different surface if the framing has changed.
6. Select the right DJI gain mode and range
Choose a measurement mode from the expected target temperature and the result you need. High Gain has the tighter published accuracy class, but the target must remain within its usable range. Low Gain accommodates hotter targets. A saturated measurement or an observation-only mode should not be presented as a valid quantitative result.
Integrated 640 × 512 native thermal detector, 16-bit R-JPEG and a 5 m-to-infinity thermal focus range.
View at SpeedyDrone →
1280 × 1024 native thermal detector, R-JPEG and measurement modes with model-specific focus and filter conditions.
View at SpeedyDrone →Scroll within the table on a phone to read the complete mode reference.
| Camera / mode | Measurement range | Published accuracy | Use boundary |
|---|---|---|---|
| H30T High Gain | -20°C to 150°C | ±2°C or ±2% | Measurement when the target fits the range |
| H30T Low Gain | 0°C to 600°C | ±5°C or ±3% | Wider temperature span, broader accuracy class |
| H30T High-Res | Not a thermometry mode | Temperature measurement not supported | Observation of small temperature differences |
| Matrice 4T High Gain | -20°C to 150°C | ±2°C or ±2% | Measurement when the target fits the range |
| Matrice 4T Low Gain | 0°C to 550°C | ±5°C or ±3% | The upper limit is not H30T's 600°C |
For all numerical accuracy entries, use whichever term is greater and retain the applicable manufacturer conditions. These ranges concern the temperature being measured. They are not the aircraft's allowable ambient operating-temperature range, and they do not make every surface inside the range suitable for the same method.
What the H30T Infrared Density Filter changes
The optional filter is for extreme-temperature work, not an everyday accuracy upgrade. DJI lists filtered High Gain at -20°C to 450°C and filtered Low Gain at 0°C to 1600°C, and recommends Low Gain for filter use. The filter reduces received infrared signal and degrades image quality; DJI says it is unsuitable for normal-temperature observation.
Under the FAQ's stated test conditions, filtered accuracy is ±80°C or ±8% at temperatures up to 1000°C, and ±100°C or ±10% above 1000°C, whichever is greater. That solves a range problem while introducing a much broader accuracy envelope. Record the filter installation as part of the camera configuration and evaluate whether the resulting uncertainty is acceptable for the task.
7. Do not confuse sensitivity with accuracy
NETD is a sensitivity/noise metric, not an absolute-temperature guarantee. H30T and Matrice 4T list sensitivity of ≤50 mK under their stated optical conditions. Since 50 mK corresponds to a temperature difference of 0.05°C, the number is easily misread. It does not mean an unknown surface is measured to ±0.05°C.
Native detector resolution, output resolution and absolute accuracy are also separate. H30T has a native 1280 × 1024 detector. Matrice 4T has a native 640 × 512 detector and can produce a 1280 × 1024 output with Super Resolution. Do not treat the enlarged output as twice as many independent detector samples or a promise of doubled radiometric accuracy.
Digital zoom enlarges the captured samples. A colour palette or adjusted display span changes how a pattern is displayed. Neither recovers missing target information. Two images can look very different because their automatic scales differ even when the underlying temperatures are similar. Retain the scale and original radiometric data with presentation images.
FFC is not a substitute for calibration evidence
DJI describes Flat-Field Calibration as optimizing thermal image quality so changes are easier to observe. It addresses camera image uniformity, not an incorrect surface model, a reflection or a target too small to sample. Do not repeatedly force FFC to make an unstable reading look convincing, and follow the manufacturer's procedure.
Where your programme requires a calibration record or reference check, document that requirement separately. A pre/post-capture check against a suitable reference can reveal drift or an inconsistent setup. It does not, by itself, correct all surfaces in the scene or turn the aircraft into a traceable calibration system.
8. Record weather and actual operating state
Weather can change the target itself, not just the radiation path. Sunlight can heat one face of an asset; wind can cool it; moisture can change surface behaviour and cooling. These may be real temperature changes rather than camera errors. The reviewer needs to know whether the pattern relates to the condition being investigated.
Record the asset's load, run time and recent startup, shutdown or maintenance where relevant. Comparing a lightly loaded component with a heavily loaded one may not answer the intended condition question. Similarly, repeating the same route after a weather change does not create a controlled before/after experiment.
A repeatable flight route is not automatically a repeatable thermal measurement.
For Canadian outdoor programmes, include precipitation, snow or ice cover, wet surfaces, sun exposure and the transition between indoor and outdoor conditions when they affect capture. Snow covering a roof changes the observed surface. Water or condensation on the lens can compromise the view. Postpone or qualify results when the documented method cannot be met, rather than calling every missing anomaly a clean inspection.
There is no universal weather threshold or waiting period for all assets in this reference. Set those requirements through the applicable method and qualified reviewer. SpeedyDrone's solar farm thermal inspection guide addresses application planning; the powerline and utility inspection workflow covers asset context. Neither replaces the measurement controls here.
9. Match the claim to the evidence and threshold
An observation identifies a thermal feature. An interpretation connects that feature to operating context and assumptions. A diagnosis identifies an asset condition and may justify corrective action. Those are different levels of claim. A thermal anomaly alone should not be reported as proof of a loose connector, wet insulation or a defective solar cell.
If emissivity, target coverage, range, focus or the relevant operating state is unknown, preserve the anomaly and record the limitation. Escalate for qualified review or corroborating inspection rather than producing an unsupported precise temperature.
Define the decision rule before reviewing the result
The asset owner or qualified discipline should establish the applicable threshold, comparison basis and escalation action. A temperature difference between two regions, an absolute maximum and a trend over time are not interchangeable criteria. Decide which is meaningful for the asset and what conditions make the comparison valid.
Illustration only: suppose a project uses an 80°C threshold, and a reviewed result is 79°C with a defensible expanded uncertainty of ±4°C. The interval extends from 75°C to 83°C, so it crosses the threshold. The number alone does not settle a pass/fail conclusion. Apply the agreed decision rule, seek corroboration or recapture; do not invent a rule after seeing the result.
The ±4°C value in that illustration is assumed, not a DJI specification or an uncertainty estimate calculated by this guide. The purpose is to show why proximity to a threshold matters. Higher-consequence decisions need stronger evidence, and “no anomaly detected under the recorded conditions” is not the same claim as “the asset is safe.”
10. Troubleshoot the cause before accepting the number
A useful troubleshooting record states what was observed, which control was tested and whether the capture became fit for its intended purpose. Do not delete a suspicious frame simply because another frame looks better. Retain enough context to explain the change, including any settings or operating-state differences.
Scroll within the table on a phone to read the check and reporting action.
| Observed issue | Investigate | Next check | Report boundary |
|---|---|---|---|
| Patch changes with viewpoint | Reflected surroundings or changed geometry | Compare safe alternate views and identify reflector locations | Do not assign an absolute value until the surface method is defensible |
| Small hot part cools as distance increases | Target/background mixing, focus and atmosphere | Check native pixel coverage and repeat from a permitted closer position | The earlier peak may be under-resolved; do not “fix” it with digital zoom |
| Critical target exceeds the range | Saturation or inappropriate gain | Use an appropriate supported range/configuration and recapture | Do not use a clipped maximum as the true peak |
| Temperature shifts across the whole scene | Target state, environment, settings or camera uniformity | Check recorded conditions, compensation, FFC procedure and a suitable reference | A global change is not automatically an asset deterioration trend |
| Same route produces different colours | Palette or display-span differences | Compare original data and scale, not screenshots alone | Colour contrast is not a numerical temperature comparison |
| Anomaly is absent on a repeat visit | Load, weather, moisture, geometry or capture completeness | Verify that relevant conditions and target coverage were comparable | Absence under changed conditions does not prove repair or safety |
If recapture cannot resolve the limitation, retain a clear status such as “qualitative observation only” or “measurement held for review,” along with the reason. A recorded access or weather limitation is useful information for the next team. An unexplained confident number is not.
11. Use a field measurement acceptance checklist
These checks support an owner-approved method; they are not a certification checklist. A team can use them before departure, at the target and before leaving the site. Record unresolved items explicitly. Passing a checklist also does not authorize a flight, approach distance or work on an energized asset.
Identify the component, surface, purpose, expected temperature and the reviewer-approved acceptance/decision rule.
Confirm a supported measurement mode, unsaturated range, focus, adequate target pixels and a defensible surface/reflection method.
Retain range/view, operating load, weather, settings, repeat capture and any reference-check result required by the method.
Preserve radiometric evidence and context. Release only the claim the method supports; recapture, qualify or escalate unresolved measurements.
Before leaving the target
- Confirm that the specific hot area is resolved, in focus and inside the supported range.
- Retain an alternate safe view when reflection or geometry could change interpretation.
- Capture matched visible context with the target and measurement region identifiable.
- Repeat critical observations under a recorded, consistent setup.
- Save the original radiometric file where supported and note unresolved assumptions.
Do not add a universal number of repeats or a fixed stabilization time to this checklist. Select those through the camera procedure and application method. If the camera or target has not reached the required stable condition, document that and wait or postpone as appropriate.
12. Keep a reusable measurement record
A report needs enough information for another reviewer to understand which surface was measured, how it was sampled and what could change the result. The following fields can be copied into a project inspection log. They do not prescribe a retention policy or replace an organisation's evidence-management requirements.
Scroll within the table on a phone. Record unknown inputs as unknown, not as default facts.
| Record group | What to retain | Why the reviewer needs it |
|---|---|---|
| Identity and evidence | Project/asset/component ID, surface, location, timestamp/time zone, operator, original radiometric filename, visible-image reference | Connects the reading to one actual component and its source evidence |
| Camera configuration | Aircraft/payload, lens, firmware/app/analysis version, native resolution, gain/range, filter state, focus status | Establishes the applicable capability and specification envelope |
| Geometry and sampling | Camera-to-target range and its basis, viewing side/angle description, target dimensions, native pixel coverage, measurement-region/statistic | Shows whether a small target and repeat observations are comparable |
| Compensation assumptions | Emissivity/value basis, reflected-temperature method where applicable, atmospheric temperature, humidity, any window/path limitation | Exposes inputs that can bias the inferred surface temperature |
| Target state and environment | Load/run time, recent change, sun exposure, wind, precipitation/moisture/snow, environmental observation source | Separates actual state changes from camera/setup effects |
| Quality and decision | Repeat/reference checks as required, uncertainties/limitations, observation/interpretation level, threshold/rule, reviewer, follow-up action | Explains why a claim was released, qualified or held |
Report the limit as clearly as the finding
Useful wording is specific: “An apparent warm region was observed on the identified component under the recorded load; absolute temperature was not accepted because the reflective surface method remained unresolved.” This is illustrative report language, not a real inspection result. It tells a reviewer what is known and what still needs attention.
Preserve the original R-JPEG and distinguish it from a presentation export. Post-processing may revise supported compensation inputs, but it cannot reconstruct spatial information that the target never occupied, recover a saturated peak or establish missing operating context. SpeedyDrone's Matrice 4T R-JPEG workflow covers the capture-to-handoff process in more detail.
Use analysis software that explicitly supports the exact camera and file. DJI's H30T FAQ identifies DJI Thermal Analysis Tool 3 V3.3.0 or later for H30T photos. Keep the actual tool version with the report and test that the original file opens and exposes the expected measurement functions before standardizing the workflow.
13. Select equipment around the measurement method
Start with the smallest relevant hot area, the minimum safe stand-off distance, surface finish, expected temperature and required deliverable. Then assess native sampling, focus range, supported thermometry and evidence handling. Buying a more capable aircraft does not remove emissivity or reflection uncertainty.
Matrice 4T offers an integrated thermal inspection path. H30T provides a higher native thermal pixel count and its own focus, gain and filter conditions. Evaluate the actual target coverage at the permitted range before concluding which is suitable. For payload-generation decisions rather than measurement principles, use the H30T versus H20T inspection comparison.
The DJI Matrice 4TD (RC PLUS 2) SP PLUS + listing provides a thermal aircraft route that can support repeated remote inspection with a compatible dock deployment. Repetition still needs comparable load, geometry and environmental conditions. For an H30T aircraft configuration, the Matrice 400 SP Plus Combo is an aircraft-package option, not a complete ready-to-measure thermal system. Its current listing excludes TB100 batteries and the BS100 charging station; confirm the complete payload and operating configuration before ordering.
Use the DJI Enterprise Series Hub to connect the method to an aircraft/payload workflow. Confirm current availability and exact package contents with SpeedyDrone, especially for enterprise payloads. Do not interpret a listed item, a successful checkout or a maximum specification as proof of field suitability.
Frequently asked questions
Does ±2°C mean every DJI thermal reading is within 2°C?
No. DJI's High Gain specification is ±2°C or ±2%, whichever is greater, under applicable manufacturer conditions. It is not a field error budget. H30T's stated test uses a blackbody at 13 m in a windless 25°C laboratory. An actual inspection adds surface emissivity, reflected surroundings, target coverage, atmosphere and method considerations. Record those controls and the basis for accepting the result. If an absolute temperature is decision-critical, a qualified reviewer should establish the applicable method and uncertainty assessment instead of copying the camera specification into the report.
Can I use emissivity 0.95 for every inspection?
No. A default value is an assumption, not evidence about the inspected surface. Painted, roughened, oxidized and polished surfaces can behave differently even when the base material is the same. Identify the relevant finish and document the value's basis. For a low-emissivity reflective component, evaluate the reflected environment and use an appropriate surface/reference method where authorized. If that cannot be resolved safely, keep the result qualitative or request corroborating inspection. Do not adjust emissivity simply to force the displayed temperature to match an expected answer.
How far away can a drone measure temperature reliably?
There is no universal distance. The answer depends on the hot area's projected size, camera/lens IFOV, native resolution, focus, reflection geometry, atmosphere and application method. Screen the pixel coverage at the permitted stand-off range, then validate the other controls. FLIR's 3 × 3 pixel guidance is a useful starting concept, not a DJI-certified working distance. A rangefinder measurement establishes distance to its hit surface; it does not prove that the thermal region contains enough uncontaminated target pixels or that the component's absolute temperature is reliable.
Does digital zoom or Super Resolution fix a small target?
No. Digital enlargement cannot recover independent target samples that were never captured. Matrice 4T has a native 640 × 512 thermal detector, even when its output is 1280 × 1024 with Super Resolution. H30T's native detector is 1280 × 1024, but it still needs adequate target coverage and focus at the working range. If the relevant hot region is too small, reassess the lens/platform, permitted geometry or measurement method. Enlarging an export can help presentation, but it is not evidence of improved absolute-temperature accuracy.
Should I always select High Gain?
Use High Gain when the expected target fits its supported range and the method requires the tighter published accuracy class. It is not suitable if the critical temperature clips or exceeds that range. Matrice 4T and unfiltered H30T High Gain cover -20°C to 150°C. Their Low Gain upper limits differ: 550°C for Matrice 4T and 600°C for H30T. H30T High-Res is an observation mode without temperature measurement. A filter changes the configuration and performance envelope. Confirm the range and record the actual mode rather than selecting whichever image looks clearest.
Is a 50 mK camera accurate to 0.05°C?
No. NETD describes thermal sensitivity relative to noise, not agreement with an unknown surface's absolute temperature. A sensitive camera can reveal subtle contrast while still having a much larger absolute accuracy specification. Do not use NETD as a report uncertainty, a calibration tolerance or a smallest guaranteed field temperature change. A comparison between visits also needs comparable load, geometry, surfaces, compensation and environment. The camera's ability to display a faint pattern is different from a method's ability to support a quantitative decision about that pattern.
Can analysis software correct a bad capture?
Only some limitations can be addressed after capture. Compatible software can use original radiometric data to analyse supported regions and compensation inputs, with the assumptions documented. It cannot recover an unresolved small target, reconstruct a saturated peak, remove all reflection ambiguity or supply missing load and weather records. A screenshot or ordinary presentation JPEG is not a substitute for the supported original R-JPEG. Preserve the original, record the analysis version and export settings, and retain any unresolved limitation with the conclusion. Recapture may be necessary when the source evidence is inadequate.
How do we compare repeated thermal inspections?
Define the same component and surface, comparable operating state, measurement statistic, geometry, native target coverage and compensation method. Retain environmental context and original files. A repeat route helps return to the asset, but does not control its electrical load, sun exposure, wind or moisture. If these differ materially, describe the comparison limitation rather than calling the temperature change deterioration or improvement. Establish the trend rule and escalation process with the qualified asset reviewer before collecting repeated data, and record when a visit cannot meet that method.
Can a thermal anomaly prove an electrical or building defect?
A thermal image can document a feature and help direct further inspection. Defect diagnosis generally needs asset context, relevant operating conditions, visible evidence, history, application requirements and a qualified reviewer; corroborating tests may also be needed. State whether the evidence supports an observation, an interpretation or a diagnosis. Do not label a low-confidence absolute value as a confirmed defect, and do not treat the absence of an anomaly as a safety certificate. Use the recorded limitations to decide whether to recapture, monitor or escalate through the owner's inspection process.
Sources and scope
Manufacturer specifications and software support are model-specific. FLIR material supports general thermography principles; it is not a specification for DJI equipment. NIST TN 1297 is general metrology guidance, not a Canadian inspection certification.
- DJI H30 Series FAQ: accuracy test conditions, modes, focus, inputs, FFC and software support.
- DJI H30 Series specifications: native detector, optics, R-JPEG and filtered/unfiltered ranges.
- DJI Matrice 4 Series specifications: thermal resolution, focus, ranges, accuracy and file formats.
- FLIR measurement parameters: emissivity, reflected temperature and atmospheric inputs.
- FLIR emissivity reference: surface behaviour and the illustrated equal-temperature example.
- FLIR distance:size reference: target sampling, practical 3 × 3 guidance and digital-zoom limitations.
- FLIR OEM sensitivity reference: NETD versus detectable thermal contrast.
- NIST TN 1297: evaluation and reporting of measurement uncertainty.
The field controls, acceptance sequence and record layout are editorial method-planning guidance. Application standards, qualified thermography practice, asset-owner requirements and site safety procedures govern the actual inspection.
Build the measurement method before choosing the drone.
Tell us the target size, surface, minimum stand-off distance, expected temperature and required evidence. These are more useful equipment-selection inputs than a detector-resolution number alone.
