DJI Terra Thermal 2D Reconstruction Guide: From R-JPEG Images to Thermal Maps & Temperature Analysis

1. Define the deliverable before the flight

Start with the decision the map must support. A site overview used to locate possible anomalies has different acceptance requirements from a numerical temperature comparison or an asset location that a field crew must find precisely. Agree on that distinction with the owner and reviewer before choosing the route, camera or report format.

Terra's dedicated workflow combines visible-light aerial triangulation, thermal alignment, a temperature-bearing orthophoto and map annotations. It also generates a synchronized visible map for comparison. This is more useful than a folder of isolated images when the question concerns a large site, but it introduces a dependency: an error in the reconstructed geometry can place a plausible temperature observation on the wrong component.

Write a short project brief containing the site and asset identifiers, inspection question, smallest relevant target, required coverage, location tolerance, temperature-review method and person authorized to accept the result. Specify whether the output is a screening map, a repeat-inspection comparison or evidence for a separate specialist assessment. Do not leave “accurate thermal map” as the whole specification; neither the intended accuracy nor the verification method is clear.

For example, a solar owner might need suspected anomalies assigned to identifiable rows and modules so a technician can revisit them. A roof owner might need areas for follow-up investigation rather than a moisture diagnosis. These are planning examples, not customer case studies. In each case, the map should preserve the evidence needed for the next person to question the interpretation.

2. Check the camera, files and software support

Use the current dedicated Thermal Infrared support list, rather than assuming that any DJI thermal camera or any R-JPEG can enter this task. DJI's current English manual lists Mavic 3T/TA, Matrice 3TD, Matrice 4T/TD, Zenmuse H20N, H20T and H30T. It explicitly excludes Matrice 30T, the Zenmuse XT series and third-party devices. That restriction concerns this reconstruction task; it does not judge their usefulness for a different inspection workflow.

The import requires both visible JPG/JPEG photographs and thermal JPG files containing radiometric data. The manual identifies visible filenames commonly ending in _V or _W and thermal filenames generally ending in _T. These suffixes help with sorting, but are not proof of valid content or a matching pair. Preserve the original names and directory relationships instead of renaming every file to a generic “inspection image.”

Each group must contain at least ten images before reconstruction can start, and the counts should correspond. Ten is a software minimum, not enough to specify a site's coverage. Equal counts also cannot prove that the same ground was photographed at corresponding moments. Inspect capture positions, timestamps and the source images when a pair looks doubtful.

Keep untouched originals separate from working copies. A palette screenshot or an ordinary JPEG exported for a presentation is not a replacement for the radiometric original. Check for missing position/attitude information, absent strips, truncated files and changes made by an intermediary application. The related Matrice 4T R-JPEG evidence workflow covers capture and preservation in more detail.

3. Plan the mission around the thermal target

DJI recommends a DJI Pilot 2 Orthophoto mission with both visible and thermal lenses selected. Plan ground sampling distance from the thermal camera, not the sharper-looking RGB image. The limiting question is whether the target occupies enough useful thermal pixels for the intended interpretation. A small component can be obvious in the visible photograph while occupying too little thermal area to support a meaningful reading.

Use a representative capture to check that geometry before surveying the whole site. Review visible sharpness, thermal target coverage and the ability to identify the same object in both images. Digital enlargement and super-resolution presentation do not supply missing native radiometric detail. The current manual requires High Gain or Low Gain acquisition; images taken in Super Clear mode are not supported for this reconstruction.

Extra overlap is particularly useful where solar rows or roof surfaces look nearly identical to the matching algorithm. It cannot cure a reflection, a target that is too small, or an asset operating in an unsuitable state for the inspection question. Likewise, flying along the exact boundary can leave the edge with weaker overlap than the centre. Plan a buffer deliberately and identify inaccessible portions rather than hiding them in the final crop.

For repeat inspections, record weather, surface condition, viewing geometry, operating state and the time of collection. As a practical project control, note whether snow, standing water, shadow or recent weather changes the exposed surface being compared. Canadian seasonal conditions can make two well-stitched maps poor temperature comparisons. The owner or qualified reviewer should define the collection window and acceptable conditions for the specific task.

The flight remains subject to its own approvals. DJI's altitude and distance guidance does not determine a Canadian operation category or authorize access to a roof, solar site or substation. Use Transport Canada's operation-category guidance for the intended aircraft, airspace and proximity to people, and obtain the relevant site permissions. A software-supported recurring workflow does not itself authorize remote operation.

4. Build and inspect the Terra project

Create a Thermal Infrared project from the Reconstruction Project area, then import the folder containing the paired visible and thermal data. Use a clear project name and a stable storage path. DJI warns that special characters such as a number sign in the project path can cause reconstruction errors. Keep the working project separate from the protected original-data folder.

After parsing, address the import warnings before starting a large computation. Enable the photo-capture-point display and look for missing coverage, unexpected positions or an incomplete flight strip. Check image pairing and missing POS information. Terra supports POS import/export and local PPK, but these are processing inputs rather than independent proof of the delivered positional accuracy.

Confirm the output coordinate system against the owner's requirements. The import manual describes an arbitrary coordinate system as the default when images lack POS, and a known coordinate system when they contain POS. Check the actual selected system, units and any required datum information. For a map that must line up with existing assets, compare identifiable locations and agree on a verification method rather than relying on a plausible basemap overlay.

Terra performs AT using the visible images. Review that result before concentrating on thermal colours: weak light, blur, low texture or missing overlap in the visible set can undermine a map even when each thermal file opens correctly. Define the region of interest without cutting away areas the brief requires. Inspect the accepted coverage against the mission boundary, including its edges.

The current parameter manual makes a useful distinction about resolution. High, Medium and Low control the visible 2D map; the thermal model uses High by default and is not affected by that selector. Increasing a setting is therefore not a general remedy for insufficient thermal target detail. Standalone and cluster computation are available, but choose the resources needed for the dataset and installed entitlement, then retain the execution record and quality report.

This thermal task produces 2D outputs. It does not generate a thermal 3D mesh, point cloud or Gaussian splatting model. Specify a separate, verified process if the owner needs three-dimensional thermographic interpretation rather than implying that Terra's other reconstruction features are automatically part of this project.

5. Keep material and parameter scopes separate

Set emissivity according to the surface being evaluated and the review method. Terra's presets are convenient starting assumptions, not measurements of an individual panel, coating or piece of metal. A single default value cannot represent every surface in a mixed industrial scene. Different settings can produce different converted temperatures at the same location without a physical change in the asset.

The reconstruction applies each selected material across the full map and generates a separate temperature layer for it. Region-specific emissivity is not supported at this stage. You can add a custom material within the documented 0.10–1.00 range, but choosing a value still requires a defensible basis. Avoid presenting a newly created material name as evidence that its emissivity is correct.

Environmental parameters can come from the original datasets or be entered using Custom. Review ambient air temperature, reflected temperature and relative humidity, and record where they came from. Edited files can lose readable metadata. Manual entry may resolve a parameter gap in a supported dataset; it does not convert an unsupported camera into a compatible one or recreate missing radiometric observations.

This distinction is central to review. Map annotations can override the global material. Photo measurement has independent parameters, including distance and environmental settings. If a reviewer adjusts a source-photo reading, document that change and decide what must be reprocessed or reannotated before release. Do not leave the report containing an earlier map value while describing the adjusted photo value as if the two were interchangeable.

Low-emissivity metal deserves particular caution. DJI describes how a clear sky's reflected temperature can differ substantially from ambient air temperature, causing biased readings. If a layer contains missing temperatures, a different emissivity may help investigate the software result, but raising it until the gap disappears is not validation of the real surface temperature. Keep diagnostic trials distinct from accepted settings.

6. Measure on the map, then inspect the original

Use a point for a localized observation, a line for a temperature profile along a feature, or a polygon for a region. Terra reports a line's minimum, maximum and average temperatures and distances; an area adds projected area and temperature statistics. Choose geometry that matches the inspection question. A polygon crossing several materials can mix unlike observations and make its average less useful than separate, clearly named regions.

Set and record the temperature unit. Palettes and display ranges help make patterns visible, but they do not correct material assumptions, atmospheric effects or poor target coverage. When comparing screenshots, keep the units, palette, display range and selected material layer consistent. Otherwise the same physical observation can look different simply because the rendering changed.

For an important observation, use Link to Original Photo to display the corresponding visible and thermal images. Enter photo measurement mode to inspect the target directly, review alternative source views and check that the marked feature is the intended asset. The gallery can help select a view closer to the image centre, but the reviewer must still assess target identity, angle, reflections and the actual surface seen.

Keep spatial and thermal confidence separate. An annotation can be associated with the correct row yet have uncertain thermometry; a credible source-photo pattern can also be poorly located in the map. State which question has been checked. Where the next action is consequential, the owner should arrange appropriate qualified interpretation or field verification.

DJI's thermal support FAQ reports an approximately 2°C maximum map-versus-photo difference under identical measurement parameters. This is a relative comparison, not a blanket ±2°C absolute field-accuracy guarantee. Sensor behaviour, target size, surface properties and conditions still matter. For a numerical review, compare the map and original using matched assumptions, record discrepancies and investigate them. Do not adopt that manufacturer figure as your project's acceptance threshold without a defined basis.

The thermal measurement accuracy guide explains these uncertainty sources beyond the mapping task. A map supports finding and locating an inspection question; neither a hot-looking pixel nor a report automatically diagnoses its cause.

7. Accept the evidence in four stages

Use a release checklist that distinguishes processing success from suitability for the agreed purpose. The following is an editorial project-control template to adapt with the owner and qualified reviewer; it is not a thermography standard. Set the required coverage, location checks and measurement-review criteria before processing, so an attractive result cannot redefine what counts as a pass.

Inspect the centre and boundaries, not just the most impressive view. Look for seams, repeated structures displaced onto adjacent assets, abrupt temperature discontinuities, missing strips and areas where the visible and thermal footprints disagree. Review representative normal areas as well as the selected anomalies. A strong-looking anomaly can distract from a broader registration problem.

Assign an explicit status: accepted for the stated screening purpose, held for investigation, or requiring recollection. Include the excluded areas and reasons. If the map passes for localization but not quantitative comparison, say so in the cover note and report. This allows a useful partial result without overstating what has been established.

For a repeat campaign, compare capture conditions, geometry, material settings, software version and parameter sources before interpreting a temperature change. A changed setting is an alternative explanation that should be resolved or disclosed. Where a comparison is not defensible, retain both observations and arrange a controlled revisit instead of presenting the numerical difference as asset deterioration.

8. Troubleshoot the dependency that failed

Investigate the earliest failed stage rather than repeatedly changing the palette or export settings. Begin with files and support, then visible AT and coverage, then thermal parameters and measurement interpretation. Record the symptom, test, change and outcome; preserve the original dataset while you test a working copy.

Consult the installed version's execution record, known issues and current DJI manual when a failure does not match the input checks. Save the exact message, version, processing stage and reproducible steps. A later software version can change an issue or supported input; an old forum workaround is not proof that the present workflow needs it. If the capture lacks usable visible imagery or the required surface observation, a new mission may be the only credible remedy.

9. Deliver a map that someone else can review

Package the original evidence, the project configuration and the conclusions together. A PDF alone hides much of the capture and processing context. DJI's map report export supports PDF and XLSX: the PDF includes device information, measurement parameters and annotated views; XLSX contains measurement data. “Raw measurement data” in a spreadsheet does not mean it replaces the original radiometric photographs.

A practical handoff has five groups: protected visible/thermal originals and flight records; the Terra project with its version and settings; visible/thermal map outputs and coordinate information; annotations with PDF/XLSX and the quality report; and a release note identifying reviewer, purpose, gaps and next actions. Use a manifest to identify the actual delivered files and the version of the map referenced by the report.

When the owner needs a GIS layer, test the intended export in the receiving application. DJI documents tiled TIFF output for large maps, but a successful import or correct colour display alone does not prove that numeric temperature values, units and material context survived. Verify georeferencing, sample values and the software's interpretation of the output; deliver accompanying settings and documentation. Avoid promising a universally interchangeable temperature raster.

A minimum annotation record

For each consequential finding, record a unique ID, site/asset location, map version, annotation geometry, source-photo IDs, unit and selected material, environmental-parameter source, observed pattern, review status and next action. Keep an observation distinct from an inferred cause. If several alternative source views were checked, retain their identifiers rather than only the most convincing screenshot.

For another inspection date, create a new dataset and change log rather than overwriting the accepted package. Record altered flight conditions, settings and software. An auditable comparison should let a future reviewer distinguish a physical change from a change in capture or conversion assumptions. The solar inspection resource and utility inspection guide cover the wider site workflow around that handoff.

10. Match the configuration to the deliverable

Choose the capture system after defining target coverage, site access and review requirements. The following are three current SpeedyDrone-listed options on DJI's supported thermal-input list. They address different deployment needs; the list is not a ranking or a promise that every firmware, route or licence combination is ready. Confirm paired acquisition and process a representative dataset from the actual configuration.

For a team considering the Mavic platform, the current store route is the DJI Mavic 3 Thermal Advanced (M3TA) listing; DJI's manual includes 3TA in the supported thermal group. Verify the exact variant rather than treating the legacy URL text as its current model name. Older supported payloads may remain relevant to existing fleets without becoming the preferred new purchase.

Geometric surveying is a separate selection question. The DJI Matrice 4E is an RGB mapping platform, not a source of this task's thermal images. Neither its mapping role nor RTK on another aircraft automatically certifies the location or temperature accuracy of a thermal project.

SpeedyDrone lists DJI Terra Standard. Confirm the actual edition, account/device binding, update terms, workstation requirements and Thermal Infrared entitlement before purchase. The Terra licence guide supports that discussion. If the job only needs a few source-image observations, a complete site reconstruction may add little; choose the workflow that answers the evidence question.