Drone forensics: flight logs and media

Where evidence sits on a small drone, how DJI and open-autopilot logs differ, what EXIF GPS shows, and what logged times and positions can and cannot prove.

Updated

In short

A small drone spreads evidence across the aircraft's internal storage or SD card, the controller, any paired phone or tablet, and manufacturer cloud accounts, according to SWGDE's Best Practices for Drone Forensics (final version 2025-07-31), and the flight logs hold times, GPS positions and altitude. DJI's log formats are proprietary: the sources describe encrypted DAT files and AES-encrypted app records from log version 13, and the DatCon developer says the onboard files of several later aircraft cannot be processed by DatCon, while ArduPilot and PX4 publish their formats. Logged positions can be inaccurate, so SWGDE says examiners need to be ready to validate them against other sources, and logged times are often relative clocks that need converting. Public guidance is thin, and the academic work reviewed here examined specific models such as the DJI Phantom III, so check any claim against the model and firmware in front of you.

Where the evidence sits

SWGDE lists these sources: an SD or microSD card (many drones have external or internal slots for media or flight logs), a flash chip on the board reachable by chip-off or in-system programming, synchronised devices such as phones, tablets and controllers, and cloud services that store media or back up flight logs. It also lists FAA Remote ID broadcasts, which carry the aircraft serial number or session ID, positions and a UTC time mark (section 5.1.5). SWGDE adds that mobile devices used to fly the drone "may contain flight log data and media" and should be within the search authority.

INTERPOL's 2020 framework adds that removable cards can carry flight-path data and geotagged photographs, and that a drone may use its handset to send photographs or video to cloud services such as iCloud or Google Photos (INTERPOL, January 2020). SWGDE says analysing drone data is virtually no different from analysing a disk image or a mobile device, and that what makes it seem different is the scarcity of parsing tools (section 6). For the phones and tablets themselves, see acquisition types and BFU vs AFU.

Powering on some aircraft alters them. SWGDE states that on some popular models a new flight log is created at power-on and that, if log storage is full, older logs are deleted. The DROP authors observed this on a Phantom III: with the internal card filled to capacity, switching the drone on deleted the 12 oldest DAT files and zeroed their space (Clark et al., 2017). They concluded that removing the internal card was the most sound acquisition method for that aircraft. SWGDE separately warns that an attached USB cable will most likely not give a full physical image, and that ISP or chip removal may be needed.

DJI's privacy policy (last updated 2025-03-01, as read on 2026-11-01) lists "Device-Related and Flight-Related Information", says users may provide location data when they "synchronize your flight record", and says after-sales staff may ask for flight data (DJI). The policy gives a general retention standard (the period necessary to fulfil its obligations) but no retention period or field list specific to flight records, and no source read here gives one.

DJI logs: DAT and TXT

For the Phantom III, Clark et al. describe two flight records. The aircraft writes FLY###.DAT files to its internal card (a new file at each power-on). The DJI GO app on the controlling device writes DJIFlightRecord_YYYY-MM-DD_[HH-MM-SS].txt, whose file name gives the flight start. The paper calls both formats proprietary, encoded and encrypted, and DROP, a Python parser, decodes the DAT files. The DROP repository (MIT licence, last pushed 2022-08-30 per GitHub, no tags) states that it handles Phantom III DAT files. Its README says the structure was learned by reverse engineering DatCon.

Later aircraft differ. The DatCon developer states that the onboard DAT files of the Mavic Air, Mavic 2, Mavic Air 2, Mavic Mini and Mavic Mini 2 are encrypted and cannot be processed by DatCon or CsvView, and that the DAT the Fly app makes for the Mavic Mini 2 is encrypted too (datfile.net, undated; latest download listed 2023-03-19 on the downloads page). SWGDE says only that data may sit in "proprietary, encrypted, and encoded DAT files" and that the acquisition method can decide whether data arrives encoded.

For the app's TXT records, dji-log-parser states that from log version 13 records are AES encrypted and need a keychain obtained from DJI's API with a developer API key (README, tag v0.5.7). Keychains can be saved with the log for offline use, so a first decryption needs a network request to DJI. The README does not say what that request transmits.

The library source at the same tag shows an HTTPS POST of JSON to a dev.dji.com keychains endpoint (a default the caller can override), with the API key in an Api-Key header (keychain/api.rs). The JSON holds a version number, a department code and encoded keychain blocks; per lib.rs, the library by default reads the version and department from the log's auxiliary data and builds the blocks from the log's key-storage records. That is how the code reads at v0.5.7; DJI's server side is not documented in the sources read. The README says the "general data" section is not encrypted for any version, and the 0.5.7 API docs list its fields, among them start time (UTC), a latitude and longitude, product type, and aircraft, camera, controller and battery serial numbers. The project's last release was v0.5.7 (2025-04-26), and its tool entry lists it as unmaintained.

Which DJI models and firmware versions use which log version is not documented in any source read here.

Open autopilot logs

ArduPilot records two kinds of log. Dataflash logs, with .bin or .log extensions, are written on the autopilot, by default after the vehicle is first armed, to an SD card or dataflash chip, or streamed over MAVLink (ArduPilot). Telemetry logs (tlogs) are recorded by the ground station on its own computer when connected (ArduPilot); Mission Planner names them by date and time (Mission Planner docs). The aircraft and the ground station therefore hold separate records of one flight. The ArduPilot Dataflash logs page also describes parameters that make logging less predictable: older logs can be deleted at start-up to free space (LOG_FILE_MB_FREE, default 500 MB), and logging can start at power-on rather than at first arming (LOG_DISARMED).

PX4 writes ULog (.ulg) files. By default it starts a log when the vehicle arms and creates one file per arming session on the SD card, and it supports encrypted logging (PX4 Guide, v1.17 stable docs). Old logs can be deleted in stable firmware too: the PX4 v1.17 parameter reference says for SDLOG_DIRS_MAX that the system deletes old logs if free space runs low (minimum 300 MB). The PX4 Guide's main-branch (development) Logging page, read 2026-11-01, describes rotation that is on by default and deletes the oldest .ulg files first to keep free space, controlled by SDLOG_ROTATE (main-branch Logging page). That text is not in the v1.17 stable Logging page, so check which firmware version the aircraft ran. The ULog specification is public and self-describing.

UAV Log Viewer opens MAVLink and dataflash logs. Its README at tag v1.0.3 describes no forensic workflow, hashing or evidence handling (README). Mantas and Patsakis report writing an open-source tool for ArduPilot logs, and say DROP had become the default tool for DJI logs as of 2021 (arXiv preprint). That claim sits awkwardly with the other sources: DROP covers Phantom III DAT files only, and the DROP paper itself says that, among the hobbyist DAT tools it found, DatCon "yields the most comprehensive output".

Media metadata

ExifTool documents EXIF GPS fields including measurement mode and dilution of precision (GPS tags). It also lists XMP tags DJI uses for drone images: absolute and relative altitude, gimbal and flight pitch, roll and yaw, flight speeds, latitude and longitude, and RTK flag and standard-deviation fields (DJI tags). Its DJI table also lists protobuf-format timed metadata (djmd and dbgi), with serial number and model fields for the Avata 2. Whether a given file contains any of these must be checked per file.

Limits. ExifTool describes itself as "Read, Write and Edit Meta Information" (ExifTool), so metadata can be altered with common tools. This guide's inference is that a value in a file needs custody records and corroboration. The EXIF table lists the time-zone offset for DateTimeOriginal as a separate tag, OffsetTimeOriginal (EXIF tags), so DateTimeOriginal alone does not state a zone. ExifTool is actively developed; its history records DJI timed-GPS decoding added in versions 13.52 (2026-02-27), 13.54 (2026-04-01) and 13.60, and it labels 13.55 as the latest production release and 13.60 (2026-10-28) as a development release.

What times and positions can and cannot show

Timestamps often are not wall-clock time. In the Phantom III DAT, each record carries an internal bus-clock tick, and the flight status record carries flight time in milliseconds since flight start. Clark et al. use that to correlate DAT and TXT files and to find the Unix time of events, matching GPS rows to 5 decimal places. ArduPilot's GPS message has TimeUS ("time since system startup") plus GMS and GWk (milliseconds into the GPS week, weeks since 5 January 1980) (log messages). ULog message timestamps are in microseconds, and a UTC offset appears among the predefined information keys as time_ref_utc. Converting to wall-clock time is a derivation step to document and test.

Positions are estimates. SWGDE states that an uncalibrated GPS at startup may log inaccurate locations, and lists checks: recovery location, third-party observations, the photographic and video record, satellite count, and GPS records from synchronised devices. Mantas and Patsakis quote accuracy of about 1.5 m for standard GPS, 40 cm for differential and 1 cm for RTK, and note drones can be spoofed. They also argue that logs on the aircraft and the ground station may contradict each other, from sensor differences and, they say, because commands in most UAVs are not cryptographically authenticated, and that logs could therefore be crafted on either device. That is their position, not a tested finding in this guide.

Logs tie data to devices. Clark et al. report linking the user to a specific device from extracted metadata, including serial numbers in the TXT record. Tying a device to a particular person would need other evidence; that is this guide's inference, not a finding in the sources.

Published guidance and its age

  • SWGDE 21-F-002-1.2. Approved as final 2025-07-31. Its history table dates the first draft to 2020 and shows drafts released for public comment from 2021 to 2024; the last round drew one comment, resolved with no changes. It says a wide disparity exists in tools' ability to parse these systems, and that its artefact list may not apply to other vendors or custom-built drones. It points to NIST's drone data sets for validation and recommends primary and secondary tools.
  • NIST CFReDS Drone Data Set. Built by VTO Inc.; both the DSIAC repost and VTO describe DHS S&T sponsorship. Counts differ between pages: the portal entry's short description says images from 60 drones, and its long description says VTO purchased 79 drones in 30 models. A 2018 DSIAC repost says 60 drones in twenty models in its data-set section and 14 models in its article text. Check the data itself. VTO says the program ran 18 months for DHS and NIST hosts a copy (VTO, 2020). The portal entry is dated 2020, and the portal's data API gives an upload timestamp of 2020-10-14.
  • Academic work. DROP (2017) covers one aircraft. Azhar, Barton and Islam (2018) examined a Phantom 3 Professional and a Parrot AR.Drone 2.0 with open-source tools (abstract). Bouafif et al. (NTMS 2018) is paywalled and unread here, as is a 2026 ICMCSI paper on integrating a drone flight-log parser into iLEAPP. Mantas and Patsakis wrote in 2021 that the standards they surveyed contain no reference to drones, and that techniques tied to popular models may soon become obsolete. Their paper was later published in Array (2022); this guide uses the arXiv v1 text and did not check it against the published version.

Tools

Tool Status in this directory What it does here
DatCon, CsvView uncertain (latest 4.3.0, 2023-03-19) Convert and plot supported DJI DAT files; encrypted DATs listed above are not supported
dji-log-parser unmaintained Parses DJI TXT logs; v13+ needs a DJI-issued keychain
DroneXtract unmaintained (last commit 2023-07-19) Reads exported CSV, KML and GPX; its README sends users to Airdata's online Flight Data Analysis tool to convert TXT logs, which means sending the log to a third party. The DROP authors (2017) called uploading evidence to the online converter Healthy Drones, which its site now says is Airdata UAV, a drawback for examiners
UAV Log Viewer active Views MAVLink and dataflash logs
ExifTool active Reads EXIF, XMP and maker-note metadata

Commercial suites make vendor statements. MSAB said in June 2024 that XRY 10.10 added drone import profiles and that XAMN Pro could import DJI and V2 Forensics drone extractions (MSAB; see XRY and XAMN). Cellebrite's Spring 2026 fact sheet lists drone forensics for its Covert Forensic Imaging Device and a drone forensics service (Cellebrite; see UFED). Paraben's E3:UNIVERSAL page, read 2026-11-01, lists DJI Go among its data sources (Paraben; see E3). None of these pages lists supported aircraft models or firmware versions, and none of the tools was tested for this guide.

SWGDE advises validating against a known data set and cross-checking with a second tool. See validating mobile forensics tools and choosing a mobile forensics tool.

Sources

  1. SWGDE 21-F-002-1.2, Best Practices for Drone Forensics (final approved version dated 2025-07-31)
  2. Clark, Meffert, Baggili, Breitinger, DROP (DRone Open source Parser) your drone: Forensic analysis of the DJI Phantom III, Digital Investigation 22 (DFRWS USA 2017)
  3. unhcfreg/DROP on GitHub (README and repository metadata read 2026-11-01)
  4. Mantas and Patsakis, Who Watches the New Watchmen? The Challenges for Drone Digital Forensics Investigations (arXiv v1 preprint, submitted 2021-05-23; this guide uses the v1 text)
  5. Mantas and Patsakis, Who watches the new watchmen? The challenges for drone digital forensics investigations, Array 14 (2022-07) (bibliographic record only; the published text was not compared with v1)
  6. Azhar, Barton, Islam, Drone Forensic Analysis Using Open Source Tools, Journal of Digital Forensics, Security and Law 13 (2018), abstract page only
  7. Bouafif, Kamoun, Iqbal, Marrington, Drone Forensics: Challenges and New Insights, IEEE NTMS 2018 (paywalled; bibliographic record only, not read)
  8. Putra, Studiawan, Pratomo, Integration of Drone Flight Logs Parser to iLEAPP Forensic Tool, ICMCSI 2026 (paywalled; bibliographic record only, not read)
  9. INTERPOL Digital Forensics Laboratory, Framework for Responding to a Drone Incident (January 2020)
  10. NIST CFReDS portal entry, Drone Data Set (Steve Watson / VTO Inc., 2020)
  11. NIST CFReDS portal data API (entry upload timestamp for the Drone Data Set; read 2026-11-01)
  12. DSIAC repost, NIST Builds Drone Forensics Dataset for Law Enforcement (2018-07-30)
  13. VTO Labs, Back by Popular Demand: our Drone Forensics page is back online (2020-06-24)
  14. DJI Privacy Policy (last updated 2025-03-01; read on the Thailand English site 2026-11-01)
  15. datfile.net, DatCon introduction (undated)
  16. datfile.net, DatCon downloads (latest listed 4.3.0, 2023-03-19)
  17. dji-log-parser README at tag v0.5.7
  18. dji-log-parser source at tag v0.5.7, keychain/api.rs
  19. dji-log-parser source at tag v0.5.7, lib.rs
  20. docs.rs, dji-log-parser 0.5.7, Details struct
  21. ANG13T/DroneXtract README (default branch; the repository has no git tags)
  22. ArduPilot Copter documentation, Logs (copyright 2024)
  23. ArduPilot Copter documentation, Dataflash logs
  24. ArduPilot Mission Planner documentation, Telemetry logs
  25. ArduPilot Copter documentation, Onboard message log messages
  26. PX4 Guide (main branch, development docs), Logging (read 2026-11-01)
  27. PX4 Guide (v1.17 stable docs), Logging (read 2026-11-01)
  28. PX4 Guide (v1.17 stable docs), Parameter reference (SDLOG_DIRS_MAX; read 2026-11-01)
  29. PX4 Guide (main branch), ULog file format
  30. ArduPilot/UAVLogViewer README at tag v1.0.3
  31. ExifTool GPS tag names
  32. ExifTool EXIF tag names
  33. ExifTool DJI tag names
  34. ExifTool home page (version 13.60, 2026-10-28)
  35. ExifTool version history
  36. MSAB, Now released: XRY 10.10 (2024-06-26, vendor)
  37. Cellebrite, Spring 2026 Release fact sheet (vendor)
  38. Paraben, E3:UNIVERSAL (vendor)
  39. HealthyDrones site (banner: "HealthyDrones is now Airdata UAV"; read 2026-11-01)