DJI Names 15 Onboard AI Challenge Winners, From Banana Farms to Bridge Cracks

DJI named 15 winners of its Enterprise Drone Onboard AI Challenge 2026 on August 20, including five Best Onboard AI Model Award winners and ten Industry Application Excellence Award winners, with projects ranging from banana crop counting to bridge crack detection. The contest, which ran from March to a May 31 deadline extension, asked developers to run AI models directly on DJI's enterprise aircraft, and winners receive placement in DJI's Onboard AI Solutions Catalog and fast-track access to the DJI Enterprise Ecosystem partner audit. Twelve of the 15 winners are Chinese companies or university teams, with one Colombian project, AgroCount AI, automating banana counting at a 124-acre plantation.

DJI announced the 15 winners of its Enterprise Drone Onboard AI Challenge 2026 on August 20, closing out a global contest that asked developers to run AI models directly on the company’s enterprise aircraft instead of in the cloud. Judges named five Best Onboard AI Model Award winners and ten Industry Application Excellence Award winners, with projects spanning crop counting, bridge crack detection, beach litter identification, air pollution tracing, and search and rescue. When I covered the challenge launch in March, the pitch was simple: DJI had strapped 100 trillion operations per second of computing to its aircraft and wanted to see what developers would do with it. Five months and one deadline extension later, the answer is in. It says as much about where DJI’s enterprise ecosystem is heading as it does about any single winning project. Ten Judges Split 15 Winners Across Two Award Tiers Ten experts from the AI and drone industries judged the entries, naming five Best Onboard AI Model winners and ten Industry Application Excellence winners. Beyond hardware prizes, winners receive placement in DJI’s Onboard AI Solutions Catalog and fast-track access to the DJI Enterprise Ecosystem partner audit. The challenge, announced by DJI on its official contest site https://developer.dji.com/innovation-contest/ , covered solutions deployable on the Matrice 4 and 4D series https://dronexl.co/2025/01/08/dji-launches-ai-powered-matrice-4-series-enterprise-drone/ , Dock 3 , Matrice 400 , and the Manifold 3 onboard computer https://dronexl.co/2025/10/08/dji-matrice-4-manifold-3-obstacle-sensing-module/ . DJI said frontline operators are “best positioned to develop practical applications for our products,” which is why it opened its SDK and onboard computing power to outside developers in the first place, a bet I broke down when the contest opened in March https://dronexl.co/2026/03/10/dji-developers-build-future-aerial-ai/ . The announcement also arrived late. DJI originally planned to name winners at the end of June, then extended the submission deadline to May 31 and pushed winners to July. The list landed August 20, weeks behind even the revised schedule. AgroCount AI Turns a Week of Banana Counting Into Onboard Math Daniel Tovar’s AgroCount AI, an Industry Application Excellence winner, was developed and validated at Finca La Suiza, a 124-acre 50-hectare commercial banana plantation in Colombia. The job it automates traditionally takes four to six field workers four days, or about seven days of counting plants manually from aerial imagery. Those labor figures come from DJI’s announcement, and independent validation isn’t public yet, but the architecture is concrete: the planned system pairs the Matrice 4E with the Manifold 3 to process imagery and GPS-tag individual plants while the drone is still flying. Growers would get field data on landing rather than a memory card full of homework. DroneXL has already documented the Matrice 4E doing serious mapping work https://dronexl.co/2026/06/09/dji-matrice-4e-webodm-native-species-flora-mapping/ , so a crop-counting workflow on the same airframe is a short step, not a leap. Hangzhou New Modal Packs Nine Traffic Inspections Into One System Hangzhou New Modal Technology took a Best Onboard AI Model award for a nine-in-one fusion algorithm built on DJI FlightHub 2 and the Manifold 3. The system folds traffic enforcement, road maintenance, facility management, and traffic flow monitoring into a single closed-loop inspection workflow. Closed-loop is the operative phrase. The system detects an issue, collects evidence, generates the alert, and supports the follow-up, all inside one workflow rather than nine separate drone missions. The Winner List Runs Straight Through China Twelve of the 15 winners are Chinese companies or university teams by name, from Hangzhou New Modal to Lanzhou University’s Grass Flying Team. Daniel Tovar’s Colombian project is the only winner DJI explicitly places outside China, and the announcement does not disclose how many teams entered or from where. That entry data is the real gap in the record. DJI billed this as a global challenge, and its own press release leads with the one unmistakably international winner. Which tells you DJI noticed the optics too. Whether American developers entered and lost, or never entered at all, is a question DJI’s announcement leaves open, and it matters for reading what this contest actually measured. DroneXL’s Take The way I read this announcement, the winning projects are the least interesting part. What DJI just demonstrated is that the onboard AI era for working drones has arrived, and it is arriving almost entirely inside an ecosystem American operators are being ordered to exit. While Chinese traffic departments get closed-loop aerial inspection and a Colombian banana farm gets plant-level counting in real time, US agencies are navigating an FCC import ban that already has legacy fleets scrambling for grandfathering clarity https://dronexl.co/2026/07/05/fcc-legacy-import-ban-dji-grandfathering/ . The security concerns behind that ban are not invented. The Covered List proceeding is a real docket, and DJI’s data practices deserve the scrutiny they are getting. But banning the hardware does not ban the capability. It just decides who gets it, and this winner list is an early scoreboard of exactly that. Watch the Section 232 tariff decision due September 3 for whether Washington doubles down on walling off Chinese drone hardware while this developer ecosystem compounds without American participation. If it does, the gap between what a drone can do in Hangzhou and what one is allowed to do in Houston gets wider on a schedule you can put on a calendar. Source: DJI DroneXL uses automated tools to support research and source retrieval. All reporting and editorial perspectives are by Haye Kesteloo.