Repeatable UgCS flights for 3D sorghum phenotyping

The challenge

Breeders judge sorghum varieties by the size and shape of each panicle and how it changes as the crop matures. Measuring by hand is slow and does not scale: a single trial can hold hundreds of plots and thousands of plants, and the panicles change every week.

The team planned to build a 3D model of every plot from drone video and detect each panicle automatically. That only works if every flight is the same. A small change in height, speed or path would show up as a false change in the plants.

The solution

The team built the missions once in UgCS, grouped by treatment-replication block, and reloaded the saved missions for each weekly campaign. Each plot got two passes:

• a Circlegrammetry orbit at a 6 m radius with the camera at 45 degrees, capturing the plot centre from many angles;
• a double-grid photogrammetry pass in a cross-hatch pattern, giving even coverage across the plot.

Both flew at about 5 m above ground. The drone recorded video while moving at 2.5 to 4.5 m/s, so the circular pass took about 20 seconds and the grid pass about 30 seconds. Methods that stop for each photo can take five minutes per plot.

The video frames went through photogrammetry and 3D reconstruction, and a deep learning model the team called SegVoteNet found and measured each panicle in the point cloud.

The results

• Panicle detection reached 0.850 mAP@0.5 IOU on real crop data, against 0.679 for the VoteNet baseline in the study's ablation test.
• Panicle height matched lab-based 3D reference models at r = 0.984; length and width tracked at r = 0.938 and r = 0.894.
• Weekly campaigns ran for 8 weeks, from seed-head emergence to maturity.
• The data separated varieties: Mr-Buster and Sentinel-IG had similar panicle heights, but Mr-Buster had more than double the panicle radius, with fewer panicles.

Where this applies in India

Repeatable low-altitude missions suit crop research where plots are measured many times a season:

• plant breeding trials at agricultural universities and research institutes;
• seed company field trials comparing varieties;
• crop growth monitoring in experimental plots;
• 3D mapping of orchards and plantations.

See also our case study on UgCS Circlegrammetry mapping efficiency. SPH Engineering's UgCS training courses are also available through HobiTech.

The equipment, available from HobiTech

Flight planning	UgCS - Drone Flight Planning Software: see UgCS PRO, UgCS EXPERT and UgCS ENTERPRISE
Drone	DJI Mavic 3 Enterprise was used in the study. Ask us for current compatible drones.
Analysis	SegVoteNet, a deep learning model built by the research team

Case study by SPH Engineering, published June 2026, based on James, Chandra and Chapman (2025), Plant Phenomics, 7, 100050. This page summarises it in our own words; the research was carried out by the University of Queensland, not by HobiTech. Read the original case study on SPH Engineering ↗