Field-scale boll-opening maps from UAV and Sentinel-2 imagery for harvest-aid timing in cotton

What it is: maps of the percent of open bolls across whole cotton fields, updated weekly from UAV RGB and multispectral imagery and Sentinel-2 time series, and turned into zone-timed harvest-aid prescriptions: which part of a field is ready for defoliation, and when.

Why it matters: growers time harvest aids from a few field-average readings. Fields do not mature evenly, so part of a field is treated too early (yield and fiber-quality loss) while another part waits. A map lets growers and spray drones treat each zone on its own date.

How it works: a deep learning detector counts open bolls in UAV imagery; ground counts of open and total bolls calibrate the counts to percent open bolls; a lightweight temporal attention model trained on the UAV maps extends the estimate to free 10-m Sentinel-2 imagery, so fields without UAV flights can be mapped. A thermal-time model forecasts the date each zone reaches the application threshold.

Proof to date: our group has published UAV open-boll detection from LiDAR and RGB imagery (Current Plant Biology, 2025) and satellite yield mapping for cotton (Smart Agricultural Technology, 2026). Field-scale mapping and zone-timed prescriptions are at proof-of-concept stage.

Next steps: on-farm strip trials with spray-drone application on commercial fields in the Texas High Plains, comparing zone-timed with uniform application for defoliation, lint yield, fiber quality, product use, and cost. We are open to sponsored research, co-development, and pilot trials with input, equipment, and digital-agriculture partners.