Answers are grounded in this paper and cite their sources.
Breeding, conservation, and the control of Africanized bees require knowing the evolutionary lineage of honey bee colonies. Wing-geometric morphometrics is the standard low-cost method, but 19 landmarks must be placed by hand on every wing. We present WingNet, software that places these landmarks automatically and identifies the lineage of each colony (A, C, M, or O) with the standard reference method. WingNet was trained on colonies from south-western Asia and Kazakhstan and tested on 2800 colonies (54,290 wings) from ten collections on four continents, most imaged by other laboratories. The lineage identified from WingNet landmarks was consistent with that from manual landmarks of the same wings for 88.4% of colonies (Cohen’s κ = 0.84), against 32.5% for a conventional detector. Three safeguards support routine use: An automatic quality check excluded 92% of the wings with gross landmark errors. A correction for systematic differences between WingNet and a new laboratory, estimated from 2–5 colonies digitised by hand, raised agreement from 72% to 94% in Ecuador and from 79% to 88% in north-western Europe. Referring unclear colonies for genetic testing left 96% of the European colonies identified automatically, with 99% agreement. WingNet is freely available as a web application.
No takes yet. Share an insight, caveat, or question.
Alnori et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: