Holstein-Friesian re-identification using multiple cameras and self-supervision on a working farm

Download PDF
Open Webpage

Phoenix Yu, Tilo Burghardt, Andrew W. Dowsey, Neill W. Campbell

Published: 01 Jan 2025, Last Modified: 07 Nov 2025Comput. Electron. Agric. 2025EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•Cattle Re-ID achieves ∼<math><mo is="true">∼</mo></math>96% accuracy on a farm using contrastive loss function.•Requiring no human labelling, self-supervised cattle ID matches supervised accuracy.•Adding more cameras boosts the Re-ID system’s accuracy.•First multi-view on-farm cattle Re-ID dataset and source code released.