Real-time facial identification of Jersey cows using an edge computing device

Accurate individual identification of cattle is crucial for monitoring their productivity, health, and welfare, particular in the context of precision livestock farming. However, most existing cattle identification methods, such as ear tags and collars, are contact-based and can restrict the animals' movement, causing discomfort or pain. A promising alternative is machine vision, which offers a non-invasive and reliable approach to identify cattle, such as Holstein dairy cows. Nevertheless, in commercial farm settings, identifying a group of cows, particularly those without distinct coat pattern like Jersey cows, remains challenging. In this study, we propose and implement an improved YOLOv5-based method for real-time facial identification of multiple Jersey cows in the feeding area using cameras. Our approach achieves a remarkable accuracy of 95.1% when tested on the NVIDIA Jetson platform, running at 5 FPS. Furthermore, we present an application pipeline designed for the on-farm environment, enabling the monitoring of feed intake of individual cows along with facial identification. Overall, our machine vision and deep learning-based approach provides a non-invasive, non-contact, real-time, and accurate tool for identifying and monitoring each cow. This study lays the foundation for the development of an affordable smart cattle management system on farms, which has significant implications for precision livestock farming and animal welfare. © 2024 11th European Conference on Precision Livestock Farming. All rights reserved.