Development of a Dairy Cattle Facial Recognition System Based on CNN for Digital Livestock Data Management

Abstract

Dairy cattle data management plays a crucial role in supporting operational efficiency and the sustainability of the livestock sector. However, conventional practices at Mamad Jaya Farm in animal identification still present severe operational and animal welfare challenges. The existing traditional marking methods carry high physical risks for the livestock, frequently leading to torn ears from ear-tagging, severe skin irritation from branding, and high human-error risks in manual paper-based record-keeping. To mitigate these invasive drawbacks, this study aims to develop an artificial intelligence-based facial recognition system for dairy cattle as a non-invasive solution to support digital and integrated livestock data management. A Convolutional Neural Network (CNN) architecture was uniquely implemented for the identification process due to its superior capacity to automatically and hierarchically extract complex spatial biometric features from facial images without manual feature engineering. The research methodology involved collecting a multi-angle facial dataset from a herd of 15 dairy cattle at Mamad Jaya Farm, Karangpawitan, Garut, which was then processed using Roboflow. The developed model was integrated into a web-based livestock platform named BonvaLink. Empirical testing on 15 distinct cattle facial images demonstrated that the system achieved an individual identification accuracy of 93.33%, with the majority of correct predictions yielding robust confidence scores. These results indicate that the CNN-based biometric approach is highly effective and reliable in recognizing individual dairy cattle identities under practical barn environments