Automated dairy cattle identification:A YOLOv5-Siamese framework for Deoni breed

Abstract

Accurate identification of indigenous cattle breeds is important for livestock management, selective breeding, and conservation of valuable genetic resources. Deoni cattle, a recognized indigenous breed of India, exhibit considerable phenotypic variation that can make visual breed identification challenging and subjective. The present study investigated the feasibility of using a deep learning-based approach for automated Deoni cattle identification from images. A hybrid framework combining object detection for feature localization and a Siamese neural network for similarity learning was developed. A dataset comprising images of Deoni cattle, other cattle breeds, and non- cattle animal species was assembled and evaluated under varying field conditions, including differences in pose, illumination, and background. The model extracted discriminative visual features from facial and morphological characteristics and mapped them into an embedding space to distinguish similar and dissimilar images. Performance was assessed using standard classification and verification metrics. The proposed framework achieved effective discrimination between Deoni cattle and non-target animals within the available dataset and demonstrated consistent performance across different image conditions. However, the study was conducted using a relatively limited dataset, and further validation on larger and more diverse populations is required to establish broader generalizability. The findings indicated that deep learning-based image analysis can serve as a promising non-invasive tool for cattle breed identification and may support digital livestock record systems, breed characterization, and conservation initiatives.