Cattle Breed Identification Using Deep Learning: A Comparative Study of ResNet50, EfficientNet-B0 and Vision Transformer
Abstract: Accurate cattle breed identification is important for livestock management, breeding programs, genetic conservation, and agricultural applications. Manual identification based on physical characteristics can be time-consuming and requires expert knowledge. This study presents a deep learning-based approach for automated cattle breed identification using images of 41 cattle breeds. A dataset containing 5,946 images was prepared and divided into training, validation, and testing sets. Images were resized to 224 ? 224 pixels and appropriate preprocessing and augmentation techniques were applied. Three pretrained deep learning architectures, ResNet50, EfficientNet-B0, and Vision Transformer (ViT), were trained using transfer learning and evaluated on the test dataset. The obtained test accuracies were 87.20% for ResNet50, 90.55% for EfficientNet-B0, and 94.87% for Vision Transformer. Among the evaluated models, ViT achieved the highest classification accuracy. A Gradio-based graphical user interface was also developed to allow users to upload cattle images and obtain breed predictions. The results demonstrate the potential of deep learning-based approaches for automated cattle breed identification while highlighting the challenges associated with fine-grained classification across visually similar breeds.
Keywords: Cattle Breed Identification, Deep Learning, Image Classification, Vision Transformer, Transfer Learning
How to Cite?: Dr. Sheik Imran, Namratha V, Mahima Danappa Kasambi, Nagashree C N, Meghana Venkatesh Telkar, "Cattle Breed Identification Using Deep Learning: A Comparative Study of ResNet50, EfficientNet-B0 and Vision Transformer", Volume 15 Issue 10, October 2026, International Journal of Science and Research (IJSR), Pages: 1258-1262, https://www.ijsr.net/getabstract.php?paperid=SR261019115733, DOI: https://dx.doi.org/10.21275/SR261019115733