EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

A Deep Learning Approach for Cattle Classification to Enhance Livestock Monitoring

Farah Kareem, Syed Umaid Ahmed, Muhammad Farrukh Shahid, M. Hassan Tanveer

International Journal of Innovations in Science and Technology · 16.05.2026 · 562

DOI: 10.33411/ijist/1802 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Innovations in Science and Technology, online erschienen am 16.05.2026. Verfasst von Farah Kareem und 3 weiteren Autor:innen. Die Arbeit nennt 30 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The classification of cattle from frontal face images is an important step toward automated livestock management, disease tracking, and optimization of farm productivity. In this study, we compare the performance of three deep learning architectures, Vision Transformer (ViT), Swin Transformer, and ResNet18, for large-scale, multiclass cattle classification. The proposed system is trained and evaluated on a cow dataset from Karachi, Pakistan, which includes approximately 459 distinct classes, making it one of the largest publicly available cow image datasets. All the models were trained and tested in similar experimental conditions, which ensures a fair comparison. The highest classification accuracy was achieved by the Vision Transformer with 96.27%, compared to 95.86% and 94.48% in both ResNet18 and Swin Transformer, respectively. In addition, the ViT model attained a macro precision of 0.94, recall of 0.95, and F1-score of 0.94, while ResNet18 achieved 0.94, 0.95, and 0.94, and Swin Transformer achieved 0.92, 0.93, and 0.92, respectively. The training process converged within 100 epochs, with final training and validation losses of 0.0176 and 0.2420 for ViT, 0.0228 and 0.2172 for ResNet18, and 0.0527 and 0.3159 for Swin Transformer, indicating stable learning behavior across models. The obtained results indicate that transformer-based architectures effectively capture fine-grained facial features in cattle compared to traditional CNNs. In addition, the Top-5 accuracies of all models were more than 99%, which highlights the appropriateness of all models to large-scale, multiclass cattle identification. Hence, the proposed work illustrates that these models can improve classification accuracy, which will aid in accurate livestock tracking, traceability, and help to identify diseases in their early stages, which will eventually increase productivity and the development of sustainable agriculture. The current research also fits in with the United Nations Sustainable Development Goals, SDG 2 (Zero Hunger) and SDG 12 (Responsible Consumption and Production), as it contributes to effective livestock management and sustainable agricultural practices.

Zitieren

APA

Kareem, F., Ahmed, S. U., Shahid, M. F., & Tanveer, M. H. (2026). A Deep Learning Approach for Cattle Classification to Enhance Livestock Monitoring. International Journal of Innovations in Science and Technology, 562. https://doi.org/10.33411/ijist/1802

Vancouver

Kareem F, Ahmed SU, Shahid MF, Tanveer MH. A Deep Learning Approach for Cattle Classification to Enhance Livestock Monitoring. International Journal of Innovations in Science and Technology. 2026:562. doi:10.33411/ijist/1802

BibTeX
@article{Kareem2026,
  author = {Kareem, Farah and Ahmed, Syed Umaid and Shahid, Muhammad Farrukh and Tanveer, M. Hassan},
  title = {A Deep Learning Approach for Cattle Classification to Enhance Livestock Monitoring},
  journal = {International Journal of Innovations in Science and Technology},
  year = {2026},
  pages = {562},
  doi = {10.33411/ijist/1802}
}
Kostenlos starten →