Image-Based Breed Recognition of Indian Cattle and Buffaloes Using YOLOv12 and Roboflow Cloud Platform

Abstract

India\'s livestock sector faces significant challenges in accurate breed identification, with traditional manual methods achieving only 65-70% accuracy while consuming 40-60% of Field Level Workers\' (FLWs) time. This paper presents BreedVision, an AI-powered web system for automated cattle and buffalo breed recognition using YOLOv12 object detection trained on 3,683 annotated images covering 15 major Indian breeds. The model achieves 69.8% mean Average Precision at IoU 0.5, with breed-specific accuracies reaching 87% for Bargur, 85% for Dangi, and 83% for Ongole and Alambadi. The system integrates ReactJS frontend with WebRTC camera capture, Python Flask backend with OpenCV preprocessing, and Roboflow cloud inference, providing real-time classification with confidence scoring and Excel export for government BPA system integration. Field testing demonstrates 40-60% reduction in FLW workload while improving data accuracy to 70% overall.

Introduction

India’s livestock sector, with over 193 million cattle and 109 million buffaloes, supports rural livelihoods and contributes 4.11% to GDP. Accurate breed identification is critical for breeding, insurance, and market pricing, but traditional manual methods are slow, error-prone, and unable to scale.

This research introduces BreedVision, an AI-powered web system for real-time livestock breed classification. Key contributions include: a dataset of 3,683 annotated images covering 15 breeds; YOLOv12 model achieving 69.8% mAP with breed-specific accuracies up to 87%; a web platform integrating ReactJS, Flask, OpenCV, and Roboflow; confidence scoring and Excel export for government integration; and validation showing a 40–60% reduction in field worker workload.

The system architecture uses a three-tier model (frontend, backend, cloud ML inference). Images are preprocessed and sent to Roboflow for YOLOv12 inference, returning breed labels, bounding boxes, and confidence scores. The web app displays results with color-coded confidence and allows data export.

Results show high accuracy for distinctive breeds like Bargur (87%) and Dangi (85%), while rare breeds with limited data had lower performance. Field testing demonstrated 45–60% time savings for frontline workers, with real-time inference within 1.2 seconds. The system is scalable, cross-platform, and user-friendly, providing a robust AI solution to improve livestock breed management and reduce human effort.

Conclusion

BreedVision demonstrates AI\'s transformative potential in modernizing livestock management. The system achieved 69.8% mAP with breed-specific accuracies up to 87%, reduced classification time from 5-10 minutes to <2 seconds, and enabled FLWs to process 25-30 animals daily versus 10-15 with manual methods. Data accuracy improved from 65-70% to 70% overall, supporting NDLM objectives and BPA integration.

References

[1] S. H. Mon et al., \"AI-enhanced cattle identification through back patterns,\" Nature Scientific Reports, vol. 15, pp. 12345-12358, 2024. [2] Kumar et al., \"Ensemble learning for cattle breed identification,\" J. Agricultural Technology, vol. 20, pp. 145-162, 2024. [3] M. Novak et al., \"Cattle identification methods using YOLOv8,\" Computers and Electronics in Agriculture, vol. 215, pp. 108-119, 2025. [4] Y. Zhang et al., \"Modern YOLO variants for livestock detection,\" IEEE Access, vol. 13, pp. 45678-45692, 2025. [5] D. Reis et al., \"Real-time object detection with YOLOv8,\" arXiv:2305.09972, 2023. [6] J. Redmon and A. Farhadi, \"YOLOv3: An incremental improvement,\" arXiv:1804.02767, 2018. [7] G. Jocher et al., \"Ultralytics YOLOv8,\" GitHub, 2023. [8] K. Simonyan and A. Zisserman, \"Very deep CNNs for image recognition,\" arXiv:1409.1556, 2014. [9] R. Johnson et al., \"Computer vision in precision livestock farming,\" Biosystems Engineering, vol. 234, pp. 89-112, 2024. [10] Ministry of Fisheries, Govt. of India, \"National Digital Livestock Mission,\" Tech. Rep., 2024. [11] NBAGR, \"Breed Purity Assessment System Guidelines,\" ICAR Tech. Rep., 2023. [12] P. K. Mishra et al., \"Indigenous cattle breeds of India,\" Indian J. Animal Sciences, vol. 93, pp. 875-892, 2023. [13] M. Rieke, \"WebRTC: Real-time communication,\" IEEE Internet Computing, vol. 18, pp. 8-14, 2014. [14] M. Grinberg, Flask Web Development, O\'Reilly Media, 2018. [15] Roboflow Inc., \"Computer Vision Platform Documentation,\" 2024.

Copyright

Copyright © 2025 Aashish Chauhan, Ajit Kumar Gupta, Garvit Paliwal, Gautam Singh, Harshit Singh, Dr. Umesh Dwivedi. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.