Deep Learning-Driven Cattle and Buffalo Classification Using CNN- YOLO Hybrid Framework for Smart Livestock Management

Abstract

Accurate identification and classification of livestock animals is a fundamental requirement in modern agricultural management. This paper proposes an image- based deep learning framework to automatically distinguish between cattle and buffaloes using Convolutional Neural Networks (CNN) integrated with a YOLO-based object detection pipeline. A curated dataset of 2,500 annotated images— comprising 1,300 cattle and 1,200 buffalo images—was utilized for training and evaluation. The preprocessing pipeline applies resizing to 224×224 pixels, normalization, and data augmentation techniques including random rotation, horizontal flipping, and brightness adjustment to enhance model generalization. The CNN model extracts multi-scale hierarchical features, while YOLO localizes individual animals within cluttered farm backgrounds prior to classification. Experimental evaluation demonstrates an overall classification accuracy of 93.2 %, with precision of 0.92 and recall of 0.94. The proposed system significantly reduces dependence on manual observation and supports real-time, automated livestock monitoring. These results confirm the viability of deep learning for scalable and accurate animal type classification in smart farming environments