<![CDATA[Body condition score assessment serves as a critical metric for evaluating the health, nutritional status, and overall well-being of beef cattle, playing a pivotal role in herd management and productivity optimization. Traditional manual BCS assessment methods are inherently subjective, labor-intensive, and impractical for large-scale operations, thereby necessitating an automated and data-driven approach. This study investigates the performance of YOLOv10-based deep learning models, incorporating the convolutional block attention module (CBAM) and spatial pyramid pooling-fast cross-stage partial connections (SPPFCSPC) to enhance feature extraction, classification accuracy, and computational efficiency in BCS estimation. A total of 432 annotated images representing five BCS categories (1–5) were used for model training and evaluation. The models were assessed using precision, recall, and F1 Score, with expert-labeled ground truth ensuring robustness. Results show that the YOLOv10x variant achieved the highest classification accuracy of 88.2%, highlighting its superior detection capability. YOLOv10m exhibited a balanced trade-off between accuracy and computational efficiency, achieving an F1 Score of 79.2%. The integration of CBAM improved precision but slightly reduced recall, whereas SPPFCSPC enhanced recall at the expense of increased computational complexity. Notably, YOLOv10n achieved the fastest inference time of 1.0 ms but with a lower accuracy of 82.4%, underscoring the trade-off between model depth and real-time applicability. These findings validate the effectiveness of attention-based and multi-scale feature learning strategies for improving the automation of BCS classification in beef cattle.]]>
Copyrights © 2025
