Deep Learning-Based approach for Detection and Measurement of Challenging Linear Traits in Sahiwal Cows

Abstract

 Accurate selection of high-yielding dairy animals is essential for improving productivity and guiding future breeding programs. Linear type traits, which refer to specific anatomical features of cattle, have long been used to estimate milk production potential and assist in breed identification. However, certain traits—such as hoof angle and hump area—are particularly challenging to measure manually, often leading to subjective errors and inconsistencies across individuals. Literature suggests that while visual estimation of linear traits may suffice for breed prediction, actual measurement is crucial for assessing breed purity and production-related characteristics. In this study, we present a deep learning-based approach using the YOLOv8 object detection model to automatically detect and accurately measure the hoof angle and hump area of Sahiwal cattle from single-animal images. A dataset of 200 annotated Sahiwal cow images was used to train the model, achieving high precision in trait localization and quantification. Hoof angle, which influences reproductive performance, and hump area, a key indicator of breed purity in Sahiwal cattle, were specifically targeted due to their difficulty in manual measurement. This work offers a reliable and automated solution to overcome the limitations of manual assessment, contributing to improved livestock evaluation through advanced computer vision techniques.