Optimising automatic cattle weight estimation based on computer vision through animal temperament assessment

Conventional cattle weighing approaches are labour-intensive, time-consuming, and can cause stress, which negatively affects animal health and production. Automated, contactless methods using computer vision and machine learning technologies have been proposed to address these issues. These methods enable the accurate estimation of animal weight by analysing RGB-D images from a depth camera, thereby eliminating the need for manual weighing and reducing stress in animals. Temperament assessments in beef cattle, particularly exit velocity and flight speed, provide critical insights into growth performance, feed efficiency, and carcass quality. The potential of incorporating speed features into morphometric data can lead to a stable improvement in the quality of the live cattle weight prediction model. This study aimed to develop an algorithm for estimating cattle movement speed using the YOLOv8n neural network on RGB-D images, integrate these speed features with traditional morphometric data, and evaluate the resulting improvements in predicting live cattle weight. This study showed that for the ExtraTrees model, R 2 increased by 0.04 on the test, while MAE decreased by 1.9 kg. The proposed model is a promising approach for practical application due to the combination of high performance and automation.