Open Access

ARTICLE

Calf Posture Recognition Using Convolutional Neural Network

Tan Chen Tung¹, Uswah Khairuddin¹, Mohd Ibrahim Shapiai¹, Norhariani Md Nor^2,*, Mark Wen Han Hiew², Nurul Aisyah Mohd Suhaimie³

1 Malaysia-Japan International Institute of Technology, Universiti Teknologi Malaysia, Kuala Lumpur, 54100, Malaysia
2 Faculty of Veterinary Medicine, Universiti Putra Malaysia, Selangor, 43400, Malaysia
3 Faculty of Bioresources and Food Industry, Besut, 22200, Malaysia

* Corresponding Author: Norhariani Md Nor. Email:

Computers, Materials & Continua 2023, 74(1), 1493-1508. https://doi.org/10.32604/cmc.2023.029277

Received 01 March 2022; Accepted 20 May 2022; Issue published 22 September 2022

Abstract

Dairy farm management is crucial to maintain the longevity of the farm, and poor dairy youngstock or calf management could lead to gradually deteriorating calf health, which often causes premature death. This was found to be the most neglected part among the management workflows in Malaysia and has caused continuous loss over the recent years. Calf posture recognition is one of the effective methods to monitor calf behaviour and health state, which can be achieved by monitoring the calf behaviours of standing and lying where the former depicts active calf, and the latter, passive calf. Calf posture recognition module is an important component of some automated calf monitoring systems, as the system requires the calf to be in a standing posture before proceeding to the next stage of monitoring, or at the very least, to monitor the activeness of the calves. Calf posture such as standing or resting can easily be distinguished by human eye, however, to be recognized by a machine, it will require more complicated frameworks, particularly one that involves a deep learning neural networks model. Large number of high-quality images are required to train a deep learning model for such tasks. In this paper, multiple Convolutional Neural Network (CNN) architectures were compared, and the residual network (ResNet) model (specifically, ResNet-50) was ultimately chosen due to its simplicity, great performance, and decent inference time. Two ResNet-50 models having the exact same architecture and configuration have been trained on two different image datasets respectively sourced by separate cameras placed at different angle. There were two camera placements to use for comparison because camera placements can significantly impact the quality of the images, which is highly correlated to the deep learning model performance. After model training, the performance for both CNN models were 99.7% and 99.99% accuracies, respectively, and is adequate for a real-time calf monitoring system.

---

Keywords

---