Article

A deep learning-based approach for feeding behavior recognition of weanling pigs

MinJu Kim1, YoHan Choi2, JeongNam Lee3, SooJin Sa2,*, HyunChong Cho4,**
Author Information & Copyright
1The University of Queensland, Queensland 4072, Australia.
2Rural Development Administration, Cheonan 31000, Korea.
3Kangwon National University, Interdisciplinary Graduate Program for BIT Medical Convergence, Chuncheon 24341, Korea.
4Kangwon National University, Dept. of Electronics Engineering and Interdisciplinary Graduate Program for BIT Medical Convergence, Chuncheon 24341, Korea.
**Corresponding Author: SooJin Sa, Rural Development Administration, Cheonan 31000, Korea, Republic of. E-mail: soojinsa@korea.kr .
**Corresponding Author: HyunChong Cho, Kangwon National University, Dept. of Electronics Engineering and Interdisciplinary Graduate Program for BIT Medical Convergence, Chuncheon 24341, Korea, Republic of. E-mail: hyuncho@kangwon.ac.kr.

© Copyright 2021 Korean Society of Animal Science and Technology. This is an Open-Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/4.0/) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.

Received: Nov 11, 2021; Revised: Nov 16, 2021; Accepted: Nov 16, 2021

Published Online: Nov 23, 2021

Abstract

Feeding is the most important behavior, which shows the health and welfare of weanling pigs. Early detection of feed refusal is crucial for control of disease in the initial phase or detection of empty feeder to add feed timely. This paper proposes a real-time technique to progress detection and recognition of small-size pigs using a deep-learning-based method. The proposed model focuses on detecting pigs on the feeder in a feeding position. The conventional methods detect pigs and classify them to different behavior gestures; however, in our proposed method, these two categories are combined into one process to detect only feeding behavior in order to increase the speed of detection. Considering the high difference between pig behavior and size, adaptive adjustments are introduced to YOLO, including an angle optimization strategy between the head and body to detect the head in the feeder. According to the results, this method can detect the feeding behavior of pigs and screen the non-feeding positions with 95.66, 94.22, and 96.56 % average precision (AP) at 0.5 IoU threshold with YOLOv3, YOLOv4, and additional layer and suggested activation function, respectively. The drinking behavior was detected with 86.86, 89.16, and 86.41 % AP at 0.5 IoU threshold in YOLOv3, YOLOv4, and proposed activation function, respectively. In terms of detection and classification, the result of our study the proposed method shows a higher precision and recall compared with conventional methods.

Keywords: Convolutional neural network; Deep learning; Behavior detection; processing; Weanling pig