TY - JOUR
T1 - Estimation of cattle weight from composite image/height/length data with spatial and channel attention convolution network (SCA-ConvNet)
AU - Lan, Leibin
AU - Shen, Lei
AU - Wang, Huaxia
AU - Yao, Yudong
AU - Zheng, Peng
AU - Mao, Axiu
N1 - Publisher Copyright: © The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature 2024.
PY - 2024/9
Y1 - 2024/9
N2 - Currently, the predominant method for indirectly obtaining cattle weight data involves establishing the correlation between cattle liveweight and measurable parameters, and subsequently calculating the liveweight of cattle using specific formulas. However, the factors considered in these methods are not comprehensive, which can easily lead to biased weight estimates. Additionally, deep learning-based cattle weight estimation methods typically rely on predicting weight from limited measurable parameters or images, limiting the potential for improved performance. In this study, we propose a novel cattle liveweight estimation algorithm based on composite image/height/length data and spatial and channel attention convolution network (SCA-ConvNet). By combining measurable parameters and images on the channel dimension, our algorithm is able to provide a more comprehensive set of learnable information. Furthermore, the incorporation of channel attention and spatial attention branches in the SCA-ConvNet enhances the network’s ability to extract valuable information from the composite data. Through multi-task prediction and loss calculation, our algorithm is able to fully leverage the diverse information available in the channels during training. The experimental results demonstrate that our proposed algorithm outperforms widely used methods including multiple linear regression, Xception, and ConvNeXt in terms of mean-absolute-error, root-mean-square-error, and mean-absolute-percentage-error for cattle liveweight estimation.
AB - Currently, the predominant method for indirectly obtaining cattle weight data involves establishing the correlation between cattle liveweight and measurable parameters, and subsequently calculating the liveweight of cattle using specific formulas. However, the factors considered in these methods are not comprehensive, which can easily lead to biased weight estimates. Additionally, deep learning-based cattle weight estimation methods typically rely on predicting weight from limited measurable parameters or images, limiting the potential for improved performance. In this study, we propose a novel cattle liveweight estimation algorithm based on composite image/height/length data and spatial and channel attention convolution network (SCA-ConvNet). By combining measurable parameters and images on the channel dimension, our algorithm is able to provide a more comprehensive set of learnable information. Furthermore, the incorporation of channel attention and spatial attention branches in the SCA-ConvNet enhances the network’s ability to extract valuable information from the composite data. Through multi-task prediction and loss calculation, our algorithm is able to fully leverage the diverse information available in the channels during training. The experimental results demonstrate that our proposed algorithm outperforms widely used methods including multiple linear regression, Xception, and ConvNeXt in terms of mean-absolute-error, root-mean-square-error, and mean-absolute-percentage-error for cattle liveweight estimation.
UR - http://www.scopus.com/inward/record.url?scp=85197493762&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85197493762&partnerID=8YFLogxK
U2 - 10.1007/s11760-024-03398-5
DO - 10.1007/s11760-024-03398-5
M3 - Article
SN - 1863-1703
VL - 18
SP - 7349
EP - 7358
JO - Signal, Image and Video Processing
JF - Signal, Image and Video Processing
IS - 10
ER -