Estimation of beef marbling standard for live cattle using multi-input convolutional neural network with ultrasound images

We use ultrasound images of the loin, iliocostalis, and shoulder clod, the regions that experts examine when assessing beef marbling. A multi-input convolutional neural network combines information from these regions to estimate the beef marbling standard of live cattle.

The small deposits of fat within beef are known as marbling, and they influence the meat’s value. We study how to estimate the degree of marbling in a living animal from ultrasound images, which reveal structures inside the body. The target is the beef marbling standard, or BMS, a measure used to describe marbling.

We first carried out a field survey to learn what experts examine when estimating BMS. They looked at ultrasound images of several regions, particularly the loin, iliocostalis muscle, and shoulder clod. We therefore designed a learning model that takes images from multiple regions together, rather than relying on a single region.

The model uses a neural network that learns patterns in images. Its structure incorporates the experts’ approach by combining information from the different body regions. Experiments showed that this multi-input model could estimate BMS with high accuracy.

• Toshiki Katayama, Hirohumi Kawada, Masashi Nishiyama, Yoshio Iwai,
  Estimation of beef marbling standard for live cattle using multi-input convolutional neural network with ultrasound images, Proceedings of 15th International Conference on Quality Control by Artificial Vision (QCAV), GT3-9, Vol. 11794, pp. 309 - 316, May 2021. [Short movie]