Prediction of milk yield using visual images of cows through deep learning.

Abstract

The broad objective of the study was to determine, through deep learning, the predictability of milk yield from a cow's image data. The data size of 1238 image pairs (the side-view images and the rear-view images) from 743 Holstein cows within their first or second parity and the cows’ corresponding first lactation 305-day milk yield values were used to train a deep learning model. The data was first split into the training and testing data at the ratio of 80:20, respectively. The training data was then augmented four times more, then again split into training and validation data at the ratio of 80:20, respectively. Three principal analyses were done, i.e. the prediction of milk yield using rear-view images only, the prediction of milk yield using the side-view images only and the prediction of milk yield using a merge of the side-view and rear-view images (the combined-view images). In all three analyses, poor predictions were observed, i.e. R2 values of 0.32 for the model using the side-view image, 0.30 for the model using the rear-view images and 0.38 for the model using combined side and rear images. The mean absolute errors were 1146.4 kg, 1148.3 kg and 1112.9 kg for the side-view, the rear-view and the combined-view models, respectively. The root mean square error values were 1460.7 kg, 1480.5 kg and 1401.2 kg and the mean absolute error percentages were 17.6, 17.3 and 17.0 % for the side-view, rear-view and combined-view models, respectively. Hypotheses tests were also done to check whether there was any difference between these three prediction models. There was no significant difference in performance between all the prediction models (p>0.05), i.e. the side-view model, the rear-view model and the combinedview model. It was concluded that predicting 305-day milk yield of Holstein cows using either view has the same level of accuracy and no additional benefits are derived from using both the rear and the side views. Keywords: Computer vision; deep learning; linear conformation traits; 305-day milk yield; side-view images; rear-view images; combined-view images; Holstein cows.