The effect of segmentation on the performance of machine learning methods on the morphological classification of Friesien Holstein dairy cows

Abstract

Many classification algorithms are in the form of image pattern recognition; the approach to the complexity of the problem should be a feature of feasibility for representing images. The morphology of dairy cows greatly affects their health and milk production. The paper will apply several classification methods based on the morphology of Holstein Friesian dairy cows. To improve the accuracy of the model used, the segmentation process is the right step. In this paper, we compare several machine learning algorithms to get optimal accuracy. The algorithm used a support vector machine (SVM), artificial neural networks, random forests and logistic regression. Segmentation methods used are mask region-based convolutional neural network (R-CNN) and Canny; optimal accuracy is expected to create intelligent applications. The success of the method is measured with accuracy, precision, recall, and F1 Score, as well as testing by conducting a training-testing ratio of 90:10 and 80:20. This study discovered an artificial neural network optimal model with Canny with an accuracy of 82.50%, precision of 87.00%, recall of 79.00%, F1-score of 81.62%, and testing ratio of 90:10. While the models with the highest 80:20 ratio achieved 84.39% accuracy, 88.46% precision, 80.47%, and 83.00% F1-score with mask R-CNN with logistic regression.