Automatic individual identification of Holstein dairy cows using tailhead images

摘要

The implementation of dairy cow identification will be of great significance in precision animal management based on computer vision. In this study, a computer vision technique to identify the individual dairy cows automatically was proposed and evaluated. The tailhead image, which was used as a Region of Interest (ROI), was captured in a dairy farm. Zernike moments were used as descriptors of shape characteristics for the white pattern on the ROI. Two groups of Zernike moments were extracted from the preprocessed image and classified using four alternative classifiers, namely, linear discriminant analysis (LDA), quadratic discriminant analysis (QDA), artificial neural network (ANN) and support vector machines (SVM). The QDA classifier had the highest value, 99.7%, while the SVM classifier had the highest precision, 99.6%. Comprehensively, the QDA and SVM classifiers presented the best performance, with equal F-1 score of 0.995. These results show that the low-order Zernike moment feature, along with the QDA and SVM algorithms is an effective approach for individual dairy cow identification and has significant applications in precision animal management.