Individual cow identification study based on DTW and K nearest Neighbor classification algorithm

Individual cow identification based on computer vision using deep convolutional neural networks has a serious dependence on the amount of sample data, and there are limitations in collecting a large amount of image data under unfavorable factors such as light changes and pasture environment. This study proposes an individual cow identification method based on the DTW algorithm and K-nearest neighbor classification algorithm from the perspective of the uniqueness and Differences of the behavioral characteristics of dairy cows' activities. The cow activity data collected from the sensors worn on the cow's neck were used as input, and the similarity feature matrix between cows was generated based on the DTW algorithm after denoising with moving average and S-G filters respectively, which reduced the dimensionality of the data and decreased the number of data, and finally the similarity feature matrix was input to the K-nearest neighbor classification model for individual cow identification. On the experimental dataset, its recognition rate after S-G filter processing was 91.08%, and the scores of precision, recall, F1 score, and Mathews correlation coefficient were better than those after moving average processing. The results of this study indicate that this method can be effective for individual cow identification.