基于机器学习的奶牛颈环ID自动定位与识别方法.

[Objectives] Real-time perception and behavior analysis of individual cow information are the inevitable requirements of modern dairy cow fine breeding. Effective identification of individual cow identity is the premise and basis of the above goals. The contact-free recognition method based on the image of the cow’s biological characteristics(faces, body spots, etc.) is susceptible to external interference and the algorithm complexity is high, and the identifiable sample size is limited. Therefore, this paper proposes a method of automatic location and recognition of cow’s collar ID based on deep learning. [Methods] Aiming at the ID deflection problem of neck ring caused by cow movement, the cascade detector based on histogram of oriented gradient(HOG) feature combined with multi-angle detection method was adopted to realize the localization of cow signs. A single character image was obtained by a series of processing such as image enhancement and binary segmentation. The structure and parameters of the convolutional neural network were designed to train the character recognition model so as to complete the recognition of signage characters. The experimental data included 1 414 side-looking images of 80 cows, of which 58 were randomly selected as the training set and the images of the remaining 22 cows as the test set. [Results] The accuracy of placards was 96.98%,the recall rate was 80.23%,the accuracy of the character recognition model was 93.35%,and the recognition rate of individual cows in continuous image sequences was 95.45%. [Conclusions] The recognition model has good robustness to light change, stain contamination, rotation angle and so on, which has the potential to replace the traditional animal individual identification method. [ABSTRACT FROM AUTHOR]