A Recognition Method of Cattle and Sheep Based on Convolutional Neural Network

Abstract

Facing the problem of low recognition accuracy caused by the confusing background information and low quality of monitoring images of automatic recognition for cattle and sheep animals, this paper proposes a convolutional neural network-based animal identification method for cattle and sheep. Thus, the recognition accuracy can be improved in the case of multiple images. First, the original image data is enhanced by randomly cropping, randomly inverting angles, and randomly horizontal rollback, and then a binary classification model for cattle and sheep recognition based on the VGG-16 convolutional neural network is built. Then the relevant hyperparameters will be continuously adjusted to increase the number of iterations. A higher recognition accuracy rate will finally be achieved. To verify the effectiveness of the method, this article adopted 260 and 110 cattle and sheep pictures respectively from open resources for training and testing. The experimental results showed that the highest recognition accuracy of the test set reached 96.67%, making the average accuracy rate as high as 90.95%, approximately 5.4% higher than the accuracy rate of other traditional VGG network models. This method showed faster speed and more extensive generalization, providing a practical technological reference for cattle and sheep recognition and binary classification problems.