Individual identification of dairy cows based on deep learning and feature fusion

Individual identification of dairy cows is one of the most important prerequisites for an intelligent dairy farming. Herein, a new method of individual identification of dairy cows based on the fusion of deep and shallow features of dairy cow's trunk image is proposed. Our aim is to obtain low-cost, touch-free, and high-accuracy individual identification of dairy cows. First, the model of VGG-16 is improved. Second, the multiple overlapping pooling layer in the improved VGG-16 model is used as a feature extractor to extract the features of dairy cow's trunk image, and the global average pooling (GAP) layer was used to replace the traditional flatten layer to reduce the dimensions of the extracted shallow and deep features. Finally, the features after the dimensionality reduction are fused using the Concat method and input the fusion of new features into the support vector machine (SVM) classifier for classification. The proposed method was trained and tested on the dataset of 3151 cow trunk images captured by us. Experimental results showed that this method not only has better recognition accuracy of 99.48% than the VGG-16, AlexNet, and ResNet-50 models but also has fewer model parameters, faster convergence speed, and stronger generalization ability.