Amur tiger stripes: individual identification based on deep convolutional neural network.

The automatic individual identification of Amur tigers (Panthera tigris altaica) is important for population monitoring and making effective conservation strategies. Most existing research primarily relies on manual identification, which does not scale well to large datasets. In this paper, the deep convolution neural networks algorithm is constructed to implement the automatic individual identification for large numbers of Amur tiger images. The experimental data were obtained from 40 Amur tigers in Tieling Guaipo Tiger Park, China. The number of images collected from each tiger was approximately 200, and a total of 8277 images were obtained. The experiments were carried out on both the left and right side of body. Our results suggested that the recognition accuracy rate of left and right sides are 90.48% and 93.5%, respectively. The accuracy of our network has achieved the similar level compared to other state of the art networks like LeNet, ResNet34, and ZF_Net. The running time is much shorter than that of other networks. Consequently, this study can provide a new approach on automatic individual identification technology in the case of the Amur tiger.