基于 Transformer 的 东北虎体侧条纹个体识别.

The individual identification of Amur tiger（Panthera tigris altaica)，as the world’s largest Felidae animal and an endangered species, is a key step in answering many major questions in evolutionary biology. Although traditional methods such as iris and DNA analysis have been proposed for individual identification of the Amur tiger, these methods face challenges in remote acquisition and sample collection, and heavily rely on manual identification. With the development of computer vision technology, deep learning has become a powerful tool for animal individual recognition. In this study, a deep learning based approach was used for individual identification of the Amur tiger. We collected surveillance video images of 20 individual Amur tigers in the Heilongjiang Siberian Tiger Park and used the Mask R-CNN algorithm to automatically detect and segment the feature proposals in each image to construct the Amur tiger stripe dataset (ATSD). On the basis of this dataset, multiple classification networks based on CNN and Transformer were applied independently to identify individuals of Amur tigers. The results showed that the Transformer based classification network had a better performance on identification of Amur tiger stripes, with an accurate rate of 91. 49%. This method has good applicability to complex environments under reduced shooting conditions, and has the potential to expand ecological investigation and non-invasive sampling design, providing technical support for wildlife protection and management. [ABSTRACT FROM AUTHOR]