Individual identification of wild raptors using a deep learning approach: A case study of the white-tailed eagle

Individual identification is essential for elucidating animal population structures, tracking population dynamics, and uncovering social networks. Advances in computational technology have enabled the application of deep learning-based methods for individual wildlife identification. However, accurately identifying individual animals in complex wild environments remains a significant challenge. Motivated by the need for accurate and efficient identification of individual animals in the wild, a deep learning-based individual identification framework, the object tracking–face extraction–sampling–recognition (OFSR) approach, is proposed. This framework uses deep learning to extract facial features and a multitask module with cross-task information sharing to integrate supplementary data, enhancing individual identification accuracy. By employing the OFSR framework, we identified individual white-tailed eagles in the Jingxin Wetland during the overwinter period. Our results demonstrated that the OFSR framework could accurately identify individual white-tailed eagles in wild environments, achieving an accuracy exceeding 93 %. In addition, in the multitask module of the OFSR framework, age recognition is used to increase the individual identification accuracy, successfully separating recurring and new individuals and increasing the accuracy by 2 % without adding extra costs. Our results demonstrate the potential of deep learning in identifying individual animals in complex wild environments, and the proposed OFSR framework is universally applicable to other raptors. The findings highlight that the added multitask module increases the accuracy of identifying individual animals. Our framework could improve the accuracy of identifying individuals in complex wild environments, offering a promising method for population detection and conservation research involving wild animals.