A Novel Approach To Lion Re-Identification Using Vision Transformers

In recent years, technology has played a crucial role in wildlife and ecosystem conservation, significantly decreasing the time and effort required for wildlife monitoring, population estimation, poaching prevention, habitat mapping and, specifically, animal re-identification. This study proposes a novel approach that can be used in wildlife research and conservation to track individual animals over time by employing deep learning techniques. The challenges associated with animal re-identification in diverse natural environments such as variability in appearance, species diversity, accuracy and reliability can be addressed by leveraging the capabilities of advanced deep learning models, namely Vision Transformers and Convolutional Neural Networks. When trained on a Lion wildlife dataset, the Vision Transformer demonstrated significantly better performance compared to the Convolutional Neural Network in terms of accuracy, precision, recall and training time. This research contributes towards ongoing ecological initiatives to improve population monitoring, anti-poaching efforts, and habitat protection.