Computer Vision for Biodiversity Monitoring: Improving Accuracy in Tracking Endangered Species of Wildlife in Impacted Ecosystems

The relentless destruction of ecosystems, climate change, and human interference has put biodiversity at risk, resulting in the swift downfall of already endangered species within affected regions. Conservation of biodiversity is crucial for the protection of ecosystems, yet, tracking biodiversity using manually conducted field surveys or camera trapping techniques is highly inefficient in terms of time, labor, and accuracy. This article proposes a novel approach for improved tracking of endangered species using AI driven data fusion for multi-modal deep learning architecture. This model employs remote sensing techniques such as LiDAR, audio, and RGB and thermal imaging to reinforce species detection, behavior analysis, and mitigation of localization issues. For object detection and tracking, the proposed system implements the utilization of YOLOv8 + DeepSORT, while image recognition is solved using Vision Transformers (ViTs) and species interaction analysis is conducted using hybrid CNN-GNN models. In addition, the model employs Contrastive Learning and Meta-Learning methods in the rare samples, along with Federated Learning that allows for model training on various data sets without compromising privacy. The results obtained from the experiments reveal that the proposed framework is remarkably more efficient than the current existing methods.