Computer Vision (CV) algorithms and Federated Learning frameworks for Animal Monitoring
Monitoring wild animals in agricultural areas is crucial for sustainable management and wildlife preservation, as they offer substantial obstacles. Using new deep learning algorithms for real-time animal spotting can improve wildlife monitoring efforts. However, these models need to be kept up-to-date, constantly trained on newer data, especially when they are prone to environmental shifts depending on the location in which they are deployed. This research work proposes exploring state-of-the-art and novel object detection models with emphasis on diverse deep learning algorithms, as well as data augmentation, hyperparameter fine-tuning. Additionally, as models will be deployed in numerous locations, model consistency across the edge and far-edge devices is necessary, prompting the need for inter-device communication and adopting methods such as federated or aggregated learning for model weight averaging, aiming for constantly refined model with minimized power and computational impact.
Back to Alumni