UAV (Unmanned Aerial Vehicle)-based Object Tracking with prototypical networks using Deep Learning

Unmanned Aerial Vehicles (UAVs) are increasingly utilized in fields such as surveillance, disaster management, and environmental monitoring, but object detection in UAV imagery faces challenges like varying altitudes, perspectives, occlusions, and environmental noise. This research introduces a novel multi-modal deep learning framework that combines a Kolmogorov–Arnold Networks (KAN)-based VGG-11 model with Prototypical Networks to address these challenges. The KAN- based VGG-11 model efficiently extracts high-dimensional feature representations from multi-modal inputs, while Prototypical Net- works enable few-shot learning, allowing the system to detect and classify new objects with minimal labeled data. This approach integrates visual to enhance detection accuracy and robustness in complex conditions. Evaluated on the UAVDT dataset, the pro- posed system demonstrates improved object detection accuracy, computational efficiency, and operational resilience compared to traditional CNN-based models, making it highly suitable for real-time UAV applications in diverse fields, including security, disaster response, and environmental monitoring.