Deep Learning Techniques for Quadcopter-Based UAV Systems: Applications and Challenges

Abstract

Unmanned aerial vehicles (UAVs), particularly quadcopters, integrated with deep learning techniques have transformed aerial monitoring across diverse domains, including security surveillance, target tracking, traffic monitoring, precision agriculture, and disaster management. This review examines the evolution of object detection methods from traditional handcrafted feature-based approaches to modern deep learning frameworks, highlighting the development and application of one-stage and two-stage detectors in UAV imagery. Key challenges associated with deploying deep learning on UAV platforms, including limited onboard computational resources, energy constraints, communication latency, complex backgrounds, small-object detection, class imbalance, and arbitrary object orientations, are critically discussed. The review further explores recent applications of UAV-based deep learning systems in traffic surveillance, crop health assessment, flood and wildfire monitoring, and post-disaster damage evaluation. In addition, operational considerations such as obstacle avoidance, battery management, adverse weather conditions, flight control, noise suppression, and communication range are examined. Overall, the reviewed literature demonstrates that the combination of quadcopters and deep learning provides accurate, scalable, and cost-effective solutions for intelligent aerial monitoring. Future advances in lightweight models, edge computing, multisensor fusion, and autonomous decision-making are expected to further enhance the performance, reliability, and real-world deployment of UAV-based monitoring systems.