Optimal Deep Learning Model for UAVs: A Case Study

Unmanned Autonomous Vehicles (UAVs), particularly autonomous flying drones, have garnered significant attention in the field of Artificial Intelligence. Advancements in electronic technology, marked by smaller, cheaper, and more efficient components, have led to substantial progress in UAV research. These versatile devices find applications in diverse fields, ranging from flood monitoring and algae spread detection in water bodies to forest trail identification. In this work, we focus on autonomous flying drones, conducting a comprehensive case study to assess the efficiency, robustness, and accuracy of UAVs. Our research is substantiated through a series of experiments, providing a detailed analysis of the software and hardware architecture employed in the study. We present the implementation algorithms and compare three state-of-the-art approaches: TrailNet, InceptionResnet, and MobileNet, in terms of accuracy, robustness, power consumption, and inference time. Our study reveals that MobileNet outperforms the other algorithms, offering superior results with minimal computational requirements and power consumption. Additionally, we address the challenges encountered during our research and discuss potential avenues for enhancing safety features and overall performance in future work.