Drone Classification Using Gated Recurrent Unit
Natthaphon Nakngoen, Pakkarawin Pongboriboon, Natthapat Inthanop, Jeeradet Akharachaisirilap, Tim Woodward, Nonthapat Teerasuttakorn · 2023
Nowadays, the trend of using unmanned aerial vehicles (UAVs) is increasing dramatically due to commercial, engineering, medical, logistical, security, and military applications. This paper proposes the use of Gated Recurrent Unit (GRU) to classify drones based on their sound, and methods to collect drone sounds for machine learning applications. This paper concludes that the performance of GRU is better than traditional RNN and is on par with papers using CNN and CRNN networks even though the GRU network has 4 times less parameters than RNN, CNN, and CRNN networks. We also found out that while collecting sound for machine learning applications, it is important to include environmental noise to make it more robust. Our approach is a strategic parameter optimization, allowing us to streamline the model's complexity and speed up training without compromising accuracy. The paper proceeds with an introduction to the motivation and background of the research, followed by a summary of related works and challenges in drone sound classification. We then describe the GRU architecture, parameter optimization, and the data collection process involving environmental noise. Experimental results and comparative analysis are presented, along with potential applications and future research directions. Overall, this research presents a compelling solution for drone sound classification, leveraging the advantages of the GRU model, optimizing its parameters for improve efficiency, and emphasizing the importance of environmental noise during data collection. We believe that our work will contribute significantly to the field of drone-related research and inspire further advancements in UAV classification and sound-based machine learning applications.