Acoustic Drone Detection Based on Transfer Learning and Frequency Domain Features
Mohamad Yaacoub, Hadeel A. Younes, Mostafa Rizk
Abstract
Currently, drones are widely used in various sectors because of their affordability, flexibility and ability to carry payloads during their flights. Nevertheless, drones are excluded from several regions. Although anti drone systems, which are based on radar technology or visual detection, are deployed, drone intrusions are still recorded due to their small size and ability to maneuver. In this paper we investigate the detection of drones based on their acoustic signature and exploiting the advances of deep learning. Convolution neural networks (CNNs) are adopted to recognize drone sounds. Transfer learning is used to fine-tune pre-trained CNN on a custom acoustic dataset to classify sounds and detect drone acoustic features. The obtained results demonstrate the effectiveness of our proposed approach as a promising solution to identify the presence of a drone. A mean average precision of 0.88 has been achieved when testing the trained CNN on unseen sound recordings.