Single Node Detection on Direction of Approach
Sungyoun Seo, Seunghyun Yeo, Heejae Han, Youlim Ko, Kar Ee Ho, Eric T. Matson · 2020
While unmanned aerial vehicles (UAVs) present us with many benefits in the modern world, the immature use and malicious exploitation of this technology increase its potential to cause harm and fear. For that reason, in order to mitigate threats, the demand for technology that enables the detection of UAVs is ever increasing. Many studies focus on detecting the presence of UAVs and the direction of approach (DOA), but in order to overcome the issues caused by acoustic sensor limitations, researchers used a combination of sensors and nodes. This paper proposes a different method to overcome the limitation with a single node in place of the previous approach of using multiple and high-priced acoustic sensors. Within this study, data augmentation and Mel-frequency cepstral coefficients (MFCCs) were used in the dataset. We investigated convolutional neural network (CNN), convolutional recurrent neural network (CRNN) and ResNet-50 to identify UAV’s DOA on a Raspberry Pi 3 which has low computational resources. Our empirical results confirm that the CRNN classifier is more suited to operate in a real detection system since it can predict DOA in only 0.429 seconds with an accuracy of 97.60% on a Raspberry Pi 3.