Development of a new Light-Weight Convolutional Neural Network for acoustic-based amateur drone detection

Abstract

In recent years, interest in autonomous Unmanned Aerial Vehicles (UAVs) has increased, especially due to their use in different application areas, together with their ability to operate autonomously. The widespread use of UAVs also brings serious security threats. The prevention should be taken for early detection of UAVs-based threats, especially in military areas and governmental buildings, where high-level security is required. In this study, a sound processing algorithm with a new Light-Weight Convolutional Neural Network (LWCNN) model is proposed for the detection of UAVs, by distinguishing them from air noises, animal noises, and vehicle noises in environments with high spatial noise density. The most important advantages of the proposed model are its low complexity and a small number of parameters, as well as its high classification performance. The Mel Frequency Spectrum coefficients are applied to extract the sound features required to distinguish the sounds of UAVs. The features obtained from the proposed LWCNN model are used as input to the Support Vector Machines (SVM), where different objects are classified. With the proposed method, it is possible to detect UAVs and distinguish them from different sounds such as birds, airplanes, helicopters, background noises, and storms. Experimental results show that the proposed methodology obtains better results than other methods proposed in the literature.