Time-Frequency Analysis and Recognition for UAVs Based on Acoustic Signals Collected by Low-Frequency Acoustic-Electric Sensor

The utilization of unmanned aerial vehicles (UAVs) is progressively expanding, thereby introducing novel risks to public safety and even the battlefield environment. Consequently, it becomes imperative to detect noncooperative UAVs. The employment of UAV sound signals for identifying nearby UAVs has garnered significant attention due to its comprehensive coverage, independence from obstructions such as trees and cost-effectiveness. However, practical implementation is hindered by the rapid attenuation of sound, limited acquisition range, and insufficient sound dataset. Therefore, urgent research is required on the generation and propagation of UAV sounds in order to enhance sensor performance and signal processing algorithms while enriching the available UAV sound dataset. This article analyzes the time-frequency characteristics of UAV sound generation and propagation. On this basis, a low-frequency acoustic sensitive acoustic sensor based on piezoelectric ceramic double vibrator is adopted. The sensor has an amplification effect on the low-frequency high-energy signals emitted by UAV, and the detection distance reaches 500 m. Then, UAV sound datasets at different altitudes are collected, and a sparse Mayer filter is designed to generate the time–frequency feature map. In order to reduce the influence of environmental noise, vehicle noise, bird chirp, rain, and other environmental sounds are collected as negative samples. The deep residual network is used to train and identify UAV sounds, and the accuracy of the test set reaches 99.7%. This system can accurately identify whether there is a UAV in the range of 500 m, which has important value in the field of security and other fields.