Acoustical unmanned aerial vehicle detection in indoor scenarios using logistic regression model

Abstract

In this study, the data obtained from the acoustic measurements were used to train a model based on logistic regression in order to detect a quadrotor’s vehicle in indoor environment. To simulate a real environment, we made sound recordings in a shopping center. The sounds related to two scenarios were recorded: only anthropic noise and anthropic noise with background music. Later, we reproduced these sounds in an indoor environment of the same size and characteristics as the shopping center. During the simulation test, a drone placed at different distances from the sound level meter was turned on at different speeds to identify their presence in complex acoustic scenarios. Subsequently, these measurements were used to implement a model based on logistic regression for the automatic detection of the unmanned aerial vehicle. Logistic regression is widely used in pattern recognition of the binary dependent variable. This model returns high value of accuracy (0.994), indicating a high number of correct detections. The results obtained in this study suggest the use of this tool for unmanned aerial vehicle detection applications.

Investigation of Metrics for Assessing Human Response to Drone Noise

“…Schäffer et al [ 15 ] also point out pure tones and high-frequency broadband noise as important contributors to drone noise annoyance. Tonal and high-frequency drone noise is also likely to increase its noticeability in existing soundscapes [ 19 , 20 , 21 ] and therefore lead to noise annoyance. None of these sound features are appropriately accounted for in current aircraft noise metrics, neither L Aeq -related nor even more sophisticated metrics, such as the effective perceived noise level (EPNL) [ 22 ].…”

Duck Eggshell Crack Detection by Nondestructive Sonic Measurement and Analysis

“…Logistic regression is a statistical method for modeling the probability of a binary dependent variable, and it is utilized in supervised learning methods [26]. It can also be used to consider the influence of several factors on the probability of any event [27][28][29], and this is used in acoustics [30][31][32].…”

Drone Model Classification Using Convolutional Neural Network Trained on Synthetic Data

“…Ciaburro et al [ 6 ] performs a study to detect a flying drone indoors. Iannace et al [ 7 ] expands on it by using logistic regression, and reports a drone detection accuracy of 99.4% at distances of up to 9 m in a noisy environment. Jamil et al [ 8 ] use a combination of audio and visual features to detect drones, thunderstorms, birds and planes.…”