Deep Learning-based drone acoustic event detection system for microphone arrays

In recent years, drones have brought about numerous conveniences in our work and daily lives due to their advantages of low cost and ease of use. However, they have also introduced significant hidden threats to public safety and personal privacy. Effectively and promptly detecting drone is thus a crucial task to ensure public safety and protect individual privacy. This paper proposes a method that combines beamforming algorithm with Deep Learning neural network to achieve the detection of drone acoustic event using microphone array technology. The aim is to achieve maximum coverage and accuracy in drone detection. The proposed approach utilizes beamforming algorithm to perform directional audio capture of the drone sound signal acquired by the microphone array. It then extracts features such as Log-Mel spectrogram and Mel-Frequency Cepstral Coefficients from the audio signal, which are subsequently input to a Convolutional Neural Network for classification. The final detection result is obtained through this process. The study also incorporates experimental analysis to assess the impact of different frontend processing algorithms, dataset compositions and feature selections on the detection performance. To provide a more specific and pronounced indication of the accomplishment of the drone sound event detection task, a novel evaluation criterion is introduced, termed as the Machine- Human Ultimate Distance Ratio. This criterion is employed to assess the detection effectiveness of the drone sound event detection task. The results demonstrate that the detection range and accuracy of the drone sound event detection system based on Deep Learning and microphone array surpass those of single-microphone sound event detection method. The proposed detection approach achieves effective detection within a range of up to 135 m in the surrounding environment.