Adaptive filtering of UAV acoustic signals based on MFDE-LOF fundamental frequency filtering and PSO-VMD harmonic reconstruction

Abstract

The rapid proliferation of drone technology has significantly impacted public security, necessitating the development of advanced Unmanned Aerial Vehicle (UAV) filtering techniques. This paper proposes an innovative adaptive filtering method for UAV acoustic signals in diverse noisy environments. Optimizing the Local Outlier Factor (LOF) parameters of Multiscale Fluctuation Dispersion Entropy (MFDE) scatterplots enables the removal of outlier points, thereby filtering out noise fundamental frequencies in the Time-frequency spectrum. Furthermore, Variational Mode Decomposition (VMD) with Particle Swarm Optimization (PSO)-optimized parameters is utilized to adjust the center frequency and bandwidth, enabling precise separation of acoustic signals. Intrinsic Mode Functions (IMFs) are selected based on the harmonic characteristics of UAVs for signal reconstruction, achieving adaptive filtering. Experimental results utilizing three UAVs with nine distinct noise types demonstrate notable performance enhancements, with the average Signal-to-Noise Ratio (SNR) increasing by approximately 5 dB, 8 dB, and 12 dB under 0 dB, -5 dB, and -10 dB noise conditions respectively, while spectral Cosine Similarity (CS) improves by about 0.4, 0.5, and 0.6 on average.