Optimal Design of Feature Extraction Filter for UAV Acoustic Signal Recognition

The feature extraction of the U A V acoustic signal is the basis for its effective identification. In order to solve the mismatch between the UAV acoustic signal feature and the filter bank at this stage, according to the time-frequency distribution characteristics of the blade noise, a triangular bandpass filter bank NIFB is designed. Firstly, initialize the undetermined parameters of the decision vector of its center frequency expression, take the minimum value of multi-class cross-entropy as the objective function, and use multi-island genetic algorithm to optimize it. The convolutional neural network CNN-NIFB is established, and the feature extraction result is used as its image input, and the recognition rate is trained and calculated to verify the extraction effect. The results show that when the number of filter banks is the same, the recognition accuracy of NIFB is 2.3% and 3.7% higher than that of Mel filter bank and Gammatone filter bank, respectively. In order to verify the stability of the extraction algorithm, Gaussian white noise of different intensities was added to the original data. When the signal noise ratio of the original signal and the noise was higher than 8 dB, NIFB could keep the recognition rate above 90.5%.