A Multi-feature Extraction Algorithm for Battlefield Acoustic Target Recognition

In view of the problem that the widely used single analysis domain feature extraction algorithm in battlefield acoustic target recognition often gives rise to some important information missing, and low recognition, therefore, the generation mechanism of battlefield acoustic signal was analyzed. It can be concluded that acoustic signal includes both randomness machine noise and quasi-periodic aerodynamic noise, based on this, a multifeature extraction algorithm was proposed by combining wavelet packet analysis with discrete spectrum analysis. The energy distribution of non-uniform frequency bands was extracted by means of the wavelet packet transform, combined with quasi-periodic feature in the time domain described by the discrete spectrum, thus the feature parameter was obtained, which reflected the characteristic of the target signal more comprehensively. The experimental results show that the accuracy and robustness of the multi-feature extraction algorithm applied to acoustic recognition are improved obviously, compared with that of single analysis domain feature extraction algorithm.