Iterative Sound Source Recognition Method Based on Randomness Removal

The technology of acoustic imaging based on microphone array is crucial for optimizing acoustic structure and reducing noise pollution. However, traditional methods are limited by the Rayleigh criterion and half-wavelength theory, and existing approaches exhibit limited accuracy in detecting and localizing distant low-frequency acoustic emissions. The spatial resolution and recognition accuracy need to be improved, especially when the noise influence is large. To overcome these limitations, a grid fission method (R-VGSBL) based on randomness removal is developed. First, the initial source position is obtained under the Bayesian learning framework. Further, the correlation between multi-snapshot signals is used to retain the equivalent source position with a higher probability of non-zero elements in the multi-snapshots to remove random noise interference. Then, the grid position is fissioned and iterated, updated near the high-weight grid to form a new transfer matrix until the gradient monitor is less than the threshold, then the loop is jumped out and the output source is strong. Simulation experiments verify the performance of this method, and its recognition accuracy has been significantly improved compared with the comparative method.