Pattern spectrogram segmentation by base-model convolution

Pattern spectrogram segmentation by base-model convolution applied to signals withadditive noise

The present work establishes a convolutional method by spectrogram segmentation to detect discretized signals with a spectralpattern of interest and that is contaminated with additive noise, where detection by standard spectrogram tends to not be the mostefficient procedure. To achieve this objective, a three-step methodology is proposed: generation of a base model of patterns ofinterest with similar characteristics to be identified, extracting the norm of the discrete Gabor transform of the signal to be identifiedin convolution with the patterns of the base model, threshold the result of the previous step to limit the spectral content of the noiseand identify the resulting patterns. This method is compared with the traditional convolution of sine/cosine functions (spectrogram)and the advantages and disadvantages of both are highlighted. For validation, code in Matlab 2024a of the proposed algorithm isimplemented. Finally, comments are made on future work on the identification of gravitational wave signals immersed in noisysignals in which the methodology can be applied to find these patterns.

Link: https://repository.uaeh.edu.mx/revistas/index.php/icbi/article/view/13171