Processing and preparation of observation data in the interests of highlighting the features of the dynamics of the characteristics of geoacoustic emission

The lithospheric layer deformation under the action of seismic processes affects the characteristics of geoacoustic emission. The study of the geoacoustic emission dynamics is aimed at finding signs of preseismic events. There is a problem obtained for the high-quality processing of geoacoustic emission signals and the results classification. The study is aimed at finding the best combination of pre-processing and clustering tools for the pulse flow of geoacoustic emission to identify the features of the characteristics dynamics of such a signal. The processed signals were obtained during long-term measurements in the surface lithosphere layers of the seismically active region of the Kamchatka Peninsula. To identify the variability features of geoacoustic emission signals characteristics they are converted by sructurno-linguistic into a three-dimensional image. The images are processed, compared and clustered using convolutional neural networks of various architectures. The best result is assessed by three selected quality criteria. A technique has been developed for finding the best preprocessing and clustering result. The experimental result analisys are presented.