Auto-extraction of stratified interface in the underground space based on Bayesian detection algorithm with statistical fitting of probability density by actual data

•Focuses how to extract the interfaces of subsurface layered media automatically.•For avoiding the complex calculating the propagation of the electromagnetic wave in the subsurface layered media, the method of fitting probability density function is exploited by processing lots of actual GPR echo data.•The algorithm of the Bayesian detection is constructed by the fitted probability density function. According the feature of the data echo GPR in ice circumstances, decorrelation filter is design for the preprocessing.•The methods mentioned above can be extended to the other layered circumstances besides ice. There are many man-made or natural complex layered structures in underground space. The corresponding dielectric constants of different layers are different. The phase and amplitude changes of electromagnetic wave appear on the interface when these electromagnetic waves propagate in these layered media. Efficient extraction of these layered interfaces is one of the most important applications for non-destructive detection. It is very complicated to calculate phase and amplitude changes corresponding to the electromagnetic wave propagation near the interface in the medium using electromagnetic wave propagation and scattering theory. Based on a large number of measured data, we statistically analyze the amplitude and phase characteristics of the echo in different media, and give the fitted probability density function. The probability density function is used to calculate the likelihood ratio, so that the Bayesian algorithm is utilized to extract the interfaces of different media automatically. Taking the automatic extraction of the interface between the ice and the underground space as an example to verify the algorithm, we first perform the decorrelation filtering, and then the fitting of probability density function with lots of measured data, followed by Bayesian detection. And finally the layered media interface is extracted while suppressing the related noises.