Geophysical data fusion of ground-penetrating radar and magnetic datasets using 2D wavelet transform and singular value decomposition

Abstract

This work addresses the problem of the lack of perceptibility that geophysical data may have. Data fusion allows us to combine datasets, providing an improved and more informative source of information about structures buried in the ground. After testing different approaches, a strategy was developed using ground-penetrating radar and magnetic datasets collected over the same area. Data collected at the Roman Villa of Pisões (Beja, Portugal), which is a place of easy application of geophysical methods, were used to test the method, but with problems caused by the properties of the soil. The approach was based on processing operations that allow the fusion of images obtained by different equipment widely used in medical imaging for tumor detection and image processing. The goal is to create an improved image with data fusion that has higher quality than the input images, allowing a better understanding of the object of the study. The approach is composed of two stages: pre-processing and data fusion. Pre-processing is applied to enhance the input data. It consists of removing background noise through singular value decomposition applied in the spectral domain. Then the calculation of the data entropy will highlight the differences corresponding to the spatial alignments compatible with buried structures. Then, both entropy maps of the two datasets are fused in the second processing step to produce the final image. This step involves applying the 2D wavelet transform to each entropy map, decomposing them into sub-bands. Algorithms to calculate multiresolution singular value decomposition and the image gradient are applied to the sub-bands. The processed sub-band pairs are then fused using specific fusion rules. The fused image is obtained by applying the inverse of the wavelet transform. Data fusion with the proposed approach allows us to obtain a detailed image that is sharper and of better quality than the input datasets. The increase in sharpness and quality can be quantified through the sharpness index and the BRISQUE quality index in several steps of the processing. The obtained values confirm the graphical results. Images produced by the proposed data fusion approach suggest that the perceptibility has increased, allowing us to provide conclusions about the existence of buried structures.