Feature-region recognition in ground penetrating radar images

摘要

Hyperbolic arc features in ground penetrating radar (GPR) images represent the most important evidence in identifying and locating underground pipes and other isolated,small-scale objects.Based on waveform features and other spectrum data,we propose a new method of extracting feature regions based on a gradient amplitude image.Routine pre-processing is performed to distinguish hyperbolic arc features from the background image.To highlight those regions containing hyperbolic arc features in GPR images and suppress background noise,we improved the differential valuing method by optimizing the spatial step in calculating the gradient amplitude image.Using the statistical gray-scale data of the gradient amplitude image,the proposed program auto-calculates the threshold used in distinguishing object pixels from background pixels by analyzing the histogram,and applies a binarization process to the gradient amplitude image.Using an algorithm of connected regions,we perform extension image segmentation for those regions containing hyperbolic arc features.Based on the area of the feature region,pseudo-feature-regions generated by noise are eliminated;the remaining regions are considered regions of interest.Information within the feature regions can then be used to identify and locate underground pipes and other isolated,small-scale objects from the GPR image.