Abstract
Feature extraction is the key for detecting and identifying subgrade shallow targets of railways through ground-penetrating radar (GPR). Obtaining the appropriate feature to recognize subsurface targets is difficult due to the complication of subsurface structures and the diversity of target echoes. To identify those targets effectively and efficiently, this study proposes an energy statistical feature extraction method based on the directional feature of a target echo. Directional models of a typical target echo of shallow subsurface were initially built. The directional feature of different target echoes was discussed. The direction of the hyperbolic asymptote and horizon were the major directional features in the target echo. A target classification method based on echo direction was summarized. On the basis of the analysis of the relationship between curvelet coefficient and statistical features of energy in different directions, a feature extraction method was presented to form a feature vector subspace. Finally, target classification and recognition were achieved using the nearest neighbor method. Results show that the proposed method can effectively identify the preinstalled void disease of railway subgrade. The detection accuracy met the requirements of the roadbed disease identification of railways. The method used in this study was insensitive to the echo phase, and was suitable for detecting 0° and 180° phase target echoes. The proposed method provides a new means of identifying railway subgrade diseases and is significant in developing automatic technology for subsurface target detection based on GPR. Keywords: Ground-penetrating radar, Directional feature, Curvelet transform, Nearest neighbor method, Railway subgrade diseases
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have