Performance assessment of SVM-based classification techniques for the detection of artificial debondings within pavement structures from stepped-frequency A-scan radar data

Abstract

Owing to traffic and weather conditions, pavement structures may suffer from horizontal and vertical cracks that shorten the overall lifetime of roadways. In this paper, we focus on the detection of the horizontal subsurface cracks (debondings) occurring between the first two pavement layers from Stepped-frequency Radar (SFR) data. The processing of radar data requires some refined signal analysis to detect constructive interference between overlapping echoes. It is performed from timely data by a supervised machine learning method namely, Support Vector Machines (SVM) and compared to the conventional reference method, namely, the Amplitude Ratio Test (ART), which is routinely performed at the operational level. Besides, the straightforward application of SVM on raw data is compared to the solution using physical signal features (global and local) that reduces the computational burden. Performance assessment and comparison of the processing methods are conducted on data collected on a specific test-site with three different types of artificial debondings.