In the field of civil engineering, Ground Penetrating Radar (GPR) is widely used to monitor structural integrity. With the help of Step-frequency Radar (SFR) for data acquisition and proper data processing algorithms, it is possible to detect small sub-surface defects within the pavement. In this paper, we discuss a conventional method based on the signal's amplitude, a supervised machine learning method (namely SVM) and a semi-supervised clustering based algorithm to detect said defects. The data are collected using an SFR at IFSTTAR's fatigue carousel where debondings are artificially introduced.