Abstract: The hidden dangers of road collapse and underground voids are usually difficult to notice and easy to occur in sudden. They have caused huge losses in people's lives and properties. Therefore, the detection and analysis of road collapse and underground voids are crucial. The ground-penetrating radar (GPR) is the main method for detecting the hidden dangers of urban road collapse currently by virtue of its high accuracy, high efficiency, continuous non-destructiveness and real-time imaging. A convolution type objective function FWI algorithm was proposed to solve the problem of the lower inversion accuracy and practicability caused by the inaccurate estimation of GPR conventional objective function full-waveform inversion (FWI). For the two situations, i.e., road collapse and underground voids, synthetic data models were established, and compared with the inversion results obtained using the conventional objective function FWI, which indicated that the convolution objective function FWI algorithm can still provide favorable inversion results under the condition of inaccurate excitation source estimation. Accordingly, the effectiveness of the algorithm was verified. Finally, the algorithm was used in the GPR measured data of the two groups of different defect types, and the relative dielectric constant distribution of the underground medium was obtained through analysis and inversion, which verified the practicability of the convolution type objective function FWI algorithm for measured data. In this way, the theoretical basis related to the underground anomalous body detection for the subgrade was provided, and guidance was offered for practical production.