探地雷达多特征融合的城市空洞自动识别方法

Electromagnetic waves emitted by ground penetrating radar are easily attenuated by the external environment. Complex underground municipal facilities in urban areas further increase the difficulty of cavity detection. Currently, the amplitude feature in the time domain cannot fully reflect the structure and dielectric parameters of the cavity, leading to cavity omission and misjudgment during automatic detection. To fully utilize the multi-attribute information of the ground penetrating radar signal, improve the accuracy and efficiency of automatic detection of urban cavities, the amplitude, frequency, and phase features of the reflected signal at a specific time can be extracted, and the feature fusion can improve the accuracy of cavity detection. First, the signal is converted from the time domain into the time-frequency domain using the Hilbert transform, and the amplitude, frequency, and phase profiles at a specific time are obtained by time-frequency domain calculations. Four single-feature datasets containing the original profile are obtained. Second, IA+IF, IA+IP, IF+IP, and IA+IF+IP are fused using the 2D wavelet transform method. In addition, the maximum fusion rule is used in the high-frequency domain while the mean fusion rule is used in the low-frequency domain. Finally, the YOLOv7 algorithm is used to train on these 8 datasets, and the performances of the training models are compared. The results showed that the models trained on the IA+IP and IA+IF+IP datasets had better performance than the model trained on the OP datasets; the model trained on the IA+IP datasets showed the best performance; its precision, recall, F1_score, and AP_0.5 rates were 5.0%, 7.6%, 7.8%, and 5.9% better than those of the original profile, respectively. The proposed method can depict other detailed information beyond the amplitude of the cavity and strengthen the reflection characteristics of the signal at the location of a cavity, which can improve the performance of the automatic detection algorithm.