采用短时傅里叶变换的铁路车载探地雷达数据解译方法

Data Interpretation of Ground-Penetrating Radar(GPR) via Short-Time Fourier Transform for Railway Track Detection

为了更好地实现探地雷达(GPR)在铁路路基检测中的应用,寻求一种用于识别路基不同病害特征的有效方法,克服传统人工解译方法的低效率和主观性,选取大同-准格尔铁路线中典型的雷达实测图像,采用短时傅里叶变换(STFT)方法,通过对信号进行时域和频域分析,确定合适的窗函数类型和长度,得到一组可用于对不同铁路路基病害进行分类的时频域图像,分析表明,图中不同深度处的能量分布特性与实际路基情况相符.然后,通过对图像形态特征的分析,得到一组典型的特征参数值,研究了不同目标信号在该特征参数空间的聚类特征,并将STFT方法与Welch功率谱估计法、时域局部能量法的计算结果进行了对比分析,结果证明STFT方法具有优越性,可以作为GPR数据分类的基础.

To make sure the wide applications of ground-penetrating radar（GPR） to railway road- bed inspection, an efficient method for recognizing different signal types gets necessary to over- come the low efficiency and subjectivity of the conventional artificial interpretation. Taking actual GPR data from Datong-Zhungeer railway as the example and analyzing the signals in time-domain and frequency-domain by short-time Fourier transform（STFT）, an appropriate window function with suitable type and length is deduced to obtain a set of time-frequency patterns for classifica- tion. The energy distributions in different depths of the patterns coincide with the corresponding ground truth. The classification, which focuses on the sample collection from samples of different signal types, is performed to investigate the clustering result by a selected planar eigenvectors. Compared with the conventional methods, such as the target recognition methods based on Welch algorithm or local energy analysis in time domain, the advantages of STFT for GPR data interpre- tation are demonstrated.