基于频谱能量分析的地质雷达探测图像判读

Interpretation Based on Frequency Spectrum Energy Analysis of Ground Penetrating Radar Detection Image

通过人工肉眼对地质雷达探测图像进行判读的方法易受判读人员主观性、经验性影响。为了规避这一不足,提出一种基于Counterlet等高变换和K-Means++均值聚类分析的频谱能量判读方法。以实际公路隧道为依托,经现场探测获取不良地质体的原始探测数据;采用IDSP(interactive detection-surveying prediction)探测数据分析软件生成原始图像,实施背景去除、滤波等时域、频域预处理以提高信噪比;基于子带分布系数采用Counterlet对预处理后的图像进行分解和重构,并采用K-Means++算法将重构后图像中的频率信息转化为颜色特征;利用MATLAB对颜色特征进行提取,并据此建立不良地质体颜色特征样本库,将原始探测图像与样本库进行匹配对比以实现自动判读。结果表明:采用Counterlet等高变换对多方向、多分辨率、多尺度的地质雷达图像进行分解与重构是可行的,曲线边缘逼近效果良好,重构后的图像无信息丢失;K-Means++算法能实现地质雷达灰度图像中能量—频率—色彩的转化,转化后的图像色彩突出、直观;频谱能量的匹配对比能较准确快速地实现自动判读及较好地规避个体主观性。

The interpretation method of GPR(ground penetrating radar)detection image by human eye is easily affected by the subjectivity and experience of interpreters.In order to avoid this shortcoming,a spectral energy interpretation method based on Counterlet contour transform and K-Means++clustering analysis was proposed.Based on the actual highway tunnels,the original detection data of unfavorable geological bodies were obtained through on-site detection.IDSP(interactive detection-surveying prediction)detection data analysis software was used to generate the original image,and time domain and frequency domain preprocessing such as background removal and filtering were implemented to improve the signal-to-noise ratio.Based on subband distribution coefficient,Counterlet was used to decompose and reconstruct the preprocessed image and the K-Means++algorithm was used to convert the frequency information in the reconstructed image into color features.The color features were extracted by MATLAB,and a sample library of color features of unfavorable geological bodies was established accordingly.The original detection image was matched and compared with the sample library to realize automatic interpretation.Facts and figures showed that it was feasible to decompose and reconstruct multi-directional,multi-resolution and multi-scale GPR images by using Counterlet contour transformation,the curve edge approximation effect was good and the reconstructed images had no information loss.K-Means++algorithm can realize the conversion of energy to frequency to color in the gray-scale image of GPR,and the converted image has prominent and intuitive color.The matching and comparison of spectral energy can realize automatic interpretation more accurately and quickly and avoid individual subjectivity.