堤防隐患雷达图像的RB-NN融合

RB-NN fusion of images from ground-penetrating radar for detecting hidden defects in dyke

在运用探地雷达进行堤防隐患探测的过程中,为了识别雷达波散射图谱中的隐患特征,运用径向基神经网络对图像进行滤波融合。图像融合的对象为采用探地雷达技术获取的堤防模型内部的雷达波散射图谱,同时也包括采用时域有限差分方法对堤防隐患模型进行的正向模拟演算图谱。试验结果认为:径向基神经网络方法的非线性映射特征明显,聚类分析能力强,对多维特征的雷达图谱数据拟合具有良好的适用性。针对含有较多突变点、漂移点、缺失点的样本,径向基方法表现出较好的逼近性能,且计算过程简洁、时间成本低、数据收敛的可靠性强,能够满足图像融合及可视化分析的需求,为堤身隐患雷达图谱的特征识别提供技术支持。

During the application of ground-penetrating radar to the detection of hidden defects in dyke, the filtering fusion is made with RBF（ radial basis function）neural network method, so as to identify the features of the hidden defects in the radar scattering images, the radar scattering image inside of the relevant dyke model obtained by the ground-penetrating radar is taken as the object for the image fusion, while the calculation images for the forward simulation made on the dyke hidden defect model with CFDTD（ finite-difference time-domain）method are included as well. The test result shows that the nonlinear mapping feature of RBF（ radial basis function）neural network method is distinct with a strong capacity of cluster analysis and a better adaptability for the fitting of the radar imaging data with a muhi-dimensional feature. For those samples with more abrupt-change points, drift points and missing points, the RBF （ radial basis function） neural network method presents a better approximation performance with the merits such as simple calculation process, lower time cost, strong reliability of data convergence, etc. , and then can meet the demands of image fusion and visual analysis and provide a technical support for recognizing the feature of the radar image of the hidden defects in dyke.