【摘要】 探地雷达适合于铁路路基病害的检测,但后期资料处理工作费时费力,不利于其在铁路路基检测中的推广使用。本文分析各种路基病害图像的特征,从图像中提取分段能量、方差和层面位置作为特征值。根据这些特征值的大小不但能区分各种病害类型,而且可以比较病害的发育程度。根据已知样本数据计算这些特征值,建立学习向量量化神经网络模型,通过不断调整神经元的权值和阈值对特征值进行学习,直到满足给定精度为止。应用调整好的神经网络模型对沪宁线检测数据进行测试,结果表明,该模型对路基翻浆冒泥病害的识别率达90%以上。更多还原

【Abstract】 The ground penetrating radar(GPR) is suitable for detecting hidden dangers of railway subgrade,but data processing in the later stage needs much time and effort,which is unfavourable for generalized GPR application.This paper analyzes the characteristics of GPR plots of different kinds of defects in railway subgrade,extracts segmented energy,variance and interface as the eigenvalues from GPR data.The kinds of defects can be distinguished and the extent of development of defects can be compared by the magnitude of the eigenvalues.The eigenvalues are calculated from the known sample data,the learning vector quantization network model is established.The network studies the eigenvalues by constantly adjusting the weight and threshold of neurons until the given accuracy is reached.The adjusted network is used to test the GPR data of the Shanghai-Nanjing Railway.The results indicate that the model achieves a recognition rate of subgrade mud pumping defects above 90%.更多还原