基于探地雷达数据的地层模型快速反演技术

Fast inversion technology of geological models based on ground penetrating radar data

为提高探地雷达目标检测的准确率,文中基于探地雷达数据与主特征提取技术提出了一种地层模型快速反演方法。通过将马尔科夫链蒙特卡洛(MCMC)方法改进,降低了反演探地雷达数据的计算量,提高了反演准确率。将残差块和置换注意力模型融入至MCMC算法中,由此提升了模型的去噪能力和分类判别器生成精细图像细节的能力。经过对照试验,所提出的快速反演模型在特征提取去噪方面的平均指标分别提高了0.032和0.011,而在反演能力方面的指标则分别提高了0.019、0.014和0.003 5,证明了该模型能够快速、准确地反演出地下目标的轮廓和位置。

To improve the accuracy of ground penetrating radar target detection,this paper proposes a fast inversion method for geological models based on ground penetrating radar data and principal feature extraction technology.By improving the Markov Chain Monte Carlo(MCMC)method,the computational complexity of inversion of ground penetrating radar data was reduced,and the accuracy of inversion was improved.At the same time,incorporating residual blocks and permutation attention models into the MCMC algorithm improves the denoising ability of the model and the ability of the classification discriminator to generate fine image details.After a comparative experiment,the proposed fast inversion model showed an average improvement of 0.032 and 0.011 in feature extraction and denoising,while its inversion ability improved by 0.019,0.014 and 0.0035,respectively.This proves that the model can quickly and accurately invert the contour and position of underground targets.