探地雷达(GPR)回波信号经常被各种随机噪声所干扰,引起后续目标解译的困难。近年来,模态分解方法被广泛地应用于GPR 数据噪声压制,然而,现有应用基于一维信号的模态分解需要对B-scan 的每一道分别进行处理,忽略了道之间的相关性。除此之外,在复杂场景下的去噪效果也有待提升。为此,文中提出了一种基于二维变分模态分解的GPR 噪声自动抑制方法,可以有效识别多个二维模态及其对应的中心频率。通过应用这种自适应方法,将图像分解为不同的信号模态和噪声模态,最终以带限方式准确再现输入的B 扫描图像。此外,为了进一步完善信号提取过程,采用频谱分析技术自动选择最相关的信号模态。仿真和实测数据验证表明,所提方法可以较好地实现对低信噪比GPR 数据的噪声去除,提高后续目标检测与识别等解译工作的精度。

The Ground Penetrating Radar (GPR) echo signal is often disturbed by various random noise, which causes difficulties in subsequent target interpretation. In recent years, mode decomposition methods have been widely used in GPR data noise suppression. However, the existing applications of mode decomposition based on one-dimensional signals need to process each B-scan trace separately, ignoring the correlation between traces. In addition, the denoising effect in complex scenes also needs to be improved. Therefore, an automatic GPR noise suppression method based on two-dimensional variational mode decomposition is proposed. This method can effectively identify multiple 2D modes and their corresponding center frequencies. By applying this adaptive method, the image is decomposed into different signal modes and noise modes, and finally the input B-scan image is accurately reproduced in a band-limited manner. Moreover, in order to further refine the signal extraction process, the most relevant signal modes are selected using automated spectrum analysis techniques. Simulation and measured data verification shows that the proposed method can better realize the noise removal of GPR data with low signal-to-noise ratio, and improve the accuracy of subsequent interpretation work such as target detection and recognition.

TN975

国防基础科研计划资助项目(JCKY2023906C001)