Terahertz (THz) technology, as a novel nondestructive testing (NDT) approach, has emerged great potential in thickness measurement and damage location of various non-metallic materials due to its superior detection accuracy and safety. However, when THz wave penetrates the material, the inherent physical properties of the electromagnetic wave will cause it to occur different degrees of dispersion, which will limit the temporal-spatial resolution performance of THz NDT. Thus, in this work, a dispersion suppression and compensation method of THz wave is proposed based on U-Net with residual blocks. Firstly, the transmission characteristic of THz wave is analyzed. The key of dispersion compensation is to obtain the sparse pulse sequence related to the time-of-flight of structure. Then, the U-Net with residual blocks network is proposed to capture sparse pulse sequence from dispersive THz signals. Specially, during training process, Dynamic Focus Sparse Regression Loss (DFSR Loss) and Impulse Localization Overall Score (ILOS) are proposed to enhance the attention of the network to sparse pulses. Finally, a series numerical experiments of single-layer and multi-layer structures verify the effectiveness of the proposed method in suppressing and compensating complex dispersive THz signals.

Terahertz nondestructive testing, U-Net, dispersion compensation, sparse pulse sequence, resolution improvement