Terahertz Frequency Modulated Continuous Wave Detection Based on Image Restoration and Multi-View Scanning
Objective When detecting defects in multi-layer adhesive structures using a terahertz frequency modulated continuous wave(FMCW)detection system,the unidirectional nature of reflection-based scanning can lead to overlapping information blind spots,which affects tomographic imaging quality.Specifically,defects at the same horizontal location but at different depths may overlap or be occluded.Methods To assess the effectiveness of the proposed methods,scenarios for both unobstructed and obstructed defect detection were designed for analysis.For unobstructed defects,the target structure(sample 1)is a thermal protection system composed of ceramic matrix composites,insulation felt,and metal plates,bonded sequentially with epoxy resin.Adhesive defects of various shapes and sizes are introduced into adhesive layer 1 and adhesive layer 2.For obstructed defects,polyimide foam insulation serves as the primary structure,and it is layered with low-reflective(PVC)and high-reflective(coins,key)impurities,referred to as sample 2 and sample 3,respectively.For tomographic imaging,an image restoration algorithm is proposed.Initially,mask blocks are labeled on the near-field detection results,and based on this information,target removal is performed on the far-field detection results.The regions are then filled by eliminating differences in texture between the patch elements and the background,yielding detection results containing only the far-field layer's information.For 3D reconstruction,we discuss the effects observed in all three detection scenarios.Specifically,in scenarios similar to sample 3,a multi-view scanning method is introduced to minimize blind spots and achieve more complete 3D reconstructions.Results and Discussions The original detection results confirm that overlapping information and defect occlusion are common in cases involving thin or high-reflectivity defects(Figs.5 and 7).For low-reflectivity defects,high-quality tomography imaging is achieved by selecting characteristic features from different depth layers(Fig.6).By applying the proposed image restoration algorithm,we directly output F1 as I1 and use defect information from F1 to mask and restore textures in F2,producing an imaging result that contains only the current layer's defect information(Figs.10 and 11).However,for high-reflectivity defects,some useful information is still lost during tomography imaging and 3D reconstruction.To address this,a multi-view scanning approach is proposed.Results from two scans are registered to generate a complete 3D reconstruction(Fig.13).Conclusions The unidirectional nature of reflective terahertz detection significantly affects detection due to information overlap and blind spots.We propose using image restoration and multi-view scanning to mitigate these issues.These methods are validated on two-layer adhesive structures,providing new insights into the application of terahertz nondestructive testing technology in engineering.Future studies should explore multi-layer structures,especially those involving high-reflectance occlusion defects,from various perspectives,including front,top,and side views.
NETL
NSTL
万方数据
