RCTNet-A Defect Detection Network for Tiny Debonding Defects in Terahertz Images

Due to the limitation of terahertz wavelength and serious development of terahertz devices, terahertz imaging for non-destructive testing has been severely constrained by insufficient data, low spatial resolution, blurred contours and high background noise. In this paper, a THz- TDS system is utilized to conduct experiments on defective samples of composite materials, and time-slice imaging is utilized to obtain defective terahertz image datasets. A THz-RCTNet is proposed to identify the defects with high accuracy, in which CSSPP module and modified loss function effectively increase the detection accuracy of the defects. The model can accurately identify tiny defects and achieve the high-precision localization of defects. In summary, our proposed method is conducive to further development of terahertz image based nondestructive testing technology and fullfil the requirements of realtime defect detection in industrial applications at an early stage.