Terahertz Imaging Reconstruction for Super-resolution Identification of Millimeter-scale Defects in XLPE Cable Insulation

The detection of micro-defects in cross-linked polyethylene (XLPE) composite insulation is crucial for effectively monitoring cable conditions. This study presents a deep learning approach to identify millimeter-scale defects within XLPE composite insulation using terahertz imaging reconstruction. Initially, a three-layer reflection model is established based on a simplified structure of XLPE insulation, and the terahertz transmission characteristics within this model are derived. Artificial defect models are then created for testing purposes. Subsequently, terahertz imaging tests are conducted, resulting in an image set containing typical defects. Utilizing this dataset, a deep learning-based image recognition algorithm called super-resolution residual network (SRResNet) is employed for high-resolution reconstruction of terahertz images, enabling the identification of defect sizes. The results demonstrate a 48% improvement in identifying millimeter-scale defects compared to traditional methods. This innovative approach addresses the challenge of detecting early small defects within the composite insulation of XLPE cables.