Deep learning inspection techniques for construction quality of intermediate joints in urban underground cables

During the installation of urban underground cable intermediate joints, errors made by construction personnel frequently lead to defects, including unpolished burrs on the connecting tube, uneven stripping of the outer semi-conductive layer from the main insulation, and scratches or stains on the surface of the main insulation. To mitigate these defect hazards and enhance the stability of cable operations, this paper presents a defect detection method that integrates the U-Net semantic segmentation algorithm with the Canny edge detection algorithm. First, the U-Net algorithm is utilized to segment the intermediate joint image into three components: main insulation, outer semi-conductive layer, and crimping tube. Subsequently, the Canny algorithm is employed to extract the edge information from the images of the crimping tube, main insulation, and semi-conductive layer. The edge curve is then fitted and analyzed numerically to determine the presence of burrs or uneven stripping defects. Finally, to assess the stains and scratches on the main insulation surface, the number of edge points in the connected regions is utilized to ascertain the existence of defects. Experimental results demonstrate that the integration of the U-Net and Canny algorithms effectively identifies various defects, achieving a detection accuracy exceeding 90%. This method effectively mitigates the latent operational hazards posed by construction defects in urban underground cable intermediate joints and significantly enhances the stability of the power system.