The ablation fault in the buffer layer is one of the main factors of high-voltage cable fault. Due to the complex on-site environment and the need to achieve a certain angle of relative rotation for CT detection of objects, there is no equipment for CT detection of high-voltage cables in service at present. A portable detection system for on-site application is designed by using the local CT imaging mode of X-ray source translation scanning and verified by experimental simulation and experimental platform, which realizes online CT detection of defects in high-voltage cable buffer layer in service. For the buffer layer of 110 kV high-voltage cable, the hole defect of Ø1 mm can be detected, and the defect of Ø2 mm can be detected for 220 kV high-voltage cable. After collecting 2 016 CT images with defects, combined with 802 circumferential CT images of cables with the similar structure, the method of migration learning is used to achieve automatic defect detection, with an accuracy rate of 87. 6% , a recall rate of 93. 5% , and a missed detection rate of 6. 5% , which are close to the level of circumferential CT.