Abstract: Cables and their accessories are susceptible to internal defects during manufacturing and operation, posing significant risks to the safety of a power supply system. We introduced a novel target recognition detection method utilizing terahertz time-domain spectral imaging to overcome the limitations of conventional approaches in detecting internal defects within cable composite insulation structures. Focusing on cross-linked polyethylene cable joints, we created an experimental model simulating joints with layering and metal impurities through equivalent simplification. The Terahertz frequency domain imaging and absorption spectrum imaging were conducted on the artificial model with defects, yielding corresponding imaging outcomes. Subsequently, an improved YOLOv8 model was employed to classify and recognize various defect images based on the acquired imaging data. The enhanced YOLOv8 model achieved an impressive 99.8% accuracy in detecting internal defects in cable joints, with an average accuracy of 99.5% at a joint crossover of 0.5, demonstrating substantial advancements over traditional methods. This methodology extends the application of Terahertz detection technology and object detection algorithms to non-destructive visual inspection of internal defects within cable composite insulation structures. It is revealed that the defect types and locations within cable composite insulation can be effectively identified and the method can be extended to the detection of internal defects in various layered composite insulation structures.