To address the challenges posed by the complex structures and indistinct edge features in 220 kV cable insulation layer image segmentation, this study employs a precise segmentation method based on the TransUNet model. By combining the local feature extraction capability of convolutional neural network (CNN) with the global contextual modeling strength of Transformers, the proposed model significantly enhances segmentation accuracy. Experimental results show that the method achieves an mDice of 0.9641, mIoU of 0.9538, mean precision (MP) of 0.9689, and mean recall (mRecall) of 0.9601 on the test set, representing respective improvements of approximately 5.56%, 5.74%, 6.12%, and 4.87% over the baseline UNet model. Overall, the proposed method outperforms UNet, Swin-UNet, and Attention-UNet, demonstrating its effectiveness in segmenting 220 kV cable insulation layers with complex structural features.