In response to the reliance on manual inspection in infrared imagery of transmission line insulators and the scarcity of sufficient anomalous data samples,an unsupervised anomaly detection framework (PySobel FR) integrating the PySobel image enhancement algorithm and feature reconstruction mechanism is proposed.The PySobel algorithm combines Gaussian filtering,downsampling,upsampling,and Sobel operations to enhance edge perception while suppressing noise.The framework adopts a dual path reconstruction structure,where the teacher model serves solely as a knowledge transfer guide without directly participating in anomaly judgment.In the encoder,a deformable downsampling module is introduced to better capture complex edge features of insulators.During the decoding process of the student model,an efficient up convolution module replaces traditional convolutional attention and upsampling layers,leveraging residual connections to improve feature focus and detail recovery.Experiments on a self constructed infrared insulator dataset demonstrate the effectiveness of the proposed improvements,achieving an anomaly detection accuracy of 90.43%.