Abstract: On the premise that unmanned aerial vehicle(UAV) thermal infrared images are texturally complex, the insulator stain fault detection method was studied for transmission lines, to accurately and effectively detect insulator stain fault, reduce the risk of flashover or breakdown, and ensure smooth power supply to the grid. The edge of the initial insulator thermal infrared image was extracted using the Laplace operator and combined with the edge contour features extracted by the unified scene recovery network (USRNet) enhancement, whereby the complex background texture interference in the image was reduced, and an ultra-resolution reconstructed insulator thermal infrared image with high resolution and low texture complexity was obtained. This super-resolution image was input into a faster region-based convolutional neural network (R-CNN) detection model to detect insulator stain faults in transmission lines. The thermal infrared image edge extraction effect of the proposed method was satisfactory, and the super-resolution reconstructed image had high definition. The complex background texture interference in the image was effectively removed and overall image quality was significantly optimized, realizing accurate and effective detection of insulator stain faults, stable detection performance, and ideal average detection speed in practical applications.