In the process of creating Thangka paintings, the same line drawing can be used to produce different types of Thangka, but each type requires the artist to redraw an identical line drawing. This step is labor-intensive and time-consuming, relying heavily on manual work. Due to the difficulties in obtaining real line drawing images of Thangka and the distortion issues present in existing line drawing extraction methods, this paper proposes a novel Thangka line drawing extraction method based on Convolutional Neural Networks (CNN) combined with edge detection, named EdgCNN. This method incorporates innovative edge detection and line drawing fine-tuning modules, improving the accuracy and quality of line drawing extraction and effectively removing image noise, producing smooth lines and detailed Thangka line drawings. We conducted experiments on the Thangka1500 dataset, and comprehensive evaluation results indicate that our method outperforms existing methods in Thangka manuscript image extraction. Therefore, our method provides new technical support for the digital creation of Thangka art and the preservation of intangible cultural heritage.