In this paper, we introduce a method to do image segmentation in documents with anti-counterfeiting watermarks. We proposed a method based on the Deep Learning algorithm to remove the noise of anti-counterfeiting watermarks. The main research purpose of this paper is how to use deep learning U-NET model to remove anti-counterfeiting watermarks. The U-NET model is trained by establishing the data set of the training images and the validation images, and then the anti-counterfeiting watermark will be removed by the trained U-NET model to improve the identification of data.