Abstarct: In recent years, a lot of research has been done to increase the quality of images and video. High resolution video and images are used in many smart systems today. One of the most important applications of image enhancement is license plate recognition. Automatic license plate recognition plays an important role in traffic control and security in surveillance systems, and the presence of these surveillance cameras in cities can help identify the license plates of criminals at the crime scene and establish security. Insufficient resolution of the license plate image, along with its distortion, makes it very difficult to detect the license plate. One way to increase image resolution is image resolution. In order to correct the image of the damaged license plate, correction and reconstruction methods should be used to provide a suitable input for license plate reader software. In this dissertation, the aim is to make resolvable and correct damaged plates with the help of deep grids. For this purpose, we have suggested using the UNet network to increase the image resolution. Our data are images collected from license plates that are not of good quality for identification, and to solve this problem, a solution baxsed on deep UNet networks has been proposed. The reason for using this algorithm among other deep learning algorithms is that UNet algorithm due to accurate response, high accuracy, high speed of processing and learning, no need for large amounts of training data to learn and no need for complex and expensive hardware is. In this thesis, 1500 images of license plate images were collected to train the network and 1000 images of distorted license plate images were collected in order to measure the performance of the proposed method and PSNR, MSE and SSIM criteria were used to evaluate its ease. The percentage of increasing the resolution of the image of the license plate with low resolution was 89.161% and the percentage of reconstruction of the image of the distorted license plate was 95.3%.