Fingerprints are one of the most commonly used biometrics in public safety. The quality of the fingerprint image directly affects the performance of the fingerprint recognition system. This paper proposes a low-quality fingerprint enhancement model based on U-Net, which can effectively remove the background noise of the fingerprint image and repair the damaged ridge structure of the fingerprint. Furthermore, to train the proposed model, we create a dataset containing low-quality fingerprints and their associated ground truth. We compare the proposed method with several state-of-the-art methods on the home-made dataset and the public dataset FVC2002, and the experimental results demonstrate that the proposed method performs better.