The performance of fingerprint recognition systems mainly depends on the extraction and matching of minutiae features. However, existing research has shown that the reliability of automatic fingerprint recognition systems is often limited due to the instability of fingerprint image quality. Therefore, this study proposes an efficient algorithm based on deep neural networks, specifically designed for high-precision classification of latent fingerprints with low image quality. The experiment is based on a low-quality fingerprint database, and the results show that the proposed deep learning method exhibits superior performance in both recognition accuracy and robustness. The deep neural network based on ResNet architecture performs well in fingerprint classification tasks, with evaluation metrics of: accuracy 0.9585, precision 0.9383, recall 0.9378, and F1 score 0.9380. The research in this work provides useful references for the study of low-quality fingerprints.