Fingerprint recognition is a widely adopted biometric technology, valued for its reliability and precision in identifying individuals. However, traditional recognition methods relying on handcrafted features struggle under challenging scenarios such as overpressured fingerprints, where excessive pressure distorts ridge patterns, significantly affecting performance. To address these challenges, this study proposes a novel framework combining domain adaptation techniques and an attention mechanism. The framework aligns feature distributions between source and target domains, enhancing the model's generalizability to diverse datasets and acquisition conditions. Additionally, the attention mechanism emphasizes critical regions of the fingerprint, improving robustness to distortions. Experimental results demonstrate that the proposed model significantly outperforms the original ResNet, achieving a reduced Equal Error Rate (EER) of 0.0837 compared to 0.1840 for the baseline. Grad-CAM visualizations further validate the model's ability to focus on essential fingerprint features, even under distorted conditions. This study highlights the effectiveness of integrating domain adaptation and attention mechanisms in overcoming real-world challenges in fingerprint recognition.