<p>Visual tracking holds great significance in the field of computer vision. Despite the advancement of numerous algorithms, achieving efficient visual tracking remains a persistent challenge. In this paper, an attention-based deep particle filter with data augmentation (ADPF-DA) is proposed as an innovative technique to address complex tracking scenarios. It combines the modified particle swarm optimization (PSO) enhanced particle filter (PF) and an attention-based deep network employing adversarial data augmentation through generative adversarial network (GAN). The deep network is trained offline and utilizes transfer learning and fine-tuning techniques to enhance adaptability to various tracking scenarios. Meanwhile, the main tracking framework applies PF, with the modified PSO algorithm optimizing particle diversity and convergence. Through the fusion of deep features and traditional ones, a robust appearance model is created, assisting in distinguishing targets from the surrounding environment. Within the framework, a tracking failure detector is designed to prevent tracking loss. Furthermore, GAN is utilized to augment training data, thereby improving network training and tracking effectiveness. Through comparative analysis on the dataset, the proposed method ADPF-DA excels in tracking performance and shows superior robustness and stability in tracking challenging video sequences compared to other trackers.</p>