The prerequisite for accurate prediction and effective control of flow phenomena lies in the understanding of flow dynamics, and experimental studies provide essential data to support this process. Particle image velocimetry (PIV), as one of the important methods of measuring flow fields, plays a critical role in experimental investigations such as flow passing through a circular cylinder. PIV is a non-contact laser-optical measurement technique, however, it often faces challenges in obtaining complete and accurate flow field data when the optical path is obstructed. Particularly in PIV experiments involving flow passing through a circular cylinder, the presence of the cylinder itself and the supporting structure can significantly obscure the optical path, making it highly challenging to acquire complete PIV data. To solve this problem, we propose a deep learning-based flow field data reconstruction method, in which a deep learning framework centered on convolutional neural networks (CNNs) is used. The method aims to solve the reconstruction problem of gappy regions in flow field data by establishing a mapping relationship between flow fields with gappy regions and complete flow fields. First, the influence of gappy regions with different characteristics on the reconstruction accuracy of numerically simulated flow fields is investigated. The reconstructed flow fields are carefully compared with ground truth data through multi-dimensional assessments of instantaneous flow fields and velocity time statistics. The results indicate that the maximum L₂ error between the reconstructed flow field and the ground truth is still about 0.02. Furthermore, it is observed that as the size of the gappy region increases along the flow direction, the difficulty in reconstructing flow field increases significantly. In contrast, changes in the size of the gappy region perpendicular to the flow direction have minimum influence on the accuracy of flow field reconstruction. Additionally, the robustness of the proposed deep neural network against noise is systematically evaluated. When clean numerical simulation data are used for training, test data are generated by artificially introducing varying levels of Gaussian noise to assess the network performance under noisy conditions. The results demonstrate that the error between the reconstructed data and the ground truth increases exponentially as the noise level rises. Finally, the proposed deep neural network model is applied to real PIV experimental data, with the training data remaining clean and numerically simulated. Both instantaneous flow fields and time-averaged statistics are analyzed and compared. The results show that the network model successfully reconstructs velocity information in the missing regions and effectively corrects data errors caused by measurement inaccuracies in the backflow zones. The reconstructed experimental results show closer statistical agreement with numerical simulation data, demonstrating that the model proposed in this work, when trained solely on numerical simulation data, is capable of reconstructing missing physical information in PIV experiments. This method provides a novel approach for addressing the challenge of data reconstructionin PIV experiments.