Abstract: In the field of drag reduction in wall turbulence, the spatial and temporal correlation between turbulence structure and friction drag has gradually become a research frontier, and obtaining high-resolution near-wall flow field is the prerequisite for this research. The traditional Particle Image Velocimetry (PIV) method has limited spatial resolution and causes large errors in the near-wall area with large velocity gradient, which makes it difficult to describe the near-wall turbulence structure and accurately calculate the wall shear stress. In order to solve this problem, this paper proposes a structurally improved model based on the lightweight optical flow neural network (LFN). By introducing high-resolution supervision constraints in advance to optimize the training mechanism of each layer of the pyramid, and introducing Bicubic interpolation in the sub-pixel refinement stage to enhance the estimation accuracy of the unknown region, the proposed method can effectively improve the estimation accuracy of the unknown region, and thus the high-resolution reconstruction of the near-wall flow field in wall turbulence can be realized. The model is trained on the dataset based on artificially synthesized cross-frame particle images, and the prediction error of the near-wall region is controlled within 3.5%. By using the real PIV experimental data, the near-wall flow field with single-pixel resolution and time series characteristics is obtained, which conforms to the statistical law of wall turbulence. By analyzing the shear stress signal of the near-wall flow field, it is found that its propagation characteristics are consistent with the migration law of the near-wall dominant turbulence structure. In this paper, a high-resolution near-wall flow field is obtained by improving the optical flow model, which can create conditions for high-precision wall friction measurement and spatial-temporal correlation analysis with turbulent structures.