Abstract: For traditional scattering reconstruction algorithms such as digital image correlation based on local space and optical flow method based on the time domain have problems such as low spatial resolution and difficulty in real-time reconstruction. A real-time reconstruction method of the target flow field based on a conditional generative adversarial network called fast scattering network is proposed. This method optimizes the U-Net generator using a self-designed depth residual block and a grouping feature extraction unit and introduces speckle image pairs as auxiliary information to constrain the reconstruction results. In the dynamic interplay between the generator and discriminator, distinct loss functions are utilized to direct their optimization. The Structural Similarity Index (SSIM), average absolute loss, and perceived loss are further integrated into the loss function with the generator as the core to achieve better image reconstruction effect and noise suppression. The simulation results show that the SSIM of FastScatNet is higher than 0.956 at different reconstruction resolutions, and the inference speed is 30.2 frame/s on CPU and 49.6 frame/s on GPU, which is 9.44 times faster than the traditional DIC+OFM method. In the experiment of comparing the original noisy image and the denoised image, the SSIM difference of FastScatNet is 0.7% ~ 3.0%, which proves that FastScatNet has effective noise suppression ability.