Abstract: High-speed targets can be indirectly detected and identified by analyzing the atmospheric disturbance field caused by their maneuvers. However, the traditional disturbance field interpretation algorithm for targets has limitations, such as challenging feature extraction, long processing time, and poor generalization ability. Therefore, a real-time algorithm, FlowReconNet (flow field reconstruction network), based on generative adversarial networks, is designed for interpreting the target disturbance field and identifying it simultaneously. Firstly, the proposed algorithm generates the interpretation map and classifies the target disturbance field using a generator. Then, the discriminator verifies the authenticity of the generated output and compares it with the reference feature vector from the auxiliary enhancer. Finally, the generator is optimized using a multi-value joint loss function. Ablation experiments show that the designed generative network can effectively improve the interpretation and reconstruction accuracy of the target disturbance field. The experimental results from multi-resolution simulated disturbance images of supersonic targets demonstrate that FlowReconNet decodes 512\times 512 pixel target disturbance field images at 67.33 frame/s under a GPU environment. The algorithm achieves comparable decoding performance to traditional methods, with a speed improvement of approximately 33 times, an average peak signal-to-noise ratio (PSNR) of 29.948 dB, and a classification accuracy of 96.701%.