Sea ice change is closely related to the change of global atmosphere and ocean circulation, which plays an important role in the study of global climate change. Sea ice concentration is one of the important parameters to study the temporal and spatial change of sea ice. Accurately retrieving sea ice concentration is the innovation of this paper. At present, the high-resolution microwave-detected sea ice concentration product was provided by the University of Bremen, which was derived by the Arctic Radiation and Turbulence Interaction Study (ARTSIST) Sea Ice (ASI) algorithm based on the Advanced Microwave Scanning Radiometer for Earth Observing System (AMSR-E) 89-GHz brightness temperature data. The AMSR-E/AMSR-2 89-GHz brightness temperature data has higher spatial resolution, but it is often affected by cloud and water vapor, which affects the recognition and subsequent use of ground feature. Although the weather filters can remove some errors in the edge regions of the sea water and the sea ice, the errors of the sea ice concentration in other regions cannot be removed. The generative model of Conditional Generative Adversarial Network (CGAN) increases the utilization of image feature information through skip connection, which improves the removal of the influence of cloud and water vapor. The discriminative model can retain the image feature information and realize the non-linear mapping from the image to the image. The loss function can reduce the pixel-level loss, which can remove the influence of cloud and water vapor. Therefore, this paper proposed an improved ASI algorithm based on CGAN. Firstly, the relatively stable relationship between the 89-GHz brightness temperature data which is not disturbed or less affected by the external environment and the 36-GHz brightness temperature data was determined, and the 89-GHz brightness temperature data with large interference was screened. Secondly, based on the 36-GHz brightness temperature data with high reliability, the 89-GHz brightness temperature data with large interference was corrected through CGAN. Finally, the ASI algorithm was used to retrieve sea ice concentration. Compared with sea ice concentration retrieved by the ASI algorithm, the results showed that the improved ASI algorithm based on CGAN was feasible. Compared with sea ice distribution obtained from the Landsat 8 OLI-L1T data, the improved ASI algorithm based on CGAN significantly improves the inversion accuracy of sea ice concentration. The improved ASI algorithm based on CGAN makes use of the reliable 36-GHz brightness temperature data, which greatly reduces the error caused by cloud and water vapor, and the method effectively corrects sea ice concentration of the pixels affected by the external environment. Therefore, the improved ASI algorithm based on CGAN realizes high spatial resolution and significantly improves the inversion accuracy of sea ice concentration.