Abstract: In the study of the matching correction of buoy and satellite data, the matching error is mainly composed of two parts : measurement error and representative error. In order to reduce the interference of wave representative error and accurately correct the measurement error, this study proposes a joint correction method combining ' credible radius ' and ' confidence interval screening '. Based on the buoy observation of the National Data Buoy Center (NDBC), the significant wave height (SWH) data of China-France Oceanography Satellite (CFOSAT) are corrected. Through the systematic analysis of the matching error, the concept and judgment method of ' credible radius ' are proposed, and the credible radius is determined to be 9 km. The deep neural network correction model is constructed by using the trusted radius matching data. The experimental results show that the model trained by the joint correction method of ' credible radius ' and ' confidence interval screening ' has the best performance. The average absolute error of the verification set is reduced by 16.2%, and the MAE of the test set is reduced by 14.8%. In addition, there is a stable error reduction in all matching radius distance intervals and wave height ranges. This method effectively enhances the accuracy and reliability of CFOSAT inversion data, and provides a new technical path for the fine correction of marine remote sensing data.