The accuracy of satellite attitude data is an important factor restricting the geometric positioning accuracy of high-resolution images. For the current problems that the accuracy of satellite attitude data processing is not high and data assessment is difficult, an algorithm of satellite attitude data ground processing is proposed based on bilateral filtering and overall weighted smoothing to achieve high-accuracy processing for attitude data. Furthermore, the numbers are digital orthophoto map and digital elevation model reference data of geometric calibration field is used for accuracy assessment. By using high-resolution optical satellite of Yaogan-24 launched in November 2014 as an example, the processing results illustrate that the proposed method is robust and feasible. The relative accuracy of attitude data can reach 0.8″, and the positioning accuracy of panchromatic camera geometric correction image can increase to 15 m via using the refinement attitude data.