Feature extraction is one of the most essential parts in image processing. For traditional algorithms of feature extraction cannot satisfy the highresolution (HR) images, this paper applies bagofword (BoW) model algorithm to optimize the analysis of HR images feature extraction. First, key points are found by using SIFT algorithm. Second, feature vectors are extracted from key points. In terms of feature extraction of BoW model, this paper propose the mean ratio and Weber Local Descriptor (WLD) as new feature vector to improve the performance of ratio detection and decrease the illumination effect. In the feature extraction experiment using the database of Shifang SAR image, the result indicates that the new BoW model has better robustness and efficiency.