基于FLAKNN的雷达一维距离像目标识别

Due to the deficiency of traditional KNN algorithm in target recognition of high range profile, such as using fixed k value globally and not considering the influence of each characteristic component on classification, the target recognition performance is poor. Therefore, an improved KNN algorithm-FLAKNN, was proposed. By extracting the stable characteristics such as the size, entropy, center distance, irregularity, scaling feature and symmetry of the high range profile of the target, Fisher discriminant analysis was used to project all feature components to the low-dimensional space, so as to achieve the maximum separability among different categories. Combined with the local distribution of adjacent samples and the adjustment of k value, the principle of majority voting was finally used to determine the category of test samples. The results show that compared with the traditional KNN algorithm, this algorithm further improves the recognition performance.