A detection method for SAR targets based on extraction and dimensionality reduction of multi?scale SIFT features is proposed. Aiming at the problem that SAR target features cannot be completely described in single scale, we put Gaussian scale space and multi?group of seed points into use to achieve the extraction of multi?scale SIFT features. Meanwhile, there are description redundancies and structural redundancies in the same and different scales, so the method of sparse coding and features statistics is introduced to reduce redundancies and dimensionali?ty for feature vectors. Through quantitative analysis, the most optimal parameters of multi?scale factor and number are fixed, this makes the target features contain both the overall target contour information and the image details. Comparison with traditional target detectors, such as CFAR, SIFT features and multi?scale SIFT?PCA features etc, is performed in detail. The experimental results and their analysis show preliminarily the superiorities of the propos?al.%提出一种用于SAR图像目标检测的多尺度SIFT特征提取及降维方法. 针对在单一尺度下无法完整描述SAR目标的问题,采用高斯尺度空间和多组种子点的方式实现多尺度SIFT特征描述,并对同一尺度和不同尺度间的描述冗余和结构冗余分别采取稀疏编码和特征统计的降维方式实现去冗余处理. 在多尺度因子和尺度层数的选择上,通过定量计算选取最优描述参数,使得代表目标特征的向量既包括目标整体轮廓信息又包含图像细节描述. 与传统双参数恒虚警率、单尺度SIFT特征、多尺度SIFT?PCA等方法进行对比测试,验证了该方法的有效性.
展开▼
