许多核分类方法的决策函数可以表示为支持向量的组合,如SVM,而支持向量含有非常重要的隐私信息,因此,在分类决策时可能会暴露此类信息,同时分类速度受限于支持向量的个数,如SVM的分类复杂度为O(|SVs|).为解决上述两个问题,本文基于最小...许多核分类方法的决策函数可以表示为支持向量的组合,如SVM,而支持向量含有非常重要的隐私信息,因此,在分类决策时可能会暴露此类信息,同时分类速度受限于支持向量的个数,如SVM的分类复杂度为O(|SVs|).为解决上述两个问题,本文基于最小包含球球心在原始空间中的代理原像,提出了一种隐藏支持向量信息并能快速实现分类的SVM方法,称为隐私保护的快速SVM分类方法(Fast Classification Approach of SVM with Privacy Preservation,FCA-SVMWPP).同时提供了两种求解代理球心原像的方法,分别称为QP解法和直接解法.UCI和PIE人脸数据集的实验结果表明,本文方法可解决上述两个问题并具有较好的效果.展开更多
A method to detect airports in large space-borne synthetic aperture radar（SAR） imagery is studied.First,the large SAR imagery is segmented according to amplitude characteristics using maximum a posteriori（MAP） est...A method to detect airports in large space-borne synthetic aperture radar（SAR） imagery is studied.First,the large SAR imagery is segmented according to amplitude characteristics using maximum a posteriori（MAP） estimator based on the heavytailed Rayleigh model.The attention is then paid on the object of interest（OOI） extracted from the large images.The minimumarea enclosing rectangle（MER） of OOI is created via a rotating calipers algorithm.The projection histogram（PH） of MER for OOI is then computed and the scale and rotation invariant feature for OOI are extracted from the statistical characteristics of PH.A support vector machine（SVM） classifier is trained using those feature parameters and the airport is detected by the SVM classifier and Hough transform.The application in space-borne SAR images demonstrates the effectiveness of the proposed method.展开更多
文摘许多核分类方法的决策函数可以表示为支持向量的组合,如SVM,而支持向量含有非常重要的隐私信息,因此,在分类决策时可能会暴露此类信息,同时分类速度受限于支持向量的个数,如SVM的分类复杂度为O(|SVs|).为解决上述两个问题,本文基于最小包含球球心在原始空间中的代理原像,提出了一种隐藏支持向量信息并能快速实现分类的SVM方法,称为隐私保护的快速SVM分类方法(Fast Classification Approach of SVM with Privacy Preservation,FCA-SVMWPP).同时提供了两种求解代理球心原像的方法,分别称为QP解法和直接解法.UCI和PIE人脸数据集的实验结果表明,本文方法可解决上述两个问题并具有较好的效果.
文摘A method to detect airports in large space-borne synthetic aperture radar（SAR） imagery is studied.First,the large SAR imagery is segmented according to amplitude characteristics using maximum a posteriori（MAP） estimator based on the heavytailed Rayleigh model.The attention is then paid on the object of interest（OOI） extracted from the large images.The minimumarea enclosing rectangle（MER） of OOI is created via a rotating calipers algorithm.The projection histogram（PH） of MER for OOI is then computed and the scale and rotation invariant feature for OOI are extracted from the statistical characteristics of PH.A support vector machine（SVM） classifier is trained using those feature parameters and the airport is detected by the SVM classifier and Hough transform.The application in space-borne SAR images demonstrates the effectiveness of the proposed method.
