论文部分内容阅读
针对支持向量数据描述(SVDD)单类分类方法运算复杂度高的缺点,提出一种启发式约减支持向量数据描述(HR-SVDD)方法.以启发的方式从原有训练集中筛选出部分样本构成约减训练集,对约减训练集进行二次规划解算,得到支持向量和决策边界.通过不同宽度系数高斯核SVDD特征的讨论,证明了HR-SVDD的有效性.人工数据集和真实数据集上的实验结果表明,HR-SVDD分类精度与传统支持向量数据描述相当,但具有更快的运算速度和更小的内存占用.
Aiming at the shortcomings of high computational complexity of SVDD classification method, this paper proposes a heuristic reduced support vector data description (HR-SVDD) method. Some samples are selected from the original training set by means of heuristic Construct reduced training set and solve the quadratic programming solution of reduced training set to obtain support vector and decision boundary.The validity of HR-SVDD is proved through the discussion of SVDD features of Gaussian kernel with different width coefficients.Artificial data set and real Experimental results on datasets show that HR-SVDD classification accuracy is equivalent to that of traditional support vector data, but with faster computing speed and smaller memory footprint.