摘要
针对传统支持向量机(SVM)对图像中含有的噪声或野值样本的不敏感性问题,提出了一种基于双向隶属度的模糊支持向量机(FSVM)的方法。该方法通过计算样本类中样本与其所属类别和另一类别的中心点之间的距离,得出样本对每一类的隶属度,通过对样本隶属度的分析实现对样本点的分类。将该方法应用于对肺结节良恶性分类试验中,其参数反演结果表明,即使在噪声存在的情况下该方法也能得到较好的分类结果,而且克服了传统方法过拟合的缺点,从而也验证了该方法较强的抗噪能力和较好的分类能力。
In the support vector machine （SVM） method, the membership function has a vital infection on the classification of samples. Due to the limitation of the function own condition, this method cannot effectively distinguish the noise and outliers samples. A fuzzy support vector machine （FSVM） was developed based on the dual membership method to solve the problem. The method uses the characteristics of a specifically medical image to map the membership function which has been obtained from the method of degree membership to two different sides, mapping the membership function to obtain the membership function which can more effectively analyze the specific sample. The improved fuzzy support vector method was used to classify benign and malignant of the pulmonary nodule. The parameters inversion shows that the developed method distinguishes the noise and outlier samples more effectively, compared with the traditional fuzzy support vector machine method, and solves the over-fitting problem of traditional methods. Therefore, the results illustrate the robustness to anti-noise property and the effective classification ability of the developed method.
出处
《清华大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2014年第3期354-359,共6页
Journal of Tsinghua University(Science and Technology)
基金
国家自然科学基金资助项目(61240035
61373100)
山西省攻关项目(20120313032-3)
关键词
孤立性肺结节
肺结节良恶性
肺结节分类
solitary pulmonary nodule
benign and malignant of pulmonary nodules
classification of pulmonary nodules
