Abstract: In order to increase the accuracy of sleep architecture stage and overcome the influence in classification brought by samples shortage，all night sleep EEG data from MIT-BIH database are accepted as the research sample and totally 16 parameters including time-domain，frequency-domain and nonlinear picked up for sleep classification using 1-against-1 multi-classification method of Support Vector Machines using Sequential Minimal Optimization algorithm and selecting radial basis function as the kernel function.In contrast to expert’s manually-scored classification label，the classification result is more than 92%.It shows that Support Vector Machines can be a kind of effective algorithm in sleep stage.

摘要： 为了提高睡眠结构分期的准确度，克服分类时样本不足对分类的影响，使用MIT-BIH数据库整晚睡眠脑电数据作为研究样本，提取了时域、频域和非线性共16个参数作为分类特征，用支持向量机的一对一多类分类方法，采用顺序最小优化算法，以径向基函数作为核函数对样本分类。分类结果与专家的分类标注对比，分类准确率达到92%以上。支持向量机可作为睡眠分期的一种实用算法。