中文摘要：

使用智能手机来搜集声音和体动数据，对其进行预处理，提出联合特征提取和特征选择的TSFS方法。单纯的使用一种方法来选择特征，都会存在着一定的弊端。该方法是将特征提取和特征选择两种方法的联合，不仅可以筛选出符合实际情况的特征，而且还提高了分类的准确度。针对人类睡眠识别过程中的分类方法问题，提出基于改进二叉树的Multi—SVM睡眠分类器融合方法。单纯的使用一种分类方法，分类准确度难以得到提升。该方法是将多个SVM分类器组合成单枝的二叉树的形状，且树的每个节点都用一个二分类的SVM来分类。不仅降低了分类误差的积累，同时也提高了分类准确度。

英文摘要：

Using smart phones collect sound and body moving data, and these data are preprocessed, and the combination of feature extraction and feature selection is proposed, which is called TSFS method. Only using a method to select features, there will be some drawbacks. The method is a combination of two methods of feature extraction and feature selection, and not only can be screened out the characteristics of the actual situation, but also improve the accuracy of classification. For the classification of human sleep recognition process, a classifier fusion method of Multi-SVM sleep based on improved binary tree is proposed. Only using one classification method, the classification accuracy is difficult to be improved. The method is combining multiple SVM classifiers into a single branch of the shape of binary tree, and each node of the tree is classified by a two SVM. Not only the accumulation of classification error is reduced, but also the classification accuracy is improved.