中文摘要：

在无线体域网动作识别中，稀疏分类识别阶段待测向量稀疏表示系数的计算复杂度是影响其实时性的一个关键因素。提出一种基于压缩稀疏融合的动作识别方法，首先，对各对象动作矩阵进行训练；然后，通过稀疏融合得到融合稀疏向量；最后，将其重构后与待测动作向量做残差处理，比较残差，得到识别结果。该方法在识别阶段勿需对待测向量求解稀疏表示系数，使识别阶段算法的复杂度降低一半，实时性得到提高。实验结果表明，在降低复杂度的同时，本方法能对8种不同的人体动作进行有效识别。使用基追踪（BP）算法时，识别率与传统方法持平；使用正交匹配追踪（OM P）算法时，识别率比传统方法效果好。

英文摘要：

The complexity of computing sparse representation coefficient during recognition phase will influence the real‐time performance of activity recognition in wireless body area networks .This paper presents an activity recognition method based on compressed sparse fusion .First ,each subject’s activity matrix is trained .Then ,the fusion sparse vector is generated by sparse fusion algorithm . Finally , the recognition result will be obtained by comparing the residuals between test sample and the reconstruction of the fusion sparse vector .Because of no needing to compute the sparse coefficient of test sample for the proposed method ,the complexity will be reduced by half during recognition phase .As a result ,the real‐time performance of activity recognition will be improved .Simulation results show that the proposed method could recognize 8 different kinds human activities effectively with lowering the complexity .The recognition rate is similar to traditional method using BP algorithm ,and higher than traditional method using OMP algorithm .