摘要： 利用基于超完备字典的信号稀疏分解理论，提出一种基于稀疏分解的数据分类算法SRC。该算法通过学习不同类别数据的稀疏映射关系，把测试样本映射到高维空间中，根据稀疏重构的误差定义决策函数以确定测试样本的类别。采用UCI数据集评估该算法，并与SVM算法和Fld算法的实验结果进行对比，结果表明，SRC的分类准确率最高，不平衡数据集的实验结果显示了SRC的鲁棒性。

关键词: 超完备字典, 稀疏分解, 稀疏映射, 重构误差

Abstract: With the theory of sparse decomposition of signals over an overcomplete dictionary, this paper proposes a data classification algorithm based on sparse decomposition named SRC. By studying data sparse mapping relationships among different data classes, the test samples are mapped into a higher dimensional space. Decision function is defined according to the error of sparse reconstruction, which determines the class of test samples. It uses UCI dataset to evaluate the effectiveness of the algorithm, and compares the experimental results of Support Vector Machine(SVM) and Fld. The results show that SRC gains the highest accuracy in classification, and it has good robustness in the imbalanced dataset experiment.

Key words: overcomplete dictionary, sparse decomposition, sparse mapping, reconstruction error

中图分类号: 

• TP301.6