稀疏条件下的两层分类算法

Sparsity-inspired two-level classification algorithm

在有限样本下距离量的选择对最近邻算法(K-nearest neighbor,KNN)算法有重要影响。针对以前距离量学习泛化性不强以及时间效率不高的问题,提出了一种稀疏条件下的两层分类算法(sparsity-inspired two-level classification algorithm,STLCA)。该算法分为高低2层,在低层使用欧氏距离确定一个未标记的样本局部子空间;在高层,用稀疏贝叶斯在子空间进行信息提取。由于其稀疏性,在噪声情况下有很好的稳定性,可泛化性强,且时间效率高。通过在噪声数据以及在视频烟雾检测中的应用表明,STLCA算法能取得更好的效果。

The selection of distance greatly affects KNN algorithm as it relates to finite samples due to weak generalization and low time efficiency in the previous learning of distance. In this paper,a new sparsity-inspired two-level classification algorithm（ STLCA） is proposed. This proposed algorithm is divided into two levels： high and low. It uses Euclidean distance at the low-level to determine an unlabeled sample local subspace and at the high level it uses sparse Bayesian to extract information from subspace. Due to the sparsity in noise conditions,STLCA can have good stability,strong generalization and high time efficiency. The results showed that the STLCA algorithm can achieve better results through the application in noise data and video smoke detection.