基于自适应聚类流形学习的增量样本降维与识别

为了解决局部线性嵌入(locally linear embedding, LLE)流形学习算法无法自适应确定重构区间和不能进行增量学习等问题,提出了一种自适应聚类增量LLE(clustering adaptively incremental LLE,C-LLE)目标识别算法。该算法通过建立高维非线性样本集的局部线性结构聚类模型,对聚类后的类内样本采用线性重构,解决了LLE算法样本重构邻域无法自适应确定的问题;通过构建降维矩阵,解决了LLE算法无法单独对增量进行降维和无法利用增量对目标进行识别的问题。实验表明,本文算法能够准确提取高维样本集的低维流形结构,具有较小的增量降维误差和良好的目标识别性能。

Incremental sample dimensionality reduction and recognition based on clustering adaptively manifold learning

To solve the problem that incremental learning of locally linear embedding (LLE) cannot get reconfiguration neighborhood adaptively and powerlessly, a target recognition method of clustering adaptively incremental LLE(C-LLE) is proposed. Firstly, the clustering model of the clustering locally linear structure of high-dimensional data is build, so it is able to solve the problem of neighborhood adaptive reconfiguration. Then the proposed algorithm extracts an explicit dimensionality reduction matrix, and the problem of powerlessly incremental object recognition is solved. Experimental results show that the proposed algorithm is able to extract the low-dimensional manifold structure of high-dimensional data accurately. It also has low incremental dimension reduction error and great target recognition performance.