In this paper, we propose a method for semi-supervised classification based on a group solution to cluster analysis in combination with Laplacian regularization of similarity graph. The averaged co-association matrix obtained with the cluster ensemble is considered as a similarity matrix in the regularization context. We use a low-rank representation of the matrix that allows us to speed-up computations and save memory in the solution of the derived system of linear equations. Both theoretical studies and numerical experiments on artificial data and hyperspectral imagery confirm the efficiency of the method.
|Журнал||Journal of Theoretical and Applied Information Technology|
|Состояние||Опубликовано - 1 янв 2019|