Abstract:

RPCCL (rival penalization controlled competitive learning) algorithm has been extensively used in clustering problems of the aerial remote sensing hyper-spectral data but its performance is sensitive to the selection of the initial cluster center. This paper, therefore, further investigates the RPCCL and proposes an optimizing seed point selection which chooses non-neighbor data points of the greatest local density as seed points. Compared to the existing RPCCL with random seed points, the clustering by this RPCCL with selecting seed points can converge more stably and effectively. Moreover, it is applicable to deal with the data in red tide and oil spill aerial remote sensing hyper-spectral data image. Additionally, the efficiency of the clustering is very low because of the redundancy of high dimensions in the hyper-spectral data. Therefore a dimensionality reduction method is described to solve this problem. The experiments show the promising results of this improved RPCCL approach and the dimensionality reduction implementation.

Key words: RPCCL, clustering, seed point, data dimensionality reduction, cluster center