Improved Approach of Seed Point Selection in RPCCL Algorithm for Aerial Remote Sensing Hyper-spectral Data Clustering with Data Dimensionality Reduction

The existing RPCCL (rival penalization controlled competitive learning) algorithm has provide an attractive way to perform data clustering. However its performance is sensitive to the selection of the initial cluster center. In this paper, we further investigate the RPCCL and present an improved approach of seed point selection which chooses non-neighbor data points of the greatest local density as seed points. We compare the performance of the RPCCL clustering with the selecting seed points and with the random seed points in red tide and oil spill aerial remote sensing hyper-spectral data (ARSHD) image. The experiments have produced the promising results. Additionally, because of the redundancy of high dimensions in the oil spill hyper-spectral data, a dimensionality reduction method is also described.