Supervised K-means Clustering Analysis for Hyperspectral Imagery

We explore widely used K-means algorithm and propose two methods to improve its performance for hyperspectral clustering and analysis.A novel initialization method based on orthogonal subspace projection(OSP) is presented,which can get the suitable initial seeds for K-means clustering.In addition,we address a new cardinality estimation index which maximizes the distance ratio between intra-cluster distance and inter-cluster distance.It is used as a tool to estimate the numbers of clusters in K-means for hyperspectral data.The experimental results show that the proposed method can performs better than other traditional methods.