An Unsupervised Classification Method for Hyperspectral Image Using Spectra Clustering

Matched filtering method is successfully used in classification for hyperspectral image. However, it is hard to extract the low reflectance ground object due to atmospheric influence. In this paper, an improved unsupervised classification method was introduced. It can extract the low reflectance object such as vegetation in shadowed region and water from the hyperspectral image effectively. Firstly, using pixel purity index (PPI) to find the endmembers from hyperspectral image and computing the spectral angle between the pixel spectrum and each endmember spectrum, the pixel was classified into the endmember class with the smallest spectral angle. Then, the endmember spectra were clustered using K-mean algorithm. Finally, classes were merged according to the K-mean algorithm result and the final classification result was projected and outputted. Comparing the result with the original image and the field data, they are consistent with each other. This method can produce the objective result without artificial interference.