An improved hyperspectral classification algorithm based on pixel spatial association

Classification algorithm is an important technique that has been studied for many years in hyperspectral image processing along with the development of hyperspectral remote sensing. Traditional classification algorithms mostly focus on the differences of spectral dimension but neglecting spatial structures of geography objects. By improving the ECHO algorithm, we put forward a classification algorithm based on pixel´s spatial association. This algorithm could possibly be used as an important step in hyperspectral data classification and identification. It may also be a good approach to get rid of redundancy information in endmember extracting, which means that pure spectrum can be extracted from less pixels in the hyperspectral image. In this way, endmember extracting and spectral unmixing will take less time and become more efficient.