The methods of multispectral imagery processing

such as artificial neural network classification and independent component analysis are not applicable to high spectral imagery

because of “Hughes Phenomenon” (when training samples are fixed

the classification accuracy decreases with the increase of feature dimension) and classification accuracy under small study samples can not be effectively solved for high dimensional data. A decision-tree-based multiclass support vector machines is proposed and applied in spectral classification for the multi-classification problem of surface rocks collected from the Beiya gold mine

Heqing County

Yunnan Province. The results show that the average classification accuracy rate can be above 93.75%

suggesting that multiple classification based on decision tree classification support vector machine (SVM) can be applied to spectral classification of rocks.