Remote sensing image classification with small training samples based on grey theory

Depending on small samples, good adaptation, high classification accuracy, are important to remote sensing images classification. Grey system theory studies on the “small sample”, “poor information”, uncertain systems, which are difficult for Statistics and Probability Theory, fuzzy mathematics. The paper proposed a method, named Maximum gray slope correlation classification. The method were designed and implemented based on the gray slope correlation degree model. Then, the comparative classification tests between the gray relational classification and other conventional remote sensing classification methods were implemented using small samples. The classification results showed that the accuracy of maximum gray slope correlation is very similar to spectral angle mapper, and close to the support vector machine and neural network. The classification accuracies were sorted as following: Support Vector Machines> Neural Networks> maximum gray slope correlation > spectral angle mapper > minimum distance> maximum likelihood> mahalanobis distance. Compared with other classification methods, Maximum gray slope correlation classification is simple, and has the best combined accuracy considering every subclass.