A new Classifier for Remote Sensing Data Classification : Partial Least-Squares

This study has presented a new classifier-the Partial Least Squares (PLS) classifier including linear and nonlinear based on the Partial Least-Squares Regression theory, then explained the classification algorithm and process of this new classifier, and finally, them have been applied to classify Landsat TM remote sensing data. Results of Partial Least-Squares linear classifier showed that there exist many classify mistake among six kinds of land use types. On the contrary, the nonlinear classifier based on Gaussian kernel function gotbetter classi fication result, the overall classification accuracy is79.297% and overall Kappa statistics is 0.74213. So, to remote sensing classification, the nonlinear PLS classifier is basic feasible, however, it is necessary for us to improve its algorithms or learning process further.