Research on supervised LPP feature extraction for hyperspectral image


For the classification among different land-cover types in a hyperspectral image, particularly in the small-sample-size problem, a feature extraction method is an approach for reducing the dimensionality and increasing the classification accuracy. A supervised principal locality preserving projection (SPLPP ) feature extraction algorithms, which uses the label information of training sample in locality preserving projection (LPP), was proposed in this paper. Three main steps are involved in the proposed SLPP: firstly uses PCA to remove redundant information, and then combines the label information in LPP, finally, SPLPP projects high-dimensional hyperspectral image into a low-dimensional space. Last but not least, SPLPP uses the extracted features as inputs of classifiers (e.g. support vector machine (SVM ) and K-nearest neighbors (KNN )) to do classification. Experimental results show that the proposed SPLPP has better local information retention ability and class discrimination ability compared with PCA, LPP, LDA

Remote Sensing Technology and Application