Researches On Hyperspectral Image Classification Via Domain Transform Recursive Filter

Hyperspectral images contain hundreds of bands ranging from visible light to far-infrared,and the spectral resolution reaches nanometer level.Therefore,hyperspectral images have unique properties that other remote sensing images don't have.It can provide rich spectral information related to the physical properties of objects,which enhances human's ability to understand the objective world.However,traditional remote sensing methods face many challenges in processing hyperspectral images.First,the spectral dimension of a hyperspectral image is much higher than those of the images used in computer vision and other remote sensing applications.Second,hyperspectral images captured by satellite and airborne platforms are easily affected by many environmental factors such as lighting,shadow,and weather.In such imperfect situations,the spectral characteristic of the pixels in a hyperspectral image is usually quite complex.It means that the pixels belonging to the same object may have quite different spectra and vice verse.Facing these challenges,two edge-preserving filtering based hyperspectral image classfication methods are proposed in this work.1)Due to the complexity of the scene and unreliability of human labeling,the training set used for supervised classification may contain mislabeled samples.In order to address this problem,a novel method is introduced to detect and correct the mislabeled training samples for hyperspectral image classification.First,domain transform recursive filtering based feature extraction is used to improve the separability of the training samples.Then,constrained energy minimization based object detection is performed on the training set with each training sample serving as the object spectra.Finally,the label of each training sample is verified or corrected based on the averaged detection probabilities of different classes.Experiments demonstrate the effectiveness of the proposed method in improving classification performance with respect to the classifier trained with the original training set which contains a number of mislabeled samples.2)Edge-preserving features(EPFs)have been found to be very effective in hyperspectral image classification.However,single scale EPFs cannot well model the multi-scale information of hyperspectral images.Furthermore,the edge-preserving smoothing operation unavoidably decreases the spectral differences among objects ofdifferent classes,which may affect the following classification.To overcome these problems,in this paper,a novel principal component analysis(PCA)-based EPFs(PCA-EPFs)method for hyperspectral image classification is proposed,which consists of the following steps.First,the standard EPFs are constructed by applying edge-preserving filters with different parameter settings to the considered image,and the resulting EPFs are stacked together.Next,the spectral dimension of the stacked EPFs is reduced with the PCA,which not only can represent the EPFs in the mean square sense but also highlight the separability of pixels in the EPFs.Finally,the resulting PCA-EPFs are classified by a support vector machine(SVM)classifier.Experiments performed on several real hyperspectral data sets show the effectiveness of the proposed PCA-EPFs,which sharply improves the accuracy of the SVM classifier with respect to the original edge-preserving filtering-based feature extraction method,and other widely used spectral-spatial classifiers.