Hyperspectral Image Classification Based On Cascaded Multi-classifiers

In the classification task of hyperspectral images,the noise in the large amount of spectral channels and the imbalance sample distribution of various ground objects usually cause many problems.For example,the classification accuracy and the training efficiency are usually not in balance,and the classification accuracy of small-size samples is relatively low.To address the problem mentioned above,this paper proposes a novel classification method for hyperspectral images based on cascade multiple classifiers.First,the highly correlated high-dimensional features are converted into independent low-dimensional features by principle component analysis(PCA),which will speed up Gabor filter for texture feature extraction in next step.Then,Gabor filters are used to extract image texture information in multiple scales and directions.Each Gabor filter generates one feature map.In the feature map,a d-by-d neighborhood centered on each unclassified sample is defined,the mean and variance within the neighborhood are considered as the space information of the center unclassified sample.The spectral information and the space information are then combined to reduce the noise influence.Finally,the spectral-space combination features are input to the cascade multiple classifiers to generate the average probability distribution of each sample w.r.t.all ground object classes.Experiments on three benchmark data sets,i.e.,Indian Pines?Pavia University and Salinas,are conducted to evaluate the performance of the proposed method and many classic methods,such as SVM and CNN.The experimental results are measured by three criterion,overall classification accuracy,average classification accuracy and Kappa coefficient.The overall classification accuracies achieved by the proposed method on the three data sets are 97.24%,99.57%,and 99.46%,respectively.For overall classification accuracy,the proposed method achieves 13.2%,4.8% and 5.68% higher than SVM with RBF kernel,respectively;the proposed mthod also achives 3.27%,3.2% and 0.3% higher than CNN,respectively.The Kappa coefficients achieved by the proposed method on the three data sets are0.9686,0.9943 and 0.9956,respectively,which also validated the superiority of the proposed method compared with other methods.Experimental results indicate that the proposed method can achieve superior classification performance on highspectral images compared with classical methods,such as SVM and CNN.The training efficiency of the proposed method is also relatively high compared with other classical methods without relying on GPU.Moreover,the proposed method can obtain high classification accuracy on small-size samples.There are 19 figures,11 tables and 52 references in this thesis.