Discipline: Technology and Engineering
Subcategory: Electrical Engineering
Tilaye T. Alemayehu - University of the District of Columbia
Co-Author(s): Nian Zhang, University of the District of Columbia, Washington, DC
This paper presents the design and implementation of a new adaptive feature selection technique for spectral band selection prior to classification of remotely sensed hyperspectral images. This approach aims to integrate spectral band selection and hyperspectral image classification in an adaptive fashion, with the ultimate goal of improving the analysis and interpretation of hyperspectral imaging. The four components in the proposed adaptive feature selection, including local gradient calculation, reference cluster determination, prototype classes building using a fuzzy classifier, and relevant bands selection are presented in detail. The hyperspectral image data set from the ROSIS (Reflective Optics System Imaging Spectrometer) were used as training and testing data. We tested the effect of the approach on different number of selected spectral bands. The classification accuracy for AFS was illustrated by the ROC curve. In addition, multiple experiments were performed using a simulated hyperspectral cube composed by 123 samples and 1254 features and classification was done only for verification purposes. Cross-validation demonstrated that FEA generated an average improvement of 7% on the misclassification error when compared to full feature analysis. Moreover, in order to compare the proposed method with other methods, we applied the proposed adaptive feature selection (AFS) approach and the principal component analysis (PCA) method to the Gentle Boost classifier using different number of spectral bands after processing the ROSIS Pavia scene. The experimental results demonstrated that the classification accuracies obtained by the AFS method are higher than that of the PCA method. In addition, for each method, the higher the number of spectral bands, the higher the classification accuracy.
Funder Acknowledgement(s): This research is supported by these grants: NSF/HRD1531014 and NSF/HRD1505509.
Faculty Advisor: Nian Zhang and Freddie M.Dixon, nzhang@udc.edu
Role: I did how to improve the analysis of hyperspectra imaging and on four components of the proposed adaptive feature selection.