ABSTRACT
Joint spatial-spectral features of the hyperspectral data are significant for identifying the nonlinearly distributed classes of the hyperspectral image. So the hyperspectral data cube is decomposed by three-dimensional Discrete Wavelet Transform (DWT) (3D DWT). The contextual information is obtained by extracting 3D gray level cooccurrence matrix (3D GLCM) features. The combined 3D DWT- and GLCM-based methodology classifies the pixels in the hyperspectral images efficiently. But in remote sensing, a single pixel represents multiple classes called mixed pixels. To alleviate such mixed pixel issues, fuzzy-based approaches are useful in revealing the percentage of the classes contributing to the mixed pixel, by the membership function. Hence, it is the motivation for developing a fuzzy-inspired hyperspectral image classification algorithm. By adding, a simple fuzzy framework to the 3D DWT and 3D GLCM combination, the fuzzy-inspired algorithm is developed and is presented in this chapter. Fuzzy-inspired 3D DWT and 3D GLCM features produce the accuracy of 95.94% and 97.63% for AVIRIS and ROSIS data, respectively, while 5% of the samples of the classes were used for training of support vector machine (SVM), while GLCM and 3D DWT combination yielded only 92.18% for and 96.48% accuracies for the same data.
