Spectral–Spatial Classification With Naive Bayes and Adaptive FFT for Improved Classification Accuracy of Hyperspectral Images

This article presents a postprocessing-based spectral–spatial classification (SSC) approach for hyperspectral (HS) images. The approach effectively overcomes the limitations of traditional pixel-based classifiers by integrating spectral and spatial information to achieve improved classification results. Specifically, the proposed method uses principal component analysis to transform the HS images and the Naive Bayes (NB) classifier to quickly derive spectral-posterior probabilities. Spatial-posterior probabilities are then computed using an adaptive fast Fourier transform (AFFT) and a probabilistic closeness function. These probabilities are then combined to generate a precise SSC map. The proposed approach is available in two distinct styles: the conventional NB–AFFT–SSC method and the proposed iterationwise variable sequencing based NB–AFFT–SSC (IVS–NB–AFFT–SSC) method, which classifies one designated class in each iteration. In addition, two wrapper-based feature selection methods are proposed to obtain a set of principal components (PCs) for each class of the HS image, significantly improving classification accuracy. The approach's efficacy is demonstrated through extensive experimentation on three real HS datasets, including Washington DC Mall, Salinas-A, and Botswana. The generality of the approach has been proven through the use of other well-known machine-learning algorithms, such as support vector machine and K-nearest neighbor, as wrappers in the approach. The results confirm that the proposed approach is highly effective, with the IVS approach helping users concentrate on a particular set of PCs for the class of interest.