Although spectral–spatial classification in hyperspectral imagery (HSI) delivers high accuracy, it remains sensitive to noise and redun
dant information in raw spectral bands. To address this issue, we propose a novel unsupervised band selection method for hyperspectral
image classification. The approach begins with spectral band analysis, followed using a covariance matrix to identify and retain the most
informative bands. A deep convolutional neural network (CNN) is then applied to the selected bands to extract global features, which are
subsequently classified using both SoftMax and support vector machine (SVM) classifiers. We validated the effectiveness of our method
on the Indian Pines and Salinas-A datasets, achieving accuracies of 97.66 % and 97.91 %, representing improvements of 4 % and 2.21 %,
respectively. These results demonstrate that our method significantly outperforms existing approaches in terms of classification accuracy.