<p>Hyperspectral images (HSI) consist of hundreds of spectral bands per pixel and it offers rich spectral information. However, redundancy, noise, and computational burden increase. Therefore, feature extraction and band selection techniques are crucial in HSI to choose the most informative and non-redundant spectral features from the high-dimensional HSI data to improve classification accuracy, reduce computational load, and avoid overfitting. In this study, we propose a novel framework for feature extraction and band selection. Four techniques such as Gray-level co-occurrence matrix (GLCM), wavelet transform, local binary pattern (LBP), and local ternary pattern (LTP) are the techniques used for extracting handcrafted features. Individual and hybrid texture features are derived using these four methods, where hybrid feature extraction is achieved by concatenating the individual feature sets. Band selection is carried out using the metaheuristic Hippopotamus Optimization Algorithm (HOA) to identify the optimal spectral features. The texture features from GLCM, Wavelet, LBP &amp; LTP along with the spectral features of HOA are concatenated and used as input for the machine learning classifiers. The classification is also performed by an enhanced convolutional neural network (CNN), after which the results are analyzed. HOA has not yet been widely explored in the context of hyperspectral imaging, which presents a promising opportunity for novel research and application in this domain. The results show that classification accuracy improves when using HOA. The highest accuracy of 89.62% is achieved by combining GLCM and Wavelet texture features with HOA-selected bands. The Enhanced CNN further increases the performance, reaching an accuracy of 91.37%.</p>