The paper introduces a novel model called HybridSN (Hybrid Spectral Convolutional Neural Network) for hyperspectral image (HSI) classification. HSI classification is crucial for analyzing remotely sensed images, and it involves processing images with varying bands of spectral information. While Convolutional Neural Networks (CNNs) are widely used for visual data processing, most existing methods rely on 2D CNNs, which may not fully capture the spatial and spectral information required for HSI classification. To address this, HybridSN combines a spectral-spatial 3D-CNN with a spatial 2D-CNN. The 3D-CNN captures the joint spatial-spectral feature representation from multiple spectral bands, while the 2D-CNN learns more abstract spatial representations. This hybrid approach reduces computational complexity compared to using a pure 3D-CNN and enhances classification performance. The authors conducted rigorous experiments on three benchmark datasets: Indian Pines, Pavia University, and Salinas Scene. The results show that HybridSN outperforms state-of-the-art methods, including hand-crafted and deep learning-based approaches, in terms of overall accuracy, average accuracy, and Kappa coefficient. The model also demonstrates superior performance with limited training data and faster convergence. The source code for HybridSN is available on GitHub.The paper introduces a novel model called HybridSN (Hybrid Spectral Convolutional Neural Network) for hyperspectral image (HSI) classification. HSI classification is crucial for analyzing remotely sensed images, and it involves processing images with varying bands of spectral information. While Convolutional Neural Networks (CNNs) are widely used for visual data processing, most existing methods rely on 2D CNNs, which may not fully capture the spatial and spectral information required for HSI classification. To address this, HybridSN combines a spectral-spatial 3D-CNN with a spatial 2D-CNN. The 3D-CNN captures the joint spatial-spectral feature representation from multiple spectral bands, while the 2D-CNN learns more abstract spatial representations. This hybrid approach reduces computational complexity compared to using a pure 3D-CNN and enhances classification performance. The authors conducted rigorous experiments on three benchmark datasets: Indian Pines, Pavia University, and Salinas Scene. The results show that HybridSN outperforms state-of-the-art methods, including hand-crafted and deep learning-based approaches, in terms of overall accuracy, average accuracy, and Kappa coefficient. The model also demonstrates superior performance with limited training data and faster convergence. The source code for HybridSN is available on GitHub.