This paper proposes a hybrid spectral convolutional neural network (HybridSN) for hyperspectral image (HSI) classification. HybridSN combines a 3D CNN for spectral-spatial feature extraction with a 2D CNN for spatial feature learning. The 3D CNN captures joint spatial-spectral features from spectral bands, while the 2D CNN further learns abstract spatial representations. This hybrid approach reduces model complexity compared to standalone 3D CNNs. The model is tested on three remote sensing datasets: Indian Pines, Pavia University, and Salinas Scene. Results show that HybridSN outperforms state-of-the-art methods in classification accuracy, with high overall accuracy (OA), average accuracy (AA), and Kappa coefficient. The model also demonstrates superior performance on small training data. The proposed method is computationally efficient and achieves fast convergence. The experiments confirm that HybridSN provides better performance than 2D and 3D CNNs alone, especially in handling complex spectral-spatial data. The model is implemented with a 3D CNN followed by a 2D CNN, and the results show that the hybrid approach effectively captures both spectral and spatial information for accurate HSI classification.This paper proposes a hybrid spectral convolutional neural network (HybridSN) for hyperspectral image (HSI) classification. HybridSN combines a 3D CNN for spectral-spatial feature extraction with a 2D CNN for spatial feature learning. The 3D CNN captures joint spatial-spectral features from spectral bands, while the 2D CNN further learns abstract spatial representations. This hybrid approach reduces model complexity compared to standalone 3D CNNs. The model is tested on three remote sensing datasets: Indian Pines, Pavia University, and Salinas Scene. Results show that HybridSN outperforms state-of-the-art methods in classification accuracy, with high overall accuracy (OA), average accuracy (AA), and Kappa coefficient. The model also demonstrates superior performance on small training data. The proposed method is computationally efficient and achieves fast convergence. The experiments confirm that HybridSN provides better performance than 2D and 3D CNNs alone, especially in handling complex spectral-spatial data. The model is implemented with a 3D CNN followed by a 2D CNN, and the results show that the hybrid approach effectively captures both spectral and spatial information for accurate HSI classification.