Classifying hyperspectral images is a key area of study in remote sensing image analysis and the application of remote sensing technologies. Convolutional neural networks (CNNs) are widely employed in deep learning for visual data processing, particularly in hyperspectral image classification (HIC) where 2D CNN architectures are predominantly utilized. However, the performance of HSI classification heavily depends on both spatial and spectral information. Due to their increased computational complexity, only a few methods have explored the use of 3D CNNs. In this paper, we introduce a hybrid model that integrates both 3D CNNs and 2D CNNs for hyperspectral image (HSI) classification. The 3D CNNs extract joint spatial-spectral features, while the 2D CNNs, applied on top of the 3D CNNs, capture more abstract spatial features. This combination reduces the model’s complexity compared to using 3D CNNs exclusively. Additionally, the model incorporates spatial and channel attention mechanisms to emphasize important regions. The classification experiments were carried out on Indian Pines, Pavia University and Salinas Scene remote sensing datasets, and the results were compared with the most advanced end-to-end deep learning methods to prove the effectiveness of HyMSCA. The experimental results demonstrate that HyMSCA significantly improves classification accuracy.