Convolutional neural networks (CNN) are commonly used for precise hyperspectral image (HSI) classification. However, the representation capability of a single network structure for the spatial-spectral features of hyperspectral images still needs to be improved. We propose a hybrid CNN model to overcome the limitations of a single network model. The characteristics of the hybrid model are manifested in the following three aspects. Firstly, spatial-spectral features are extracted by a combination of 2D/3D residual networks. Secondly, a hybrid approach integrating multiscale features with attention mechanisms is implemented to enhance classification performance. Thirdly, the hybrid attention module integrates both spatial attention feature extraction and spectral attention feature extraction. The proposed model considers the advantages of 3D-CNN in extracting spatial-spectral features and the time and space efficiency of 2D-CNN. At the same time, multi-scale features and spatial-spectral attention features are fused, and the data feature mining ability of the model is improved by giving more attention to important bands and spatial regions, so as to improve the overall performance of the model. Extensive experiments have demonstrated the superiority of our method compared to other approaches.