Abstract: To address the limitations of existing convolutional neural network (CNN) methods, which exhibit insufficient utilization of spatial-spectral joint features and insufficient attention to global features in the hyperspectral image classification process, a dual-stream hyperspectral image classification network based on multiple convolution and spatial-spectral attention Transformer is designed to achieve the full utilization of both local and global features through a dual-stream structure integrating CNN and Transformer network. First, in the CNN branch, a spatial-spectral feature fusion structure based on multiple convolutions was designed to fully mine and fuse spatial and spectral dimensional features; while in the Transformer network branch, the spatial-spectral attention mechanism was used to capture the global information of the whole image. Finally, two branches achieved high-performance classification results through decision-level fusion. The test results based on four typical datasets show that the classification results of this algorithm are improved to different degrees compared with the current mainstream algorithms.