Abstract: Aiming at the problems of insufficient utilization of spatial information and a shortage of sample labels in hyperspectral remote sensing image classification, this paper proposes an algorithm to classify hyperspectral images based on multi-scale 3D-CNN (MS3D-CNN) and Convolutional Block Attention Mechanism (CBAM). The feature mapping approach is employed to fully explore and fuse spatial and spectral features of hyperspectral images from different receptive fields, which are further processed by CBAM. After that, the deep neural network is constructed based on the idea of Residual Network (ResNet), and the Dropout method is encompassed to deal with the over-fitting problem. Finally, the processed features are classified by Softmax classifier. Extensive experiments are conducted on three hyperspectral datasets of Indian Pines, Pavia University and Salinas Valley, and the classification results show that the proposed method is superior to other classical methods.