Aiming at the problems of large change scale of remote sensing images and complex image backgrounds, which lead to poor classification of remote sensing scenes, a Multi-scale Convolution Block Attention Module(MS-CBAM) and an improved residual network-based remote sensing image classification method are proposed. Based on ResNet50, this method introduces multi-scale information into the residual structure and adds multi-scale convolution block attention module to improve the classification accuracy. The average classification accuracy on the public remote sensing image datasets UCMerced_LandUse and AID reached 94.10% and 93.52%. Compared with ResNet50, the classification accuracy was improved by 4.29% and 1.83%, respectively. Meanwhile, additional experiments verify the effectiveness of the multi-scale convolutional block attention module.