In this paper,a land use classification network(MSNet)for multispectral remote sensing images is proposed,which has the characteristics of visible light band and infrared band.Based on ResNet-50 as the backbone network,different stages of the network are optimized and improved.Comparing MSNet with classical image segmentation model UNet,the results show that MSNet is better,and the evaluation indexes of accuracy,F1 score and recall rate are all in the range of(0.8,1).Experiments show that MSNet can make full use of multispectral features and improve the classification accuracy of remote sensing images.
关键词
深度学习/土地利用类型/多光谱信息/遥感影像
Key words
deep learning/land use type/multispectral information/remote sensing image
维普
万方数据