基于深度学习的多光谱遥感影像土地利用类型分类研究  

Classification of land use types in multispectral remote sensing images based on deep learning

作　　者：柴生亮 张艳秋 CHAI Sheng-liang;ZHANG Yan-qiu

机构地区：[1]兰州资源环境职业技术大学,甘肃兰州730000

出　　处：《智能城市》2024年第4期8-11,共4页Intelligent City

基　　金：兰州资源环境职业技术大学科研能力提升项目(X2023A-02)。

摘　　要：文章提出一种针对多光谱遥感影像的土地利用类型分类网络(MSNet),该网络具有可见光波段特征和红外波段特征,在ResNet-50作为主干网络的基础上对网络不同阶段进行优化和改进。将MSNet与经典影像分割模型UNet进行对比,结果显示MSNet结果更优,精度、F1分数和召回率各评价指标均在(0.8,1)区间内,实验证明MSNet可以更充分地利用多光谱特征,提升遥感影像分类准确度。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 

分 类 号：P237[天文地球—摄影测量与遥感]