基于深度学习的遥感图像土地分类研究  

Remote sensing image land classification using machine learning

作　　者：张凯 于航[2] Zhang Kai;Yu Hang(School of Computing,Changchun University of Finance and Economics,Jilin,Changchun 130122,China;School of Computing,Jilin Agricultural University)

机构地区：[1]长春财经学院,吉林长春130122 [2]吉林农业大学

出　　处：《计算机时代》2022年第9期108-110,114,共4页Computer Era

基　　金：吉林省哲学社会科学规划基金项目“大数据驱动城乡融合发展对吉林省缩小城乡差距的影响研究”(2021J32)。

摘　　要：针对采用单一特征描述技术存在的高分辨率遥感图像特征多、分类精度低的问题,提出一种基于深度学习的遥感图像土地分类模型。模型对高分1号卫星获得的遥感图像进行了预处理,采用特征级图像融合的方法将这些特征关联起来,实现遥感图像特征融合。将融合后的图像特征输入训练后的深度信念网络进行处理,由Softmax分类器获得土地分类。实验分析表明,该模型能够对所有土地类型进行清晰的分类。Aiming at the problems of multiple features and low classification accuracy of high resolution remote sensing images using single feature description technology, a land classification model of remote sensing images based on deep learning is proposed. It preprocesses the remote sensing images obtained by Gaofen-1 satellite, and uses the feature-level image fusion method to associate these features to achieve the feature fusion of remote sensing images. The fused image features are input into the trained deep belief network for processing, and land classification is obtained by Softmax classifier. Experimental analysis shows that the model can clearly classify all land types.

关 键 词：遥感图像 深度学习 深度信念网络 特征融合 

分 类 号：TP389.1[自动化与计算机技术—计算机系统结构]