摘要
为了解决由于空间分辨率低导致城市核心地物要素提取精度低的问题,笔者构建了适用于城市核心地物要素提取的卷积神经网络模型(convolutional neural network,CNNs),同时利用基于遥感影像提取的颜色、纹理和形状等多特征信息辅助城市地物要素提取。颜色、纹理和形状等多特征信息与CNNs模型提取的深层信息相结合,融合浅层与深层特征信息,充分挖掘影像数据中的高级语义信息,能够有效地提高结果精度。实验结果表明,CNNs模型减少了模型参数量与复杂度,模型收敛速度快,具有较高的提取精度、泛化能力和鲁棒性,总体精度为97.49%,Kappa系数为0.9575。该研究可以为城市发展与规划提供技术支持。
In order to address the issue of low extraction accuracy of urban core feature elements caused by low spatial resolution,the authors have developed convolutional neural networks(CNNs)applicable to the extraction of urban core feature elements.At the same time,multi-feature information such as color,texture,and shape extracted based on remote sensing images have been used to assist the extraction of urban feature elements.The CNNs model fully exploits the high-level semantic information in the image data by combining multi-feature information,such as color,texture,and shape,with the deep information it has extracted.This combination can significantly increase the accuracy of the results.According to the experimental findings,the CNNs model decreases the number and complexity of model parameters,converges quickly,and exhibits good extraction accuracy,generalization ability,and robustness.Its overall accuracy is 97.49%,and its Kappa coefficient is 0.9575.Thus,the research can offer technical assistance for urban planning and development.
作者
罗修杰
刘子维
徐梦霞
王明常
LUO Xiujie;LIU Ziwei;XU Mengxia;WANG Mingchang(Emergency Management Bureau of Futian District,Shenzhen Municipality,Shenzhen 518017,Guangdong,China;College of Geo-Exploration Science and Technology,Jilin University,Changchun 130026,China)
出处
《世界地质》
CAS
2023年第4期731-739,共9页
World Geology
基金
吉林省自然科学基金项目(20210101098JC)
吉林省教育厅科学研究项目(JJKH20231181KJ)联合资助。
关键词
城市核心地物要素
多特征信息
语义信息
卷积神经网络
urban core feature elements
multi-feature information
semantic information
convolutional neural network
