摘要
为准确、高效地获取城市建设用地信息,利用目前应用广泛的深度学习技术对该领域的提取效果进行对比。选用曲靖市主城区作为研究区域,以Landsat8OLI_TIRS多光谱影像为原始数据,使用卷积神经网络和BP神经网络两种分类器对影像进行城市建设用地信息提取。使用对象个数、对象面积和地表覆被吻合度3项评价方法提取精度。结果表明,应用卷积神经网络模型的城市建设用地提取具有最高精度,其测试集精度依次达到了92.99%、94.78%和89.64%,均高于常用的BP神经网络。因此,基于卷积神经网络的多光谱影像建设用地提取方法是准确获取城市建设用地信息的一种可行方法,为滇中城市建设用地提取研究提供了参考。
In order to accurately and quickly obtain information on urban construction land,this paper compares and studies the extraction effect of deep learning technology widely used in this field.In this paper,the main urban area of Qujing City is selected as the study area.Landsat8 OLI_TIRS multispectral imagers were used as the original data,and two types of classifiers,Convolutional Neural Network and Back Propagation Network,are used to construct the image of land use information extraction.The results of the three extraction accuracy evaluation methods using the number of objects,the area of the object and the degree of surface cover match show that the urban construction land extraction using the convolutional neural network model has the highest accuracy,and the accuracy of the test set reaches 92.99%94.78%and 89.64%in sequence,which were higher than the commonly used BP neural network.The results show that the method of extracting construction land multi-spectral images based on convolutional neural network is a feasible method to obtain accurate urban construction land information,which provides a reference for the research on urban construction land information extraction in central Yunnan.
作者
陈磊士
赵俊三
董智文
朱褀夫
CHEN Lei-shi;ZHAO Jun-san;DONG Zhi-wen;ZHU Qi-fu(Faculty of Land and Resource Engineering,Kunming University of Science and Technology,Kunming 650093,China;Beijing Changdi Wanfang Technology Co.,Ltd.Foshan Branch,Foshan 528305,China)
出处
《软件导刊》
2018年第11期177-180,186,共5页
Software Guide
基金
国家自然科学基金项目(41761081)
关键词
深度学习
城市建设用地
卷积神经网络
遥感影像
deep learning
urban construction land
convolutional neural network
remote sensing image
