基于深度学习的城市地理国情地表覆盖分类研究

Study on Land Cover Classification of Urban Geographical NationalConditions Based on Deep Learning

作为图像识别的研究热点,利用深度学习对遥感影像进行自动分类具有较强的应用实践价值。本文基于全卷积神经网络的深度学习框架,提出了一套城市地理国情地表覆盖分类技术方法:利用地理国情成果,构建城市遥感影像样例库,训练全卷积神经网络,实现地表覆盖自动分类,并通过相似性系数对专题地物进行变化检测。文章选取了上海局部区域作为实验对象,结果发现该方法可以有效减少时间成本,对人文和自然地理要素之间具有较好的区分度,可以为地理国情成果应用和实践提供新的思路和方法。

Deep learning is a research hotspot in image recognition,which can provide solid basic analysis data for other applied researches.This article puts forward a set of land cover classification methods of urban geographical national conditions monitoring based on dense net framework,including using geographical conditions data,building urban geographical remote sensing image sample library,training the convolutional neural network,and realizing automatic change detection by similarity coefficient.This paper selected the local area in Shanghai as experimental object,and the result shows that the study can significantly reduce time cost on urban surface coverage change detection and have better differentiation degree between the human and natural geographical elements,which may improve production efficiency and provide new ideas and methods for the application and practice of geographical national conditions monitoring results.