Open AccessAn Application of Deep Learning in Remote Sensing: Automatic Change Detection in Urban Area
Author(s) -
Yang Chun,
Bowen Zhou
Publication year - 2021
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1769/1/012012
Subject(s) - change detection , deep learning , computer science , pipeline (software) , remote sensing , change analysis , artificial intelligence , urbanization , task (project management) , geology , geography , physical geography , management , economic growth , economics , programming language
Change detection of earth’s appearance is a core task for urban management, whose main goal is to identify the change of physical materials via comparing remote sensed images (satellite images, aerial images etc.) of many time epochs. For instance, due to urbanization bare soil can be changed to building; river or lake can be filled up by soil for greening purpose. Recently, deep learning based methods for change detection have been widely applied. The standard pipeline here is: given two images of two epochs and the corresponding changed binary labels, they are feed into a deep learning model to learn the change incidents. However, this kind of change detection focuses solely separating change from no-change, yet ignores the information of “from-to”. For example, the change from soil to building is different to the change from water to soil, but in the standard pipeline they are taken as the same changed incident. In this paper, we propose a deep learning method to tackle this problem, i.e. not only detecting the change incidents, but also predict the change types (the “from-to” information). Our methods are evaluated on the competition dataset released by the SenseTime, and achieve promising detection results i.e. >70% in terms of OA.