Residual Unet for Urban Building Change Detection with Sentinel-1 SAR Data

Urban building change detection is one of the most important parts of remote sensing applications. Researching change detection method based on deep learning is an effective solution to monitor the urban expansion and recognize the specific change classes. In this paper, we propose a novel urban bui...

Full description

Saved in:
Bibliographic Details
Published in:IEEE International Geoscience and Remote Sensing Symposium proceedings pp. 1498 - 1501
Main Authors: Li, Lu, Wang, Chao, Zhang, Hong, Zhang, Bo
Format: Conference Proceeding
Language:English
Published: IEEE 01.07.2019
Subjects:
Buildings
change detection
feature combination
Feature extraction
Image color analysis
Radar polarimetry
Remote sensing
residual Unet
stacked convolutional autoencoder
Training
Urban areas
urban building
ISSN:2153-7003
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Urban building change detection is one of the most important parts of remote sensing applications. Researching change detection method based on deep learning is an effective solution to monitor the urban expansion and recognize the specific change classes. In this paper, we propose a novel urban building change detection method based on the revised residual Unet with Sentinel-1 SAR intensity images. Firstly, we present a new difference image by combing both the original intensity image and the enhanced log-ratio difference image using a non-linear function. Then, the combined difference image is sent to a revised residual Unet network to detect the building changes. By the proposed combined difference image and the revised network, our method is able to focus on the building's change while ignoring other land type changes in a large area. A pair of real bitemporal SAR images is used to test the proposed approach and the obtained experimental results confirm its effectiveness.
ISSN:2153-7003
DOI:10.1109/IGARSS.2019.8898146