Geo-Parcel-Based Change Detection Using Optical and SAR Images in Cloudy and Rainy Areas

Abstract

In this article, we deal with the problem of change detection in cloudy and rainy areas using multisource remote sensing images. While previous methods mostly focus on change detection on pixel or super-pixel levels, in this article, we introduce the concept of geo-parcel and use it as the basic processing unit for our change detection method. Concretely, we first extract geo-parcel from an optical high spatial resolution remote sensing image. Then, we divide each geo-parcel into fine-grained segments with refined boundaries using image segmentation methods. These fine-grained segments are used as the basic processing units for our change detection method. After that, an unsupervised learning-based method is adopted to obtain the difference map by comparing synthetic aperture radar images of two periods. Training samples with labels are automatically generated from the difference map. Finally, a deep neural network is trained using the generated samples and is further used to predict the refined change map. Experiments on the collected images from Gui'an, Guizhou Province, China demonstrate the effectiveness of the proposed method for change detection in a cloudy and rainy area with an overall accuracy surpasses 94%.