Urban Area Change Detection with Combining CNN and RNN from Sentinel-2 Multispectral Remote Sensing Data

Abstract

Change detection is one of the hot issues related to world observation and has been extensively studied in recent decades. The application of remote sensing technology provides inputs to systems for urban change detection primarily focus on the urban data user environment. Urban change detection refers to the general problem of monitoring the urban system and discerning changes that are occurring within that system that use to urban planners, managers, and researchers. Current methods based on a simple mechanism for independently encoding bi-temporal images to get their representation vectors. In fact, these methods do not make full use of the rich information between bi-temporal images. We propose to combine deep learning methods such as Convolutional Neural Network (U-Net) for feature extraction and Recurrent Neural Network (BiLSTM) temporal modeling. Our developed model while the validation phase gets 97.418% overall accuracy on the Onera Satellite Change Detection (OSCD) Sentinel-2 bi-temporal dataset.