SAR Image Change Detection Based on Complex Neural Network

Abstract

With the deepening of the world exploration, more and more observation projects are needed, and the requirements of observation are also higher and higher. The development of remote sensing image change detection requires higher accuracy. Synthetic aperture radar (SAR) has a special advantage in the field of image change detection because it is easy to obtain images and is not affected by atmospheric and light conditions. Therefore, in recent decades, SAR images have been successfully applied to environmental monitoring, urban research and disaster assessment. In recent years, neural network deep learning has received great attention and rapid development. The existing deep learning is basically based on the operation of real number. We need to consider how to connect the original images captured by multiple satellites to the network. Most of the existing methods use the direction of image color channel to splice images, but the color channels are not related therefore, this method will hide the association between multiple images. In this paper, a SAR image change detection method based on complex neural network is proposed. The complex neural network is used in SAR image change detection. Through the deep learning of complex neural network, the correlation between images is obtained, which effectively improves the effect of SAR image change detection. The principle of this method is that the original two groups of data in the detection problem are naturally combined into complex numbers, and then the complex convolution and full connection are used to process the data in the network, which effectively improves the accuracy of change detection.