Edge-Bound Change Detection in Multisource Remote Sensing Images

Zhijuan Su,Gang Wan,Wenhua Zhang,Zhanji Wei,Yitian Wu,Jia Liu,Yutong Jia,Dianwei Cong,Lihuan Yuan

DOI: https://doi.org/10.3390/electronics13050867

IF: 2.9

2024-02-24

Electronics

Abstract:Detecting changes in multisource heterogeneous images is a great challenge for unsupervised change detection methods. Image-translation-based methods, which transform two images to be homogeneous for comparison, have become a mainstream approach. However, most of them primarily rely on information from unchanged regions, resulting in networks that cannot fully capture the connection between two heterogeneous representations. Moreover, the lack of a priori information and sufficient training data makes the training vulnerable to the interference of changed pixels. In this paper, we propose an edge-oriented generative adversarial network (EO-GAN) for change detection that indirectly translates images using edge information, which serves as a core and stable link between heterogeneous representations. The EO-GAN is composed of an edge extraction network and a reconstructive network. During the training process, we ensure that the edges extracted from heterogeneous images are as similar as possible through supplemented data based on superpixel segmentation. Experimental results on both heterogeneous and homogeneous datasets demonstrate the effectiveness of our proposed method.

engineering, electrical & electronic,computer science, information systems,physics, applied

What problem does this paper attempt to address?

This paper mainly focuses on the problem of detecting edge boundary changes in multi-source remote sensing images, which is a challenging task for unsupervised change detection methods. Current methods mainly rely on the information of the unchanged regions, but fail to capture the connection between two different representations sufficiently. In addition, the lack of prior information and sufficient training data makes the training easily interfered by changed pixels.

To address these issues, the paper proposes an Edge-based Generative Adversarial Network (EO-GAN). This network utilizes edge information as the core and stable connection between heterogeneous representations to indirectly perform image transformation. It consists of an edge extraction network and a reconstruction network, and in the training process, additional data based on superpixel segmentation is used to ensure the extracted edges from the heterogeneous images are as similar as possible. Experimental results demonstrate the effectiveness of the proposed method on both heterogeneous and homogeneous datasets.

In conclusion, the paper attempts to address how to more effectively perform change detection in multi-source remote sensing images, particularly when dealing with heterogeneous images acquired by different types of sensors, by leveraging edge information to improve unsupervised detection methods.