Feature Decomposition-Optimization-Reorganization Network for Building Change Detection in Remote Sensing Images

Yuanxin Ye,Zhitao Fu,Bai Zhu,Miaomiao Sun,Chao Yang,Liang Zhou,Jianwei Fan

doi:10.3390/rs14030722

Abstract

Building change detection plays an imperative role in urban construction and development. Although the deep neural network has achieved tremendous success in remote sensing image building change detection, it is still fraught with the problem of generating broken detection boundaries and separation of dense buildings, which tends to produce saw-tooth boundaries. In this work, we propose a feature decomposition-optimization-reorganization network for building change detection. The main contribution of the proposed network is that it performs change detection by respectively modeling the main body and edge features of buildings, which is based on the characteristics that the similarity between the main body pixels is strong but weak between the edge pixels. Firstly, we employ a siamese ResNet structure to extract dual-temporal multi-scale difference features on the original remote sensing images. Subsequently, a flow field is built to separate the main body and edge features. Thereafter, a feature optimization module is designed to refine the main body and edge features using the main body and edge ground truth. Finally, we reorganize the optimized main body and edge features to obtain the output results. These constitute a complete end-to-end building change detection framework. The publicly available building dataset LEVIR-CD is employed to evaluate the change detection performance of our network. The experimental results show that the proposed method can accurately identify the boundaries of changed buildings, and obtain better results compared with the current state-of-the-art methods based on the U-Net structure or by combining spatial-temporal attention mechanisms.

Highlights

Introduction published maps and institutional affilImage change detection denotes the process of recognizing specific differences between multi-temporal images [1,2], which is a key technique for many applications, such as disaster assessment [3,4], land cover change detection [5,6], urban expansion monitoring [7], and so on.As an important part of the blueprint of cities, the demolition, construction and expansion of buildings are closely related to human existence
We introduce the decoupling idea into building change detection and employ the feature optimization structure to refine the main body and edge features, which greatly improves the accuracy of the boundary detection of changed buildings
Building change detection can be regarded as a binary classification task categorized into changed buildings and background

Summary

Introduction

Introduction published maps and institutional affilImage change detection denotes the process of recognizing specific differences between multi-temporal images [1,2], which is a key technique for many applications, such as disaster assessment [3,4], land cover change detection [5,6], urban expansion monitoring [7], and so on.As an important part of the blueprint of cities, the demolition, construction and expansion of buildings are closely related to human existence. It is of great significance to timely and accurately obtain the change information of buildings for human development. With the rapid development of remote sensing imaging technology, massive remote sensing images can be used for building change detection following high-precision co-registration [8]. Building change detection based on remote sensing images has become an area of immense research interest. Research on related change detection methods has made great progress, from the early pixel-based building change detection methods iations. The overall accuracy represents the ratio of correctly detected pixels to total pixels, usually expressed as a percentage. It is defined as follows: Overall accuracy = TC + TB. Where TC is the number of correctly detected pixels of changed buildings, TB is the correctly detected background pixels, FC represents pixels that belong to background but is misidentified as changed buildings, and FB represents changed building pixels classified in background.

Methods

Results

Conclusion