Consistency-guided lightweight network for semi-supervised binary change detection of buildings in remote sensing images

Qing Ding,Zhenfeng Shao,Xiao Huang,Xiaoxiao Feng,Orhan Altan,Bin Hu

DOI: https://doi.org/10.1080/15481603.2023.2257980

IF: 6.397

2023-09-19

GIScience & Remote Sensing

Abstract:Precise identification of binary building changes through remote sensing observations plays a crucial role in sustainable urban development. However, many supervised change detection (CD) methods overly rely on labeled samples, thus limiting their generalizability. In addition, existing semi-supervised CD methods suffer from instability, complexity, and limited applicability. To overcome these challenges and fully utilize unlabeled samples, we proposed a consistency-guided lightweight semi-supervised binary change detection method (Semi-LCD). We designed a lightweight dual-branch CD network to extract image features while reducing model size and complexity. Semi-LCD fully exploits unlabeled samples by data augmentation, consistency regularization, and pseudo-labeling, thereby enhancing its detection performance and generalization capability. To validate the effectiveness and superior performance of Semi-LCD, we conducted experiments on three building CD datasets. Detection results indicate that Semi-LCD outperforms competing methods, quantitatively and qualitatively, achieving the optimal balance between performance and model size. Furthermore, ablation experiments validate the robustness and advantages of the Semi-LCD in effectively utilizing unlabeled samples.

geography, physical,remote sensing

What problem does this paper attempt to address?

### The Problem Addressed by the Paper

This paper aims to address the issue of Binary Building Change Detection (BCD) in remote sensing images. Specifically:

1. **Over-reliance on Labeled Samples**:
- Many current supervised change detection methods overly depend on labeled samples, which limits their generalization ability.

2. **Shortcomings of Existing Semi-supervised Methods**:
- Existing semi-supervised change detection methods suffer from instability, complexity, and limited applicability.

To overcome these issues and fully utilize unlabeled samples, the authors propose a lightweight semi-supervised binary change detection method based on consistency guidance (Semi-LCD). This method leverages data augmentation, consistency regularization, and pseudo-labeling techniques to make full use of unlabeled samples, thereby improving detection performance and generalization ability. Additionally, Semi-LCD designs a lightweight dual-branch network structure to reduce model size and complexity while maintaining excellent detection performance. Experimental results show that Semi-LCD performs well on multiple datasets, achieving the best balance between performance and model size.