Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Hard Region Aware Network for Remote Sensing Change Detection

Zhenglai Li, Chang Tang, Xinwang Liu, Xingchen Hu, Xianju Li, Ning Li, Changdong Li

TL;DR

A novel change detection network, termed as HRANet, which provides accurate change maps via hard region mining and a cross-layer knowledge review module is introduced to distill temporal change information from low-level to high-level features, thereby enhancing the feature representation capabilities.

Abstract

Change detection (CD) is essential for various real-world applications, such as urban management and disaster assessment. Numerous CD methods have been proposed, and considerable results have been achieved recently. However, detecting changes in hard regions, i.e., the change boundary and irrelevant pseudo changes caused by background clutters, remains difficult for these methods, since they pose equal attention for all regions in bi-temporal images. This paper proposes a novel change detection network, termed as HRANet, which provides accurate change maps via hard region mining. Specifically, an online hard region estimation branch is constructed to model the pixel-wise hard samples, supervised by the error between predicted change maps and corresponding ground truth during the training process. A cross-layer knowledge review module is introduced to distill temporal change information from low-level to high-level features, thereby enhancing the feature representation capabilities. Finally, the hard region aware features extracted from the online hard region estimation branch and multi-level temporal difference features are aggregated into a unified feature representation to improve the accuracy of CD. Experimental results on two benchmark datasets demonstrate the superior performance of HRANet in the CD task.

Hard Region Aware Network for Remote Sensing Change Detection

TL;DR

A novel change detection network, termed as HRANet, which provides accurate change maps via hard region mining and a cross-layer knowledge review module is introduced to distill temporal change information from low-level to high-level features, thereby enhancing the feature representation capabilities.

Abstract

Change detection (CD) is essential for various real-world applications, such as urban management and disaster assessment. Numerous CD methods have been proposed, and considerable results have been achieved recently. However, detecting changes in hard regions, i.e., the change boundary and irrelevant pseudo changes caused by background clutters, remains difficult for these methods, since they pose equal attention for all regions in bi-temporal images. This paper proposes a novel change detection network, termed as HRANet, which provides accurate change maps via hard region mining. Specifically, an online hard region estimation branch is constructed to model the pixel-wise hard samples, supervised by the error between predicted change maps and corresponding ground truth during the training process. A cross-layer knowledge review module is introduced to distill temporal change information from low-level to high-level features, thereby enhancing the feature representation capabilities. Finally, the hard region aware features extracted from the online hard region estimation branch and multi-level temporal difference features are aggregated into a unified feature representation to improve the accuracy of CD. Experimental results on two benchmark datasets demonstrate the superior performance of HRANet in the CD task.
Paper Structure (22 sections, 13 equations, 4 figures, 2 tables)

This paper contains 22 sections, 13 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Traditional CD Methods
  4. Deep Learning-based CD Methods
  5. Hard Sample Mining Strategy
  6. Proposed Method
  7. Overview
  8. Online Hard Region Estimation
  9. Cross-layer Knowledge Review Module
  10. Loss Function
  11. Experiments
  12. Datasets
  13. Evaluation Metrics
  14. Implementation details
  15. Ablation Studies
  16. ...and 7 more sections

Figures (4)

  • Figure 1: A framework of the proposed HRANet. Initially, the bi-temporal images pass through a shared feature extractor to obtain bi-temporal features, and then multi-level temporal difference features are obtained through the TDE. The OHRE branch estimates pixel-wise hard samples corresponding of changed and unchanged regions, supervised by the diversity between predicted change maps and corresponding ground truth in the training process. CKRMs fully explore the multi-level temporal difference knowledge to enhance the feature capabilities. Finally, the multi-level temporal difference features and hard region-aware features obtained from the OHRE branch are aggregated to generate the final change maps.
  • Figure 2: Illustration of the feature aggregation module (FAM), temporal difference extracting (TDE), and online hard region estimation (OHRE) branch.
  • Figure 3: Illustration of the cross-layer knowledge review module (CKRM).
  • Figure 4: Visual comparisons of the proposed method and the state-of-the-art approaches on the LEVIR+ dataset. (a) $t_1$ images; (b) $t_2$ images; (c) Ground-truth; (d) L-Unet; (e) DSIFN; (f) SNUNet; (g) BIT; (h) MSCANet; (i) TFI-GR; (j) A2Net; (k) Ours, (l) Predicted hard region maps. The rendered colors represent true positives (white), false positives ( red), true negatives (black), and false negatives ( blue).