Hard Region Aware Network for Remote Sensing Change Detection

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

2024-10-18

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.

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

### What problems does the paper attempt to solve?

The paper aims to solve a key challenge in the remote - sensing image change detection (CD) task: detecting changes in difficult regions (i.e., change boundaries and irrelevant pseudo - changes caused by background clutter). Specifically, traditional CD methods treat all regions equally, which makes it difficult for them to accurately identify changes in these difficult regions.

To solve this problem, the authors propose a new change - detection network - HRANet (Hard Region Aware Network). HRANet achieves precise positioning and feature extraction of difficult regions by introducing an online hard region estimation branch and a cross - layer knowledge review module, thereby improving the accuracy of change detection.

#### Main contributions:
1. **Introduction of difficult - region mining strategy**: HRANet applies difficult - sample - mining techniques to the change - detection task to improve the detection accuracy of changes in difficult regions.
2. **Online hard region estimation branch**: This branch can dynamically estimate pixel - level difficult samples and be supervised according to the error between the predicted change map and the ground - truth label during the training process. These difficult - region features are then combined with multi - scale temporal - difference features to guide the model to focus on difficult changes.
3. **Cross - layer knowledge review module**: This module uses reverse - attention and conflict - attention mechanisms to effectively distill temporal - change information from low - level features to high - level features, enhancing the expressive ability of multi - scale temporal - difference features.
4. **Experimental verification**: Experimental results on two benchmark datasets show that HRANet outperforms several existing methods in the change - detection task.

### Formula representation

1. **Difficult - region supervision formula**:
\[
g_u = p_c\cdot(1 - g_c)+g_c\cdot(1 - p_c)
\]
where \( g_u \) represents the difficult - region supervision signal, \( p_c \) represents the predicted change map, and \( g_c \) represents the corresponding ground - truth change.

2. **Binary cross - entropy loss function**:
\[
L_{u}=L_{\text{bce}}(p_u, g_u)=g_u\cdot\log(p_u)+(1 - g_u)\cdot\log(1 - p_u)
\]
where \( L_u \) represents the difficult - region loss and \( p_u \) represents the predicted difficult - region map.

3. **Total training loss function**:
\[
L = L_c+L_u
\]
where \( L_c \) is the change - detection loss, including the binary cross - entropy loss \( L_{\text{bce}} \) and the Dice loss \( L_{\text{dice}} \).

### Summary

HRANet significantly improves the performance in change detection in difficult regions by introducing the difficult - region mining strategy and the cross - layer knowledge review module. The experimental results prove the effectiveness and superiority of this method.