Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SRC-Net: Bi-Temporal Spatial Relationship Concerned Network for Change Detection

Hongjia Chen, Xin Xu, Fangling Pu

TL;DR

SRC-Net tackles change detection by explicitly leveraging bi-temporal spatial relationships through a credibility-guided cross-branch Perception and Interaction Module (PIM) and a Patch-Mode joint Feature Fusion Module (PM-FFM). The methods preserve temporal correspondences during feature extraction and introduce mode-aware, patch-based fusion to prevent information loss. Empirical results on LEVIR-CD and WHU Building show state-of-the-art F1 scores with a lightweight 5.17M-parameter backbone, validating the effectiveness of cross-branch interaction and mode-aware fusion. The work offers a new paradigm for exploiting bi-temporal spatial relationships in remote sensing CD, with potential extensions to multi-temporal scenarios and edge-friendly implementations.

Abstract

Change detection (CD) in remote sensing imagery is a crucial task with applications in environmental monitoring, urban development, and disaster management. CD involves utilizing bi-temporal images to identify changes over time. The bi-temporal spatial relationships between features at the same location at different times play a key role in this process. However, existing change detection networks often do not fully leverage these spatial relationships during bi-temporal feature extraction and fusion. In this work, we propose SRC-Net: a bi-temporal spatial relationship concerned network for CD. The proposed SRC-Net includes a Perception and Interaction Module that incorporates spatial relationships and establishes a cross-branch perception mechanism to enhance the precision and robustness of feature extraction. Additionally, a Patch-Mode joint Feature Fusion Module is introduced to address information loss in current methods. It considers different change modes and concerns about spatial relationships, resulting in more expressive fusion features. Furthermore, we construct a novel network using these two relationship concerned modules and conducted experiments on the LEVIR-CD and WHU Building datasets. The experimental results demonstrate that our network outperforms state-of-the-art (SOTA) methods while maintaining a modest parameter count. We believe our approach sets a new paradigm for change detection and will inspire further advancements in the field. The code and models are publicly available at https://github.com/Chnja/SRCNet.

SRC-Net: Bi-Temporal Spatial Relationship Concerned Network for Change Detection

TL;DR

SRC-Net tackles change detection by explicitly leveraging bi-temporal spatial relationships through a credibility-guided cross-branch Perception and Interaction Module (PIM) and a Patch-Mode joint Feature Fusion Module (PM-FFM). The methods preserve temporal correspondences during feature extraction and introduce mode-aware, patch-based fusion to prevent information loss. Empirical results on LEVIR-CD and WHU Building show state-of-the-art F1 scores with a lightweight 5.17M-parameter backbone, validating the effectiveness of cross-branch interaction and mode-aware fusion. The work offers a new paradigm for exploiting bi-temporal spatial relationships in remote sensing CD, with potential extensions to multi-temporal scenarios and edge-friendly implementations.

Abstract

Change detection (CD) in remote sensing imagery is a crucial task with applications in environmental monitoring, urban development, and disaster management. CD involves utilizing bi-temporal images to identify changes over time. The bi-temporal spatial relationships between features at the same location at different times play a key role in this process. However, existing change detection networks often do not fully leverage these spatial relationships during bi-temporal feature extraction and fusion. In this work, we propose SRC-Net: a bi-temporal spatial relationship concerned network for CD. The proposed SRC-Net includes a Perception and Interaction Module that incorporates spatial relationships and establishes a cross-branch perception mechanism to enhance the precision and robustness of feature extraction. Additionally, a Patch-Mode joint Feature Fusion Module is introduced to address information loss in current methods. It considers different change modes and concerns about spatial relationships, resulting in more expressive fusion features. Furthermore, we construct a novel network using these two relationship concerned modules and conducted experiments on the LEVIR-CD and WHU Building datasets. The experimental results demonstrate that our network outperforms state-of-the-art (SOTA) methods while maintaining a modest parameter count. We believe our approach sets a new paradigm for change detection and will inspire further advancements in the field. The code and models are publicly available at https://github.com/Chnja/SRCNet.
Paper Structure (21 sections, 15 equations, 12 figures, 3 tables)

This paper contains 21 sections, 15 equations, 12 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Change Detection Networks
  4. Feature Interaction
  5. Feature Fusion
  6. Methodology
  7. Perception and Interaction Module
  8. Patch-Mode joint Feature Fusion Module
  9. Bi-temporal Sub-features
  10. Mode Perception
  11. Multi-Head Feature Fusion
  12. Network Architecture
  13. Experiments and Analysis
  14. Dataset
  15. Implementation Details
  16. ...and 6 more sections

Figures (12)

  • Figure 1: Bi-temporal remote sensing images. (a) (b) Original images. (c) Ground truth. (d) The result of our SRC-Net. (e) The result of the single-branch network (RDPNet). (f) The result of the dual-branch network without cross-branch interaction (FC-Siam-diff). (g) (h) The results of dual-branch network with cross-branch interaction (BIT, DSIFN). The false positives and false negatives are indicated in red and green, respectively. Other colors represent true positives.
  • Figure 2: Correspondence disappearance in feature concatenation.
  • Figure 3: Original input bi-temporal remote sensing images. (a) and (b) LEVIR-CD dataset, (c) and (d) WHU dataset.
  • Figure 4: The structure of perception and interaction module.
  • Figure 5: The structure of the Patch-Mode joint feature fusion module.
  • ...and 7 more figures