Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Segment Change Model (SCM) for Unsupervised Change detection in VHR Remote Sensing Images: a Case Study of Buildings

Xiaoliang Tan, Guanzhou Chen, Tong Wang, Jiaqi Wang, Xiaodong Zhang

TL;DR

This work tackles unsupervised change detection in very-high-resolution remote sensing images by leveraging Vision Foundation Models. The Segment Change Model (SCM) fuses multi-scale features from FastSAM through a Recalibrated Feature Fusion module and applies a Piecewise Semantic Attention mechanism that uses CLIP-derived semantic cues to filter pseudo-changes, followed by a pixel-wise cosine distance and global OTSU threshold to produce a binary change map. The authors demonstrate substantial mIoU gains on LEVIR-CD and WHU-CD (e.g., from 46.09% to 53.67% and from 47.56% to 52.14%, respectively), with ablations confirming the contributions of RFF and PSA. This approach shows that zero-shot semantic guidance can meaningfully enhance unsupervised change detection in remote sensing, offering a practical workflow for building-change analysis without labeled data.

Abstract

The field of Remote Sensing (RS) widely employs Change Detection (CD) on very-high-resolution (VHR) images. A majority of extant deep-learning-based methods hinge on annotated samples to complete the CD process. Recently, the emergence of Vision Foundation Model (VFM) enables zero-shot predictions in particular vision tasks. In this work, we propose an unsupervised CD method named Segment Change Model (SCM), built upon the Segment Anything Model (SAM) and Contrastive Language-Image Pre-training (CLIP). Our method recalibrates features extracted at different scales and integrates them in a top-down manner to enhance discriminative change edges. We further design an innovative Piecewise Semantic Attention (PSA) scheme, which can offer semantic representation without training, thereby minimize pseudo change phenomenon. Through conducting experiments on two public datasets, the proposed SCM increases the mIoU from 46.09% to 53.67% on the LEVIR-CD dataset, and from 47.56% to 52.14% on the WHU-CD dataset. Our codes are available at https://github.com/StephenApX/UCD-SCM.

Segment Change Model (SCM) for Unsupervised Change detection in VHR Remote Sensing Images: a Case Study of Buildings

TL;DR

This work tackles unsupervised change detection in very-high-resolution remote sensing images by leveraging Vision Foundation Models. The Segment Change Model (SCM) fuses multi-scale features from FastSAM through a Recalibrated Feature Fusion module and applies a Piecewise Semantic Attention mechanism that uses CLIP-derived semantic cues to filter pseudo-changes, followed by a pixel-wise cosine distance and global OTSU threshold to produce a binary change map. The authors demonstrate substantial mIoU gains on LEVIR-CD and WHU-CD (e.g., from 46.09% to 53.67% and from 47.56% to 52.14%, respectively), with ablations confirming the contributions of RFF and PSA. This approach shows that zero-shot semantic guidance can meaningfully enhance unsupervised change detection in remote sensing, offering a practical workflow for building-change analysis without labeled data.

Abstract

The field of Remote Sensing (RS) widely employs Change Detection (CD) on very-high-resolution (VHR) images. A majority of extant deep-learning-based methods hinge on annotated samples to complete the CD process. Recently, the emergence of Vision Foundation Model (VFM) enables zero-shot predictions in particular vision tasks. In this work, we propose an unsupervised CD method named Segment Change Model (SCM), built upon the Segment Anything Model (SAM) and Contrastive Language-Image Pre-training (CLIP). Our method recalibrates features extracted at different scales and integrates them in a top-down manner to enhance discriminative change edges. We further design an innovative Piecewise Semantic Attention (PSA) scheme, which can offer semantic representation without training, thereby minimize pseudo change phenomenon. Through conducting experiments on two public datasets, the proposed SCM increases the mIoU from 46.09% to 53.67% on the LEVIR-CD dataset, and from 47.56% to 52.14% on the WHU-CD dataset. Our codes are available at https://github.com/StephenApX/UCD-SCM.
Paper Structure (8 sections, 3 equations, 2 figures, 2 tables)

This paper contains 8 sections, 3 equations, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Overview
  4. Recalibrated Feature Fusion Module
  5. Piecewise Semantic Attention
  6. Experiments
  7. Conclusion
  8. Acknowledgement

Figures (2)

  • Figure 1: Framework of Segment Change Model (SCM).
  • Figure 2: Visualization results of different methods on WHU-CD dataset. (a) Image $X_1$. (b) Image $X_2$. (c) Ground Truth. (d) PCA-KM. (e) CNN-CD. (f) DSFA. (g) DCVA. (h) GMCD. (i) SCM (Ours). In change maps, white, black, red, and green represent TP, TN, FP, and FN, respectively.