LDGuid: A Framework for Robust Change Detection via Latent Difference Guidance

2026-05-15

AI Summary

Existing deep change detection models struggle to explicitly model task-relevant semantic differences, leading to limited robustness in complex or noise-corrupted scenarios. This work proposes LDGuid, a novel framework that introduces, for the first time, an explicit semantic difference guidance mechanism. By leveraging an adversarial autoencoder combined with an information bottleneck constraint, LDGuid pretrains a differential embedding module that extracts only the semantic discrepancies between bi-temporal inputs pertinent to change detection. This distilled semantic difference is then injected as a guidance signal into backbone networks such as U-Net, BIT, or AERNet. The approach further accommodates integration of domain knowledge, including spectral indices, and consistently achieves performance gains across multiple benchmarks—LEVIR-CD, WHU-CD, SVCD, and CaBuAr—with particularly pronounced improvements under challenging conditions such as spectral noise.

Abstract

Modern deep learning models for change detection (CD) often struggle to explicitly represent task-relevant semantic differences. This paper proposes the Latent Difference Guidance (LDGuid) framework that explicitly learns and injects semantic differences into CD models. LDGuid deploys adversarial autoencoding to implement a difference embedding (DE) module. The DE module is pretrained via the information bottleneck method, restricting it to learn only task-relevant differences between pre- and post-event samples. The learned latent difference is then used as an explicit guidance signal in the CD model. We validate LDGuid by integrating it into U-Net, BIT, and AERNet baselines for CD and evaluating it on LEVIR-CD, WHU-CD, SVCD, and CaBuAr datasets. Experimental results show that LDGuid enhances segmentation performance across all benchmarks, with particularly remarkable gains in challenging settings affected by spectral noise. The results further highlight the ability of LDGuid in incorporating domain knowledge, such as task-specific spectral indices. Our findings suggest that semantic difference learning can drastically enhance the robustness of CD in remote sensing.

Problem

• change detection
• semantic differences
• robustness
• remote sensing
• spectral noise

Innovation

• Latent Difference Guidance
• Change Detection
• Adversarial Autoencoding
• Information Bottleneck
• Semantic Difference Learning