Advanced Feature Manipulation for Enhanced Change Detection Leveraging Natural Language Models

TL;DR

The paper tackles robust change detection in bi-temporal remote-sensing imagery where environmental noise obscures true semantic changes. It advances the methodology by integrating two novel feature attention mechanisms into a diffusion-model-based change detection pipeline and introducing Flow Dual-Alignment Fusion (FDAF) to extract, align, and fuse diffusion-derived features across time using Flownet-driven warping. This approach aims to improve semantic relevance of features and suppress noise, potentially outperforming DDPM-cd baselines. The work lays groundwork for future experiments in remote sensing change detection and provides a practical framework for more reliable monitoring of land-use and infrastructure changes.

Abstract

Change detection is a fundamental task in computer vision that processes a bi-temporal image pair to differentiate between semantically altered and unaltered regions. Large language models (LLMs) have been utilized in various domains for their exceptional feature extraction capabilities and have shown promise in numerous downstream applications. In this study, we harness the power of a pre-trained LLM, extracting feature maps from extensive datasets, and employ an auxiliary network to detect changes. Unlike existing LLM-based change detection methods that solely focus on deriving high-quality feature maps, our approach emphasizes the manipulation of these feature maps to enhance semantic relevance.