Semantic decoupled representation learning for remote sensing image change detection

TL;DR

This paper introduces a semantic decoupled representation learning approach for remote sensing image change detection, leveraging semantic masks to improve object recognition and outperform existing pre-training methods.

Contribution

It proposes a novel semantic decoupling method for representation learning in remote sensing, enhancing change detection performance over traditional pre-training techniques.

Findings

Outperforms ImageNet pre-training in change detection tasks

Uses semantic masks to disentangle representations of different regions

Constructs a new dataset with semantic masks for pre-training

Abstract

Contemporary transfer learning-based methods to alleviate the data insufficiency in change detection (CD) are mainly based on ImageNet pre-training. Self-supervised learning (SSL) has recently been introduced to remote sensing (RS) for learning in-domain representations. Here, we propose a semantic decoupled representation learning for RS image CD. Typically, the object of interest (e.g., building) is relatively small compared to the vast background. Different from existing methods expressing an image into one representation vector that may be dominated by irrelevant land-covers, we disentangle representations of different semantic regions by leveraging the semantic mask. We additionally force the model to distinguish different semantic representations, which benefits the recognition of objects of interest in the downstream CD task. We construct a dataset of bitemporal images with semantic masks in an effortless manner for pre-training. Experiments on two CD datasets show our model outperforms ImageNet pre-training, in-domain supervised pre-training, and several recent SSL methods.