A Transformer-Based Siamese Network for Change Detection

TL;DR

This paper introduces ChangeFormer, a transformer-based Siamese network that improves change detection accuracy in remote sensing images by capturing multi-scale long-range details more effectively than convolutional methods.

Contribution

It proposes a novel transformer-based Siamese architecture for change detection, unifying hierarchical transformer encoders with MLP decoders for enhanced performance.

Findings

Outperforms previous change detection methods on benchmark datasets.

Achieves higher accuracy and better detail preservation.

End-to-end trainable architecture with competitive results.

Abstract

This paper presents a transformer-based Siamese network architecture (abbreviated by ChangeFormer) for Change Detection (CD) from a pair of co-registered remote sensing images. Different from recent CD frameworks, which are based on fully convolutional networks (ConvNets), the proposed method unifies hierarchically structured transformer encoder with Multi-Layer Perception (MLP) decoder in a Siamese network architecture to efficiently render multi-scale long-range details required for accurate CD. Experiments on two CD datasets show that the proposed end-to-end trainable ChangeFormer architecture achieves better CD performance than previous counterparts. Our code is available at https://github.com/wgcban/ChangeFormer.