MASNet:Improve Performance of Siamese Networks with Mutual-attention for Remote Sensing Change Detection Tasks

TL;DR

MASNet enhances Siamese networks for remote sensing change detection by introducing mutual-attention, enabling early information exchange between feature extraction branches, leading to improved performance across CNN and transformer models.

Contribution

This work introduces a mutual-attention plug-in for Siamese networks, improving early feature exchange and overall change detection performance.

Findings

Improved accuracy on multiple change detection datasets

Effective for both CNN and transformer architectures

Mutual-attention enhances early feature fusion

Abstract

Siamese networks are widely used for remote sensing change detection tasks. A vanilla siamese network has two identical feature extraction branches which share weights, these two branches work independently and the feature maps are not fused until about to be sent to a decoder head. However we find that it is critical to exchange information between two feature extraction branches at early stage for change detection task. In this work we present Mutual-Attention Siamese Network (MASNet), a general siamese network with mutual-attention plug-in, so to exchange information between the two feature extraction branches. We show that our modification improve the performance of siamese networks on multi change detection datasets, and it works for both convolutional neural network and visual transformer.