Fully Convolutional Siamese Networks for Change Detection

TL;DR

This paper introduces three fully convolutional neural network architectures, including two Siamese variants, for efficient change detection in coregistered images, demonstrating superior accuracy and speed on open datasets for Earth observation data.

Contribution

The paper proposes novel Siamese fully convolutional network architectures tailored for change detection, achieving state-of-the-art performance and significantly faster processing times.

Findings

Outperforms previous methods in accuracy

Achieves at least 500 times faster processing speed

Learns effectively from scratch with annotated images

Abstract

This paper presents three fully convolutional neural network architectures which perform change detection using a pair of coregistered images. Most notably, we propose two Siamese extensions of fully convolutional networks which use heuristics about the current problem to achieve the best results in our tests on two open change detection datasets, using both RGB and multispectral images. We show that our system is able to learn from scratch using annotated change detection images. Our architectures achieve better performance than previously proposed methods, while being at least 500 times faster than related systems. This work is a step towards efficient processing of data from large scale Earth observation systems such as Copernicus or Landsat.