Fully Convolutional Siamese Networks for Change Detection
This work addresses efficient change detection for large-scale Earth observation systems like Copernicus or Landsat, representing an incremental improvement in speed and performance.
The paper tackles change detection in coregistered images by proposing three fully convolutional neural network architectures, including two Siamese extensions, achieving better performance than previous methods and being at least 500 times faster.
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.