Fully Convolutional Siamese Networks for Change Detection

By Rodrigo Caye Daudt, Bertrand Le Saux, Alexandre Boulch. International Conference on Information Photonics, 2018

1,708 citations recorded by the source corpus.

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.

Subfields: Atmospheric Science, Computer Vision and Pattern Recognition, Media Technology

View Paper: https://www.semanticscholar.org/paper/1d385debcb66dc259119093e9a390b0870e10037

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Atmospheric Science

In Computer Vision and Pattern Recognition

In Media Technology

Citations in the corpus

End-to-End Change Detection for High Resolution Satellite Images Using Improved UNet++ (2019, background): “Two periods of images are concatenated as the input of the network, which proves to be e ﬀ ective for learning bi-temporal changes through deep supervised training [53].”
Remote Sensing Image Change Detection With Transformers (2021, methodology): “approach [3], [22]–[24], [29] concatenates the bitemporal images as a single input to a semantic segmentation net-”
Self-Supervised Change Detection in Multiview Remote Sensing Images (2021, methodology): “Recent success of deep learning techniques in change detection is mainly focused on supervised methods [10]–[12], which are often limited from the availability of annotated datasets.”
Change Detection on Remote Sensing Images Using Dual-Branch Multilevel Intertemporal Network (2023, methodology): “…of our DMINet on bitemporal RS images, several SOTA models are picked as the competitors, including FC-EF [47], FC-Siam-Di [47], FC-Siam-Conc [47], FresUNet [48], image fusion network (IFNet) [35], dual-task constrained deep siamese convolutional network (DTCDSCN) [40], densely connected…”
A Deeply Supervised Attention Metric-Based Network and an Open Aerial Image Dataset for Remote Sensing Change Detection (2021, methodology): “…of DSAMNet, the following ﬁve state-of-the-art CD methods are selected for comparison purposes and introduced in brief: 1) Fully Convolutional-Early Fusion (FC-EF): FC-EF [22] is proposed based on U-Net architecture, in which the bitem-poral images are cascaded as a multiband image for input.”
RS-Mamba for Large Remote Sensing Image Dense Prediction (2024, methodology): “On the CD task, the compared CNN-based models include FC-EF [36], FC-Siam-Diff [36], FC-Siam-Conc [36], STANet [39], DTCDSCN [54], SNUNet [55], CDNet [56], DDCNN [57], DASNet [58], DSIFN [59], and HANet [60], the compared transformer-based models include Change-Former [9], and the CNN-transformer…”
SemiCDNet: A Semisupervised Convolutional Neural Network for Change Detection in High Resolution Remote-Sensing Images (2020, background): “Note that two periods of image clips can be stacked [33], [34] or act as independent branches [35] when they are fed into different FCN architectures.”
ChangeMamba: Remote Sensing Change Detection With Spatiotemporal State Space Model (2024, methodology): “Convolutional neural network (CNN)-based methods have been dominant within the field of CD after the introduction of fully convolutional networks (FCNs) to the field by Daudt et al. [16].”
Changer: Feature Interaction is What You Need for Change Detection (2022, methodology): “[12] proposes three typical change detection models: FC-EF, FC-Siam-Conc and FC-Siam-Diff.”
Towards Geospatial Foundation Models via Continual Pretraining (2023, methodology): “The difference between the features from each pair is computed and fed into an UPerNet [41] to generate the final binary segmentation masks [30, 4].”
DynamicEarthNet: Daily Multi-Spectral Satellite Dataset for Semantic Change Segmentation (2022, methodology): “Classical approaches define axiomatic, pixel-based [4–6, 20, 35] algorithms to obtain change whereas many recent approaches are datadriven [7, 12, 33, 47].”
A Transformer-Based Siamese Network for Change Detection (2022, methodology): “In this section, we compare the CD performance of our ChangeFormer with existing SOTA methods: • FC-EF [12]: concatenates bi-temporal images and processes them through a ConvNet to detect changes.”
SNUNet-CD: A Densely Connected Siamese Network for Change Detection of VHR Images (2021, methodology): “FC-EF, FC-Siam-conc, and FC-Siam-diff [5] are baseline models for change detection, they are simple combinations of UNet and siamese network.”
A Spatial-Temporal Attention-Based Method and a New Dataset for Remote Sensing Image Change Detection (2020, background): “OSCD, introduced last year, has also driven a few studies [67,69].”
SwinSUNet: Pure Transformer Network for Remote Sensing Image Change Detection (2022, methodology): “FC-Siam-conc [20] has a siamese U-shaped structure.”