Learning a transferable change detection method by Recurrent Neural Network

Lyu, Haobo; Lu, Hui

doi:10.1109/igarss.2016.7730344

Cited by 12 publications

(8 citation statements)

References 5 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Since then, Siamese networks have been widely utilized as the part of feature extraction for CD [27] [28]. In order to improve the detection accuracy further, some methods introduce long-short term memory (LSTM) networks or recurrent neural networks (RNNs) on this basis to explore spatial-temporal relationships [29]- [31][34], and some methods introduce attention mechanisms to exploit the importance of difference feature maps and spatial positions for improving CD effect. [32][33][53] [54].…”

Section: A Rs Image Change Detection Methodsmentioning

confidence: 99%

Difference Enhancement and Spatial–Spectral Nonlocal Network for Change Detection in VHR Remote Sensing Images

Lei

Wang

Ning

et al. 2022

IEEE Trans. Geosci. Remote Sensing

View full text Add to dashboard Cite

show abstract

Section: A Rs Image Change Detection Methodsmentioning

confidence: 99%

Difference Enhancement and Spatial–Spectral Nonlocal Network for Change Detection in VHR Remote Sensing Images

Lei

Wang

Ning

et al. 2022

IEEE Trans. Geosci. Remote Sensing

View full text Add to dashboard Cite

show abstract

“…Zhang et al proposed a change detection method for SAR (Synthetic Aperture Radar) images on the basis of sparse autoencoders [11]. Lyu used the recursive neural network based on Long Short-Term Memory to solve the change detection task and Landsat data to detect the annual city dynamics [10]. Gong applied the traditional change detection method and the RBM (Restricted Boltzmann Machine) method to SAR images to improve the detection performance [12].…”

Section: Related Workmentioning

confidence: 99%

“…As deep learning technology has achieved remarkable results in the fields of natural language processing [6], image classification [7], and semantic segmentation [8], many researchers began to study the application of deep learning in change detection in high-resolution remote sensing images [9][10][11][12][13][14][15]. Although these methods have achieved good results in applying change detection in high-resolution remote sensing images, they focus mainly on the difference information between multi-temporal data and lack discrimination against pseudo-changes which have not actually occurred or are not the interested changes [16].…”

Section: Introductionmentioning

confidence: 99%

FODA: Building Change Detection in High-Resolution Remote Sensing Images Based on Feature–Output Space Dual-Alignment

Zhang

Deng

et al. 2021

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

In building change detection task, factors such as phenological changes, illumination changes, and registration errors will cause unchanged areas in remote sensing images to have obvious differences in pixels, which will lead to pseudochanges in results. Existing methods focus on the change information of multi-temporal remote sensing images, ignoring the exploration of pseudo-change problems. Therefore, FODA (Feature-Output space Dual-Alignment) method is proposed to reduce the negative effect of the pseudo-change problem by paying attention to the relationship between unchanged areas of multi-temporal images. On the one hand, FODA narrows the distance between the features of the unchanged areas in the feature space, increasing its feature extraction ability of pseudochanged areas. On the other hand, given the spatial context of image scene implicit in the output space, the ability to recognize pseudo-changes of the FODA is improved through an adversarial learning procedure. Due to its simplicity and effectiveness, FODA achieves 88.73% and 82.75% F1 scores on the LEVIR-CD dataset and WHU-CD dataset respectively. Compared with stateof-the-art methods, FODA can effectively reduce the problem of pseudo-changes and significantly improve the effect of change detection even only based on a simple backbone model.

show abstract

“…In this regard, RNN, as an important branch of the deep learning family, is a natural candidate for dealing with temporal relationships between multiple time series data in change detection problems. Reference [14] made RNN-based network to solve the multispectral change detection task, in which, the joint spectral-temporal feature representation is learned from a bi-temporal image sequence using long short-term memory network (LSTM). Reference [15] proposed a novel network architecture, which is trained to learn a joint spectral-spatialtemporal feature representation in a unified framework for change detection of multi-spectral images.…”

Section: Related Workmentioning

confidence: 99%

Remote Sensing Image Change Detection Based on Information Transmission and Attention Mechanism

et al. 2019

View full text Add to dashboard Cite

Change detection is one of the core issues of earth observation and has been extensively studied in recent decades. This paper presents a novel deep neural network architecture based on information transmission and attention mechanism. Existing methods rely on a simple mechanism for independently encoding bi-temporal images to obtain their representation vectors. In view of the fact that these methods do not make full use of the rich information between bi-temporal images, we introduce the information transmission module in the design of DNN structure for doing the transmission and interaction of information. In addition, we introduce the attention mechanism behind the information transmission module to give the corresponding attention weight to each temporal image feature so as to enhance the change information of the image, which noticeably improves final prediction. The proposed network is validated on real remote sensing image data sets. Both visual and quantitative analyses of the experimental results demonstrate competitiveness of the proposed method.INDEX TERMS Remote sensing image, change detection, deep neural network, information transmission, attention mechanism.

show abstract

Learning a transferable change detection method by Recurrent Neural Network

Cited by 12 publications

References 5 publications

Difference Enhancement and Spatial–Spectral Nonlocal Network for Change Detection in VHR Remote Sensing Images

Difference Enhancement and Spatial–Spectral Nonlocal Network for Change Detection in VHR Remote Sensing Images

FODA: Building Change Detection in High-Resolution Remote Sensing Images Based on Feature–Output Space Dual-Alignment

Remote Sensing Image Change Detection Based on Information Transmission and Attention Mechanism

Contact Info

Product

Resources

About