NestNet: a multiscale convolutional neural network for remote sensing image change detection

Yu, Xiao; Fan, Jiangwen; Zhang, Peng; Zhou, Yuke; Han, Liusheng

doi:10.1080/01431161.2021.1906982

Cited by 32 publications

(20 citation statements)

References 48 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…An operational limitation is generally due to the need for large annotated ground truth to be used for training purposes. Examples of state-of-the-art solutions based on CNNs are [62][63][64][65].…”

Section: Previous Work On Land Cover Change Detectionmentioning

confidence: 99%

Monitoring the Recovery after 2016 Hurricane Matthew in Haiti via Markovian Multitemporal Region-Based Modeling

et al. 2021

View full text Add to dashboard Cite

The aim of this paper is to address the monitoring of the recovery phase in the aftermath of Hurricane Matthew (28 September–10 October 2016) in the town of Jérémie, southwestern Haiti. This is accomplished via a novel change detection method that has been formulated, in a data fusion perspective, in terms of multitemporal supervised classification. The availability of very high resolution images provided by last-generation satellite synthetic aperture radar (SAR) and optical sensors makes this analysis promising from an application perspective and simultaneously challenging from a processing viewpoint. Indeed, pursuing such a goal requires the development of novel methodologies able to exploit the large amount of detailed information provided by this type of data. To take advantage of the temporal and spatial information associated with such images, the proposed method integrates multisensor, multisource, and contextual information. Markov random field modeling is adopted here to integrate the spatial context and the temporal correlation associated with images acquired at different dates. Moreover, the adoption of a region-based approach allows for the characterization of the geometrical structures in the images through multiple segmentation maps at different scales and times. The performances of the proposed approach are evaluated on multisensor pairs of COSMO-SkyMed SAR and Pléiades optical images acquired over Jérémie, in the aftermath of and during the three years after Hurricane Matthew. The effectiveness of the change detection results is analyzed both quantitatively, through the computation of accuracy measures on a test set, and qualitatively, by visual inspection of the classification maps. The robustness of the proposed method with respect to different algorithmic choices is also assessed, and the detected changes are discussed in relation to the recovery endeavors in the area and ground-truth data collected in the field in April 2019.

show abstract

Section: Previous Work On Land Cover Change Detectionmentioning

confidence: 99%

Monitoring the Recovery after 2016 Hurricane Matthew in Haiti via Markovian Multitemporal Region-Based Modeling

et al. 2021

View full text Add to dashboard Cite

show abstract

“…Daudt et al first proposed end-to-end fully convolutional Siamese networks for change detection, named Fully Convolutional Siamese-Concatenation (FC-Siam-conc) and Fully Convolutional Siamese-Difference (FC-Siam-diff) [33]. Thereafter, dual-task constrained deep Siamese convolutional network (DTCDSCN) [39], pyramid feature-based attention-guided Siamese network (PGA-SiamNet) [40], Siamese NestedUNet Networks [41], NestNet [20] and others have been proposed, improve the accuracy of change detection.…”

Section: Introductionmentioning

confidence: 99%

MSF-Net: A Multiscale Supervised Fusion Network for Building Change Detection in High-Resolution Remote Sensing Images

et al. 2022

Self Cite

View full text Add to dashboard Cite

Building change detection is a primary task in the application of remote sensing images, especially in city land resource management and urbanization process assesment. Due to the rich textural features of remote sensing images and the multiscale characteristics of buildings, it is still a huge challenge to effectively filter out irrelevant change information (e.g., roads) and fuse multiscale building features. To date, deep learning-based methods have demonstrated powerful capabilities in this field. To fill these gaps, this study proposes a multiscale supervised fusion network (MSF-Net), which is an attention mechanismbased approach for building change detection using bi-temporal high-resolution satellite imagery. Especially, we built a dual-context fusion module to obtain abundant global context information of buildings and suppressing irrelevant features. We also used channel attention mechanism, selective kernel convolution and multiscale supervision module to fuse multiscale feature of buildings. The ablation experiments verified the availability of these modules. The MSF-Net model has been tested on the LEVIR-CD building change detection dataset. Compared with other state-of-the-art change detection methods, the study showed that our method obtained 0.8866 and 0.8130 in F1-score and Intersection over Union (IOU), respectively. The results indicate that the MSF-Net method has stronger multiscale building feature extraction capability and suppression ability of irrelevant features, which could produce clearer building boundaries and more accurate building change maps.

show abstract

“…Zhang et al [38] proposed DifUnet++, which emphasizes the explicit representation of difference features using a differential pyramid of bi-temporal images. Yu et al [39] implemented the Nest-Net based on the UNet++. NestNet promotes the explicit difference representation using absolute differential operation (ADO).…”

Section: Introductionmentioning

confidence: 99%

“…During model training, multistage prediction and deep supervision have been proven effective strategies for achieving better performance. For instance, some attempts apply the multistage prediction strategy at the decoder's output side, such as Peng et al [27], DifUnet++ [38], NestNet [39], IFN [30], HDFNet [34], and ADS-Net [28]. The overall loss function is calculated based on the weighted sum of multistage prediction's loss.…”

Section: Introductionmentioning

confidence: 99%

An Efficient Lightweight Neural Network for Remote Sensing Image Change Detection

2021

View full text Add to dashboard Cite

Remote sensing (RS) image change detection (CD) is a critical technique of detecting land surface changes in earth observation. Deep learning (DL)-based approaches have gained popularity and have made remarkable progress in change detection. The recent advances in DL-based methods mainly focus on enhancing the feature representation ability for performance improvement. However, deeper networks incorporated with attention-based or multiscale context-based modules involve a large number of network parameters and require more inference time. In this paper, we first proposed an effective network called 3M-CDNet that requires about 3.12 M parameters for accuracy improvement. Furthermore, a lightweight variant called 1M-CDNet, which only requires about 1.26 M parameters, was proposed for computation efficiency with the limitation of computing power. 3M-CDNet and 1M-CDNet have the same backbone network architecture but different classifiers. Specifically, the application of deformable convolutions (DConv) in the lightweight backbone made the model gain a good geometric transformation modeling capacity for change detection. The two-level feature fusion strategy was applied to improve the feature representation. In addition, the classifier that has a plain design to facilitate the inference speed applied dropout regularization to improve generalization ability. Online data augmentation (DA) was also applied to alleviate overfitting during model training. Extensive experiments have been conducted on several public datasets for performance evaluation. Ablation studies have proved the effectiveness of the core components. Experiment results demonstrate that the proposed networks achieved performance improvements compared with the state-of-the-art methods. Specifically, 3M-CDNet achieved the best F1-score on two datasets, i.e., LEVIR-CD (0.9161) and Season-Varying (0.9749). Compared with existing methods, 1M-CDNet achieved a higher F1-score, i.e., LEVIR-CD (0.9118) and Season-Varying (0.9680). In addition, the runtime of 1M-CDNet is superior to most, which exhibits a better trade-off between accuracy and efficiency.

show abstract

NestNet: a multiscale convolutional neural network for remote sensing image change detection

Cited by 32 publications

References 48 publications

Monitoring the Recovery after 2016 Hurricane Matthew in Haiti via Markovian Multitemporal Region-Based Modeling

Monitoring the Recovery after 2016 Hurricane Matthew in Haiti via Markovian Multitemporal Region-Based Modeling

MSF-Net: A Multiscale Supervised Fusion Network for Building Change Detection in High-Resolution Remote Sensing Images

An Efficient Lightweight Neural Network for Remote Sensing Image Change Detection

Contact Info

Product

Resources

About