“…The current cloudy image, a mask map of clouds and cloud shadows, and a cloudless recent image of the same area were prepared in advance as the input. The mask map, which was a binary map where 1 represented the clean pixels (black), and 0 represents the clouds and cloud shadows (white), can be generated from recent CNN-based detection methods [43][44][45] or manual work. When an automatic algorithm was applied, the recall rate should be set high to detect most of the clouds; however, a relatively lower precision score will not affect the performance of the cloud removal task as a large number of clean pixels remain to train the cloud removal network.…”
Pixels of clouds and cloud shadows in a remote sensing image impact image quality, image interpretation, and subsequent applications. In this paper, we propose a novel cloud removal method based on deep learning that automatically reconstructs the invalid pixels with the auxiliary information from multi-temporal images. Our method’s innovation lies in its feature extraction and loss functions, which reside in a novel gated convolutional network (GCN) instead of a series of common convolutions. It takes the current cloudy image, a recent cloudless image, and the mask of clouds as input, without any requirements of external training samples, to realize a self-training process with clean pixels in the bi-temporal images as natural training samples. In our feature extraction, gated convolutional layers, for the first time, are introduced to discriminate cloudy pixels from clean pixels, which make up for a common convolution layer’s lack of the ability to discriminate. Our multi-level constrained joint loss function, which consists of an image-level loss, a feature-level loss, and a total variation loss, can achieve local and global consistency both in shallow and deep levels of features. The total variation loss is introduced into the deep-learning-based cloud removal task for the first time to eliminate the color and texture discontinuity around cloud outlines needing repair. On the WHU cloud dataset with diverse land cover scenes and different imaging conditions, our experimental results demonstrated that our method consistently reconstructed the cloud and cloud shadow pixels in various remote sensing images and outperformed several mainstream deep-learning-based methods and a conventional method for every indicator by a large margin.
“…The current cloudy image, a mask map of clouds and cloud shadows, and a cloudless recent image of the same area were prepared in advance as the input. The mask map, which was a binary map where 1 represented the clean pixels (black), and 0 represents the clouds and cloud shadows (white), can be generated from recent CNN-based detection methods [43][44][45] or manual work. When an automatic algorithm was applied, the recall rate should be set high to detect most of the clouds; however, a relatively lower precision score will not affect the performance of the cloud removal task as a large number of clean pixels remain to train the cloud removal network.…”
Pixels of clouds and cloud shadows in a remote sensing image impact image quality, image interpretation, and subsequent applications. In this paper, we propose a novel cloud removal method based on deep learning that automatically reconstructs the invalid pixels with the auxiliary information from multi-temporal images. Our method’s innovation lies in its feature extraction and loss functions, which reside in a novel gated convolutional network (GCN) instead of a series of common convolutions. It takes the current cloudy image, a recent cloudless image, and the mask of clouds as input, without any requirements of external training samples, to realize a self-training process with clean pixels in the bi-temporal images as natural training samples. In our feature extraction, gated convolutional layers, for the first time, are introduced to discriminate cloudy pixels from clean pixels, which make up for a common convolution layer’s lack of the ability to discriminate. Our multi-level constrained joint loss function, which consists of an image-level loss, a feature-level loss, and a total variation loss, can achieve local and global consistency both in shallow and deep levels of features. The total variation loss is introduced into the deep-learning-based cloud removal task for the first time to eliminate the color and texture discontinuity around cloud outlines needing repair. On the WHU cloud dataset with diverse land cover scenes and different imaging conditions, our experimental results demonstrated that our method consistently reconstructed the cloud and cloud shadow pixels in various remote sensing images and outperformed several mainstream deep-learning-based methods and a conventional method for every indicator by a large margin.
“…Thus, it is unreasonable to select the training samples using fixed scale. In this study, we used the super-pixel segments to construct samples database based on image interpretation at different scale (20 × 20, 40 × 40 and 60 × 60) (Shao et al 2019). In this way we not only improved the efficiency of collecting training samples, but greatly increased the number and type of samples, which contributed to improve the classification accuracy.…”
Section: Collection Of Multiscale Damage Samplesmentioning
Abstract. Accurate detection and automatic processing of earthquake-damaged regions is essential for effective rescue and post-disaster reconstruction. In this study, we proposed a Combined Super-pixel Segmentation and AlexNet Detection approach (CSSAD) for automatically extracting damaged regions from post-earthquake high-resolution images. Simple Linear Iterative Clustering (SLIC) algorithm was used to segment the high resolution images to obtain more homogeneous geo-objects. Multiscale samples database, which took the different scale effect of damaged regions into account, was constructed based on the geometric centre of each super-pixel. AlexNet, which achieved the automatic extraction of high-level features and accurate identification of target geo-objects, was used to detect the damaged regions. To enhance the localization accuracy, the output of AlexNet was further refined using super-pixel segmentations and masked out of shadow and vegetation. Compared with traditional method, the proposed approach effectively reduces the false and missed detection ratio at least 10 percent.
“…Most studies that estimate AGB at a small scale use medium and high spatial resolution remote sensing data, such as Landsat TM [27], SPOT [28], [29] or Quickbird [30]; however, these data have low temporal resolution. Remote sensing data are easily affected by meteorological factors such as clouds [31], [32]. Processing the data requires significant computing power, and data acquisition is expensive, hence, the AGB estimation at a large scale is not a straightforward task.…”
Accurate estimation of aboveground forest biomass (AGB) at a large scale is important in global carbon cycle, forest productivity, and climate change. Coarse resolution remote sensing data of long time series are often used to estimate large scale AGB, but the result is inaccurate due to the scaling effect caused by nonlinearity in data representation and the existence of mixed pixels containing different forest types and land uses. Improvement in the accuracy of AGB estimated from coarse resolution remote sensing data is urgently needed. Research on spatial scaling of AGB is still lacking, therefore, this paper proposed an approach based on structural analysis of mixed pixels and the Random Forest model (SMPRF) to increase the accuracy of AGB estimated from coarse resolution data. MODIS and SPOT 5 data were used to create forest biomass distribution maps of the study area at two scales. The scaling effect on estimating forest biomass based on remote sensing was analyzed by comparing data from these two datasets. SMPRF, which included a correction factor for the scaling effect on AGB estimated from coarse resolution MODIS data, was used to create a model that scaled from the fine resolution data (SPOT 5) to the coarse resolution data (MODIS). The results showed that the accuracy of AGB estimated from MODIS data was increased using this method. The Pearson correlation coefficient (r) for data verification increased from 0.63 to 0.89 and the root mean squared error decreased from 51.6 Mg•ha-1 to 26.8 Mg•ha-1. The difference tests showed that the changes were extremely significant (p = 0). Thus, SMPRF can significantly improve the accuracy of large scale AGB estimation based on coarse resolution remote sensing data and the feasibility of applying the method proposed in this study to related fields is verified.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.