Atmospheric fine particles (PM2.5) poses a significant threat to human health and several studies have been conducted to predict the PM2.5concentrations in time series at monitoring stations. Gridded PM2.5concentration prediction is relatively scarce. In this study, we evaluated representative ML-Based framework in daily 1 km gridded PM2.5prediction to identify suitable prediction methods. To achieve this, random forest, convolutional neural network, and transformer were selected for evaluation. The results showed that the transformers model slightly outperformed the other two models in both qualitative and quantitative evaluations. However, all three models showed poor performance during winter, as the PM2.5concentration is inherently higher and more variable. It is precisely such situation that people are most concerned about. Our research suggests that resolving "class imbalance" and "concept drift" are two potential approaches to address this issue.

Deep learning-based semantic segmentation has made great progress in understanding very-high-resolution (VHR) remote sensing images (RSIs). However, large-scale applications are still limited. The main reason is that diverse imaging modes and geographical differences make it difficult to transfer a model trained in the source domain to the target domain. To solve this problem, unsupervised domain adaptation (UDA) for VHR RSIs has received some attention, but the accuracy of cross-domain semantic segmentation still needs to be improved. Currently, one reasonable proposal for improving accuracy is to take a close look at the category-level information. In this paper, we reveal an integer programming mechanism for modeling the category-level relationship between the source and target domains. The mechanism is based on the solution of the assignment problem, and thus, the proposed method is called category-level assignment for UDA (ClA-UDA). In ClA-UDA, a category-level assignment problem with additional constraints is defined for UDA tasks, and the solution is provided. Based on the solution, an assignment-based image-to-image transferring algorithm (AIT) is first proposed to transfer the source-domain images based on the style of the target-domain images. AIT minimizes a weighted discrepancy, and provides an analytical solution for the transfer. Two assignment-based alignment losses are then introduced to align the source and target domains based on the category-level relationship in a concise way. To validate the performance of ClA-UDA, three VHR remote sensing image datasets are employed, and six UDA tasks are designed. Extensive experiments are conducted, and the results demonstrate the superiority of ClA-UDA compared to the existing methods.

It is an important task to automatically and accurately map rooftops from very high resolution remote sensing images since buildings are very closely related to human activity. Two typical technologies are often utilized to accomplish the task, i.e., semantic segmentation and instance segmentation. The semantic segmentation is to independently allocate a label (e.g., “building” or not) to each pixel, resulting in blob-like segments. On the contrary, one might model the boundary of a rooftop as a polygon to improve the shape of the rooftop by encouraging vertices of polygon to adhere to the rooftop’s boundary. Following this line of work, we present a multitask learning approach to predict rooftop corners in a sequent way using the attention learned from where the boundaries are in a given image region. The approach simulates the process of manual delineation of rooftops’ outline in a given image, which can produce accurate boundaries of rooftops with sharp corners and straight lines between them. Specifically, the proposed method consists of three components, i.e., object detection, pixel-by-pixel classification of both edges and corners, and delineation of rooftops in a sequent manner using a convolutional recurrent neural network (RNN). It is called as object-oriented, edges and corners (OEC)-RNN in this article. Three image datasets of buildings are employed to validate the performance of the OEC-RNN, which are compared with state-of-the-art methods for instance segmentation. The experimental results show that the OEC-RNN achieves the best performance in terms of overlay, boundary adherence, and vertex location between ground-truth and predicted polygons.

Recently, several recurrent neural network (RNN)-based models have been proposed to delineate the outlines of buildings from very high resolution (VHR) remote sensing images. These models first use convolutional neural networks (CNNs) to recognize the boundary fragments by learning probability maps of both edges and corners and then feed them into RNN to find and link a set of sequent corners into external boundaries of buildings. However, caused by the category imbalance of edges and corners, the local ambiguity of edge detection is very serious, which significantly affects the accuracy of predicted outline corners. To tackle this challenge, this article introduces a convolutional RNN embedded with line segment information (LSI-RNN), a novel network that aims to directly detect line segment instead of edges. To achieve this, LSI-RNN utilizes an additional cotraining branch to generate an attraction field map (AFM) by neural discriminative dimensionality reduction (NDDR) layer. Consequently, the conventional classification problem of edges is converted to a regression problem of line segments, thus solving the aforementioned issues. Experimental results over three remote sensing datasets with different spatial resolutions show that the proposed method consistently outperforms other state-of-the-art methods.

Many researchers focus on the problem of accurately and rapidly classifying regional landslide hazard, whose spectrum and shape are complicated and varied in remote sensing images. A large number of training examples with labels are necessary to construct predictive models in supervised classification, which are difficult to get strong supervision information due to the high cost of data labeling process for rapid regional landslides identification. Our methods use pre- and post-event MODIS NDVI products, and post-event SPOT-5 images to classify the landslide image during the 2008 Wenchuan Earthquake based on semi-supervised learning model, which means that only a subset of training data are given with labels to train a good learner. To examine the effectiveness of the proposed method, the results are compared with state-of-the-art support vector machine (SVM). Experimental results demonstrate that the proposed method is an accurate and rapid way to classify landslide images.

Over-segmentation has often been employed to simplify representation of images and speed the process of image analysis, where each over-segment is also called a superpixel. However, the superpixels generated by the existing algorithms rely heavily on some preset parameters, for example, number of superpixels. In this paper, we present a novel superpixel algorithm under the framework of nonparametric Bayesian image clustering, which is called Nonparametric Bayesian Iterative Clustering (NBIC). Unlike traditional approaches for image clustering, Bayesian nonparametric models provide a principled way to infer the number of clusters from observed data. Therefore, unlike traditional approach to image over-segmentation, the NBIC is free of preset number of superpixels. The NBIC to superpixel segmentation is based on histogram clustering, and local histograms are extracted as observation. For each observation, compute the probability of each class and sample a class label accordingly. The cluster probabilities depend on the number of classes in the pixel neighborhood and the distance of cluster center. This method groups pixels into meaningful regions without setting any parameters related to superpixels in the nonparametric Bayesian theory. The adaptive amount of superpixels can eventually be converged, and the size of superpixels can automatically be controlled by iterative calculation

Water is an indispensable part of urban development and ecological environment protection. Xiong'an New Area is a state-level new areas established to adjust and optimize the urban layout and spatial structure of Beijing-Tianjin-Hebei. Exploring the water changes of the Xiong'an New Area is critical to local development. In this paper, the Spectrum Matching based on Discrete Particle Swarm Optimization (SMDPSO) method—an effective approach for complex water extraction was used to extract the waters of spring in 1984, 1989, 1993, 2000, 2008, 2014 and 2018 and combined with the water extract results of 28 years to analyze the water degraded and restored areas of the Xiong'an New Area. The largest water area (280 km2) in Xiong'an New Area appeared in 1989. Water degraded areas is distributed in strong change zone and strong change zone and water restored areas is distributed in weak change zone and micro-change zone.

There are many approaches to fuse panchromatic (PAN) and multispectral (MS) images for classification, mainly including sharpening-then-classification methods, classification-then-sharpening methods, and segmentation-then-classification methods. The generalized Chinese restaurant franchise (gCRF) is a segmentation-then-classification-like method to fuse very high resolution (VHR) PAN and MS images for classification, which has the limitation the same as that of the general segmentation-then-classification methods that segmentation errors will affect the subsequent classification. The problems of gCRF are that during the segmentation step, the spatial coherence in the image plane is deficient and the global clusters without spatial position information are used for segmentation, which may lead to undersegmented and disconnected regions in the segmentation results and decrease classification accuracy. In this paper, we propose an improved model, which overcomes the problems of the gCRF during the segmentation step, to increase the classification accuracy by the following two ways: 1) building the spatial coherence in the image plane by introducing neighborhood information of superpixels to construct the subimages and 2) using localized clusters with spatial location information instead of global clusters to measure the similarity between superpixels and segments. The experimental results show that the problems of undersegmentation and disconnected segments are both alleviated, resulting in better classification results in terms of the visual and quantitative aspects.

The Disaster Monitoring and Emergency Response Service(DIMERS) project was kicked off in 2017 in China, with the purpose to improve timely responsive service of the institutions involved in the management of natural disasters and man-made emergency situations with the timely and high-quality products derived from Space-based, Air-based and the in-situ Earth observation. The project team brought together a group of top universities and research institutions in the field of Earth observations as well as the operational institute in typical disaster services at national level. The project will bridge the scientific research and the response services of massive catastrophe in order to improve the emergency response capability of China and provide scientific and technological support for the implementation of the national emergency response strategy.

Stochastic gradient descent and other adaptive optimization methods such as RMSprop, and Adam have been proved effective for training deep neural networks [1], [2]. Within each epoch of these methods, the whole training set is involved. In general, large training datasets have data redundancy. In this paper, we investigate an algorithm that reduce the training time of CNN by dropping certain samples out. Thus, it is called DropSample. This method can be viewed as a special type of truncated cross-entropy loss with a finite margin. We design experiments on several datasets to demonstrate the effects of acceleration. The results show that this method could decrease the training time of multilayer perceptrons (MLPs) and convolutional neural networks (CNNs) significantly. Despite reduced number of training samples, the accuracies of networks are similar, or even better.

The types of remote sensing data are various but each plays a significant role in the ability to analyze the landslides. To detect landslides with satellite images of two different spatial resolutions in a weakly supervised way, this paper presents a probabilistic topic model - maximum entropy discrimination latent Dirichlet allocation regression (MedLDAr) model, which is a more automated approach compared to our previous method - MedLDA model. A two-step algorithm is used to infer the model. First, before- and after-the event MODIS NDVI productions are employed to get the low-resolution vegetation-cover difference map. Second, MedLDAr model is learned by both of the NDVI difference map (i.e., the weakly supervised information) and post-event SPOT-5 images to detect landslides. Experimental results demonstrate that the proposed method is a more promising and automated way to detect landslides in vegetated regions.

Two-step ways are often used for fusing both panchromatic (PAN) and multispectral (MS) images for classification, e.g., classifying MS images sharpened by PAN images or directly pouring fine spatial details of PAN images into a classification result of MS images. In this paper, we present a unified Bayesian framework to iteratively discovering semantic segments from PAN images and allocating cluster labels for the segments using MS images. Specifically, the probabilistic generative process of both PAN and MS images is explained with a generalized metaphor of the Chinese restaurant franchise (CRF) (gCRF), in which the two iterative random processes, i.e., table selection and dish selection, are adapted to discovering semantic segments in PAN images and inferring cluster labels for the discovered segments using MS images, respectively. Our major contributions are twofold: 1) The CRF is generalized into an image fusion framework by elegantly decomposing its two random processes, and 2) the random process of table selection in the CRF is transformed into stochastic image segmentation by enforcing spatial constraints over adjacent pixels. The qualitative analysis of experimental results shows that the gCRF can effectively utilize both the spatial details of the PAN images and the spectral information of the MS images. In terms of quantitative evaluation, the gCRF is comparable with support vector machine-based supervised classification methods.

As the most commonly techniques to landslide inventory mapping, visual interpretation and geomorphological field surveys are time-consuming and labor-intensive. In this paper, a probabilistic topic model, maximum entropy discrimination latent Dirichlet allocation (MedLDA), is presented to detect landslides with satellite images of two different spatial resolutions in a weakly supervised way. A two-stage algorithm is inferred the model. First, before- and after- the event MODIS NDVI productions are employed to roughly locate probable landslides, i.e., low-resolution vegetation-cover changes. Second, MedLDA model is learned by both NDVI change values (i.e., the weakly supervised information) and post-event SPOT 5 images to detect the landslide. Experimental results demonstrate that the proposed method is a very promising way to detect landslides in vegetated regions.

Probabilistic topic models have has successfully been used to classify remote sensing images in unsupervised way. However, the relationship among pixels is ignored in these applications because of the assuption of “bag of words”. This assuption leads to “pepper and salt effect” when these models are used to classify Very High Resolution (VHR) remote sensing images. To solve this problem, a novel model name guided Chinese Restaurant Franchise is proposed by combining the traditional Chinese Restaurant Franchise and guided information which is used to describe the relationship among pixels. Gibbs sampling method is used to infer the proposed model. The efficiency of parameters of the guided information on the result is analyzed. and then the result of our model is compared with other models. The results indicate that the proposed algorithm outperforms the other comparing models in our experiment.

In this paper we introduce an object-based change detection model using correlation analysis and classification. First we use eCognition to obtain an over-segmentation map. Then linear regression is used to gain three unique types of parameters - regression coefficient, offset, and correlation coefficient which can provide valuable information about the location and numeric change value derived within the segmentation objects in the two data sets. Understandably, the two data tend to be highly correlated when little change occurs, and uncorrelated when change occurs. Then we treat the three variables as a three-band image. Finally, we perform maximum likelihood classification with training examples. The result shows that our method performs better than the methods proposed by Yang [2] and J. Im [3]. The advantages of our method are that it performs automatically without selecting threshold empirically and alleviates the “salt and pepper” effect.

Many methods have been developed to detect damaged buildings due to earthquake. However, little attention has been paid to analyze slightly affected buildings. In this letter, an unsupervised method is presented to detect earthquake-triggered “roof-holes” on rural houses from unmanned aerial vehicle (UAV) images. First, both orthomosaic and gradient images are generated from a set of UAV images. Then, a modified Chinese restaurant franchise model is used to learn an unsupervised model of the geo-object classes in the area by fusing both oversegmented orthomosaic and gradient images. Finally, “roof-holes” on rural houses are detected using the learned model. The performance of the proposed method is evaluated in terms of both qualitative and quantitative indexes.

We present a nonparametric Bayesian hierarchical model (HDP_IBPs) to classify very high resolution panchromatic satellite images in an unsupervised way, in which the hierarchical Dirichlet process (HDP) and Indian buffet process (IBP) are combined on multiple scenes. The main contribution of this paper is a novel application framework to solve the problems of traditional probabilistic topic models and achieve the effective unsupervised classification of very high resolution (VHR) panchromatic satellite images. In this framework, a VHR satellite image is first oversegmented into basic processing units and divided into a set of subimages. We use the Chinese restaurant franchise process as a construct method of the HDP to capture the latent semantic structures (i.e., classes) and the class proportion (i.e., co-occurrence of topics) for each subimage. Meanwhile, the subimages are grouped into different scenes based on the class proportion. Finally, the IBP is employed to select the most appropriate classes for each subimage from all of the classes based on different scenes in turn. The hierarchical structure of our model transmits the spatial information from the original image to the scene layer implicitly and provides useful cues of classification. The experimental results show that HDP_IBPs outperforms state-of-the-art models in terms of both qualitative and quantitative evaluations.

This study utilizes the unmanned aerial vehicle (UAV) to acquire high resolution images for feature matching, resulting in a point cloud. A progressive morphological filter is used to filter out nonground object points from point cloud. Multi-scale and different shape filter windows are adopted for the morphological filter to achieve good performance. The results show that multi-scale and multi-shape window can improve the performance of morphological filter compared with single direction filter window, as nonground objects cannot be completely removed with single direction filter window. With multi-shape or 2-D filter window, buildings can be effectively removed and ground points can be reserved.

Typically, object-based classification methods are learned using training samples with labels attached to image objects. In this letter, a semisupervised object-based method in the framework of topic modeling is proposed to classify very high resolution panchromatic satellite images using partially labeled pixels. In the stage of training, both topics and their co-occurred distributions are learned in an unsupervised manner from segmented satellite images. Meanwhile, unlabeled pixels are allocated user-provided geo-object class labels based on the learned model. In the stage of classification, each segment is classified as a user-provided geo-object class label with the maximum posterior probability. Experimental results show that the proposed method outperforms several SVM-based supervised classification methods in terms of both spatial consistency and semantic consistency.

LDA model has successfully been used to analyzing satellite images. However there are two cucial problems: (1) the number of clusters needs being given in advance, and (2) all documents share a Dirichlet prior. To solve the problems, a novel model include multiple LDAs with variable topics are proposed to cluster satellite images. Each LDA in the model is dedicated to model one kind of natural scene in satellite images. Gibbs sampling method is used to discover natural scenes and learning model parameters. The effect on number of topic estimation is analyzed and then the result of our model is compared with other models. The results indicate that the proposed algorithm outperforms the other comparing models in our experiment.

A novel model is presented to address the problem of semantic clustering of geo-objects in very high resolution panchromatic satellite images. The proposed model combines a probabilistic topic model with a multiscale image representation into an automatic framework by embedding both document and scale selections. The probabilistic topic model is used to characterize the statistical distributions of both intraclass appearance and inter-class coherence of geo-objects within documents, i.e., squared sub-images. Because the bag-of-words assumption involved in the probabilistic topic models does not consider the spatial coherence between topic labels, the multiscale image representation is designed to provide a self-adaptive spatial regularization for various geo-object categories. By introducing scale and document selections, the automatic framework integrates the probabilistic topic model and the multiscale image representation to ensure that words on a site should be allocated the same topic label no matter what documents they reside in. Consequently, unlike the traditional method of applying topic models for analyzing satellite images, the process of explicitly generating a set of documents before modeling and then combining multiple labels for a word on a given site is unnecessary. Gibbs sampling is adopted for parameter estimation and image clustering. Extensive experimental evaluations are designed to first analyze the effect of parameters in the proposed model and then compare the results of our model with those of some state-of-the-art methods for three different types of images. The results indicate that the proposed algorithm consistently outperforms these exiting state-of-the-art methods in all of the experiments.

In this paper, we present a novel object-oriented semantic clustering algorithm for VHR panchromatic satellite images using a variant of latent Dirichlet allocation model. Firstly, an image collection is implicitly generated by partitioning a large satellite image into densely overlapped sub-images. Then, the Latent Dirichlet Allocation with a hierarchy Dirichlet process is employed to model the image collection. Gibbs sampling is adopted for parameter estimation and image clustering. Specifically, the introduction of Dirichlet process is purposed to extend the LDA to an infinite mixtures model which can estimate the number of components (e.g. clusters in image analysis) automatically. Finally, the effect of the proposed algorithm is analyzed through experiments, and the results of it with the traditional K-means method over a QUICKBIRD image are compared.

In this letter, we present a novel object-oriented semantic clustering algorithm for high-spatial-resolution remote sensing images using the probabilistic latent semantic analysis (PLSA) model coupled with neighborhood spatial information. First of all, an image collection is generated by partitioning a large satellite image into densely overlapped subimages. Then, the PLSA model is employed to model the image collection. Specifically, the image collection is partitioned into two subsets. One is used to learn topic models, where the number of topics is determined using a minimum description length criterion. The other is folded in using the learned topic models. Therefore, every pixel in each subimage has been allocated a topic label. At last, the cluster label of every pixel in the large satellite image is derived from the topic labels of multiple subimages which cover the pixel in the image collection. Experimental results over a QUICKBIRD image show that the clusters of the proposed algorithm are better thanK-means and Iterative Self-Organizing Data Analysis Technique Algorithm in terms of object-oriented property.

We introduce a semi-automated algorithm to extract urban road from high-resolution RS image using the Probabilistic Topic Models. First of all, an image collection is generated from a high-resolution image by partitioning it into densely overlapped sub-images. The image collection is divided into two subsets, i.e., training images and testing images. The training images are used to estimate the number of topics, and to learn topic models. The training images are densely overlapped and are folded in using the learned topics to make sure that every pixel in each document is allocated to a topic label. Therefore, every pixel in the initial large image might be allocated multiple topic labels since it might belong to multiple sub-images. By selecting the road segments samples, several cluster centers will be assumed as labels of road objects. The semantic information can improve the extraction accuracy of road segments. The central lines of the road segments will be extracted basing on some image filter algorithms and Hough transform. Experimental results over EROS-B images show that road segments can be effectively detected by the proposed algorithm and an initial road network can be formed.

Some basic issues on content-based remote sensing image retrieval are discussed in this paper. The framework, processing flow and levels are proposed based on theory of CBIR and characteristics of RS image. Oriented to the practical demands, five retrieval patterns including template-based, attribute-based, metadata-based, semanteme-based and integrated retrieval are proposed. The contents and features that can be used in content-based remote sensing image retrieval include color, shape, texture, spectra, spatial relation, metadata and relative rules and knowledge. Among those features, spectral features, spatial features and metadata are the main aspects of RS image differing from common images.