Search

With the rapid development of the energy industry and the growth of the global energy demand in recent years, the development of the photovoltaic industry has become increasingly significant. However, the development of the PV industry is constrained by high land costs, and land in central cities and industrial areas is often very expensive and unsuitable for the installation of PV equipment in large areas. With this background knowledge, the key to evaluating the PV potential is by counting the rooftop information of buildings, and an ideal solution for extracting building rooftop information is from remote sensing satellite images using the deep learning method; however, the deep learning method often requires large-scale labeled samples, and the labeling of remote sensing images is often time-consuming and expensive. To reduce the burden of data labeling, models trained on large datasets can be used as pre-trained models (e.g., ImageNet) to provide prior knowledge for training. However, most of the existing pre-trained model parameters are not suitable for direct transfer to remote sensing tasks. In this paper, we design a pseudo-label-guided self-supervised learning (PGSSL) semantic segmentation network structure based on high-resolution remote sensing images to extract building information. The pseudo-label-guided learning method allows the feature results extracted by the pretext task to be more applicable to the target task and ultimately improves segmentation accuracy. Our proposed method achieves better results than current contrastive learning methods in most experiments and uses only about 20–50% of the labeled data to achieve comparable performance with random initialization. In addition, a more accurate statistical method for building density distribution is designed based on the semantic segmentation results. This method addresses the last step of the extraction results oriented to the PV potential assessment, and this paper is validated in Beijing, China, to demonstrate the effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]

Short-term origin–destination (termed as OD) prediction is crucial to improve the operation of urban rail transit (termed as URT). The latest research results show that deep learning can effectively improve the performance of short-term OD prediction and meet the real-time requirements. However, many advanced neural network design ideas have not been fully applied in the field of short-term OD prediction in URT. In this paper, a novel parallel interactive attention network (termed as PIANet) for short-term OD prediction in URT is proposed to further improve the short-term OD prediction accuracy. In the proposed PIANet, a novel omnidirectional attention module (termed as OAM) is proposed to improve the representational power of the network by calculating the feature weights in the channel–spatial dimension. Moreover, a simple yet effective feature interaction is proposed to improve the feature utilization. Based on the two real-world datasets from the Beijing subway, the comparative experiments demonstrate that the proposed PIANet outperforms the state-of-the-art deep learning methods for short-term OD prediction in URT, and the ablation studies demonstrate that the proposed OAMs and feature interaction play an important role in improving the short-term OD prediction accuracy. [ABSTRACT FROM AUTHOR]

Accurate air quality prediction is of great significance for pollution prevention and disaster prevention. Effective and reliable prediction models are needed not only for short time prediction, but are even more important for long time-series future predictions. In the long time series, most of the current models might not function as accurately as in the short period and thus a new model is required. In this paper, the new PM2.5 predictor is proposed to achieve accurate long time series PM2.5 prediction in Beijing. The predictor simplifies the input parameters through Spearman correlation analysis and implements the long time series prediction through Informer. The results show that AQI, CO, NO2, and PM10 concentrations are selected from the air quality data, and Dew Point Temperature (DEWP) and wind speed are incorporated from two meteorological data to better improve the prediction efficiency by almost 27%. By comparing with LSTM and attention-LSTM models, the model constructed in this paper performs well in different prediction time periods, with at least 21%, 19%, 28%, and 35% improvement in accuracy in four prediction time series: 48 h, 7 days, 14 days, and 30 days. In conclusion, the proposed model is proved to solve the problem of predicting long time series PM2.5 concentrations in the future, which can make up for the shortcomings of the currently existing models and have good application value. [ABSTRACT FROM AUTHOR]

The effective refuge area is a key indicator in the study of emergency shelters. Accurately extracting the effective refuge area and analyzing the effectiveness of emergency shelters are of great significance for site selection, spatial distribution, and the evaluation of suitability. Beijing is one of only three capitals in the world located in a high-seismic-intensity zone of magnitude 8. The fast and accurate monitoring of effective refuge areas and an analysis of the effectiveness of emergency shelters are conducive to evacuation planning and disaster prevention and mitigation, and they promote the construction of a resilient city. However, the extraction of effective refuge areas in existing studies is not only a time-consuming and labor-intensive task but also has accuracy and efficiency problems, resulting in less precise validity analyses. In this paper, a remote sensing monitoring technology system for the effective refuge areas of emergency shelters is proposed based on multi-source data. Different methods were used to extract various land features, such as buildings and collapsed areas, water, dense areas of understory vegetation, and steep slope areas that cannot be evacuated, to obtain the effective refuge area at a detailed scale, in combination with the service radius of emergency shelters, the population distribution, and the actual road network, the criteria for effectiveness analysis were established for the effective open space ratio, capacity, per capita accessible effective refuge area, and population allocation gap. Taking the area within the Fifth Ring Road of Beijing as an example, the effectiveness of emergency shelters was analyzed at both the whole scale and a local scale. The results show that the effective refuge areas of different emergency shelters in Beijing vary significantly, with the smallest effective refuge area being located in Rings 2–3 and the largest one being located in Rings 4–5; between different regions, there are differences in the effectiveness. This study provides a feasible method for the fast, accurate, and detailed extraction of the effective refuge areas of emergency shelters and also provides a reference for emergency planning for disaster prevention and mitigation. [ABSTRACT FROM AUTHOR]

The profound impacts of severe air pollution on human health, ecological balance, and economic stability are undeniable. Precise air quality forecasting stands as a crucial necessity, enabling governmental bodies and vulnerable communities to proactively take essential measures to reduce exposure to detrimental pollutants. Previous research has primarily focused on predicting air quality using only time-series data. However, the importance of remote-sensing image data has received limited attention. This paper proposes a new multi-modal deep-learning model, Res-GCN, which integrates high spatial resolution remote-sensing images and time-series air quality data from multiple stations to forecast future air quality. Res-GCN employs two deep-learning networks, one utilizing the residual network to extract hidden visual information from remote-sensing images, and another using a dynamic spatio-temporal graph convolution network to capture spatio-temporal information from time-series data. By extracting features from two different modalities, improved predictive performance can be achieved. To demonstrate the effectiveness of the proposed model, experiments were conducted on two real-world datasets. The results show that the Res-GCN model effectively extracts multi-modal features, significantly enhancing the accuracy of multi-step predictions. Compared to the best-performing baseline model, the multi-step prediction's mean absolute error, root mean square error, and mean absolute percentage error increased by approximately 6%, 7%, and 7%, respectively. [ABSTRACT FROM AUTHOR]

Air quality plays a vital role in people's health, and air quality forecasting can assist in decision making for government planning and sustainable development. In contrast, it is challenging to multi-step forecast accurately due to its complex and nonlinear caused by both temporal and spatial dimensions. Deep models, with their ability to model strong nonlinearities, have become the primary methods for air quality forecasting. However, because of the lack of mechanism-based analysis, uninterpretability forecasting makes decisions risky, especially when the government makes decisions. This paper proposes an interpretable variational Bayesian deep learning model with information self-screening for PM2.5 forecasting. Firstly, based on factors related to PM2.5 concentration, e.g., temperature, humidity, wind speed, spatial distribution, etc., an interpretable multivariate data screening structure for PM2.5 forecasting was established to catch as much helpful information as possible. Secondly, the self-screening layer was implanted in the deep learning network to optimize the selection of input variables. Further, following implantation of the screening layer, a variational Bayesian gated recurrent unit (GRU) network was constructed to overcome the complex distribution of PM2.5 and achieve accurate multi-step forecasting. The high accuracy of the proposed method is verified by PM2.5 data in Beijing, China, which provides an effective way, with multiple factors for PM2.5 forecasting determined using deep learning technology. [ABSTRACT FROM AUTHOR]

Day by day pollution in cities is increasing due to urbanization. One of the biggest challenges posed by the rapid migration of inhabitants into cities is increased air pollution. Sustainable Development Goal 11 indicates that 99 percent of the world's urban population breathes polluted air. In such a trend of urbanization, predicting the concentrations of pollutants in advance is very important. Predictions of pollutants would help city administrations to take timely measures for ensuring Sustainable Development Goal 11. In data engineering, imputation and the removal of outliers are very important steps prior to forecasting the concentration of air pollutants. For pollution and meteorological data, missing values and outliers are critical problems that need to be addressed. This paper proposes a novel method called multiple iterative imputation using autoencoder-based long short-term memory (MIA-LSTM) which uses iterative imputation using an extra tree regressor as an estimator for the missing values in multivariate data followed by an LSTM autoencoder for the detection and removal of outliers present in the dataset. The preprocessed data were given to a multivariate LSTM for forecasting PM2.5 concentration. This paper also presents the effect of removing outliers and missing values from the dataset as well as the effect of imputing missing values in the process of forecasting the concentrations of air pollutants. The proposed method provides better results for forecasting with a root mean square error (RMSE) value of 9.8883. The obtained results were compared with the traditional gated recurrent unit (GRU), 1D convolutional neural network (CNN), and long short-term memory (LSTM) approaches for a dataset of the Aotizhonhxin area of Beijing in China. Similar results were observed for another two locations in China and one location in India. The results obtained show that imputation and outlier/anomaly removal improve the accuracy of air pollution forecasting. [ABSTRACT FROM AUTHOR]

Geological hazards often occur in mountainous areas and are sudden and hidden, so it is important to identify and assess geological hazards. In this paper, the western mountainous area of Beijing was selected as the study area. We conducted research on landslides, collapses, and unstable slopes in the study area. The surface deformation of the study area was monitored by multi-temporal interferometric synthetic aperture radar (MT-InSAR), using a combination of multi-looking point selection and permanent scatterer (PS) point selection methods. Random forest (RF), support vector machine (SVM), convolutional neural network (CNN), and recurrent neural network (RNN) models were selected for the assessment of geological hazard susceptibility. Sixteen geological hazard-influencing factors were collected, and their information values were calculated using their features. Multicollinearity analysis with the relief-F method was used to calculate the correlation and importance of the factors for factor selection. The results show that the deformation rate along the line-of-sight (LOS) direction is between −44 mm/year and 28 mm/year. A total of 60 geological hazards were identified by combining surface deformation with optical imagery and other data, including 7 collapses, 25 unstable slopes, and 28 landslides. Forty-eight of the identified geological hazards are not recorded in the Beijing geological hazards list. The most effective model in the study area was RF. The percentage of geological hazard susceptibility zoning in the study area is as follows: very low susceptibility 27.40%, low susceptibility 28.06%, moderate susceptibility 21.19%, high susceptibility 13.80%, very high susceptibility 9.57%. [ABSTRACT FROM AUTHOR]

Passenger flow prediction is a critical approach to ensure the effective functioning of urban rail transit. However, there are few studies that combine multiple influencing factors for short-term passenger flow prediction. It is also a challenge to accurately predict passenger flow at all stations in the line at the same time. To overcome the above limitations, a deep learning-based method named ST-RANet is proposed, which consists of three spatio-temporal modules and one external module. The model is capable of predicting inbound and outbound passenger flow for all stations within the network simultaneously. We model the spatio-temporal data in terms of three temporal characteristics, including closeness, period, and trend. For each characteristic, we construct a spatio-temporal module that innovatively integrates the attention mechanisms into the middle of residual units and convolutional neural networks (CNNs) to extract and learn spatio-temporal features. Subsequently, the results of the three modules are integrated using a parameter matrix method, which allows for dynamic aggregation based on data. The integration results are further combined with external factors, such as holidays and meteorological information, to obtain passenger flow prediction values for each station. The proposed model is validated using real data from Beijing Subway, and optimized parameters are applied for 30-min granularity passenger flow predictions. Comparing the performance against 5 baseline models and verifying with data from multiple lines, the results indicate that the proposed ST-RANet model shows the best results. It is demonstrated that the method proposed in this paper has high prediction accuracy and good applicability. [ABSTRACT FROM AUTHOR]

The predictions from time series data can help us sense development trends and make scientific decisions in advance. The commonly used forecasting methods with backpropagation consume a lot of computational resources. The deep echo state network (DeepESN) is an advanced prediction method with a deep neural network structure and training algorithm without backpropagation. In this paper, a Bayesian optimization algorithm (BOA) is proposed to optimize DeepESN to address the problem of increasing parameter scale. Firstly, the DeepESN was studied and constructed as the basic prediction model for the time series data. Secondly, the BOA was reconstructed, based on the DeepESN, for optimal parameter searching. The algorithm is proposed within the framework of the DeepESN. Thirdly, an experiment was conducted to verify the DeepESN with a BOA within three datasets: simulation data generated from computer programs, a real humidity dataset collected from Beijing, and a power load dataset obtained from America. Compared with the models of BP (backpropagation), LSTM (long short-term memory), GRU (gated recurrent unit), and ESN (echo state network), DeepESN obtained optimal results, which were 0.0719, 18.6707, and 764.5281 using RMSE evaluation. While getting better accuracy, the BOA optimization time was only 323.4 s, 563.2 s, and 9854 s for the three datasets. It is more efficient than grid search and grey wolf optimizer. [ABSTRACT FROM AUTHOR]

Air pollution has become a critical factor affecting the health of human beings. Forecasting the trend of air pollutants will be of considerable help to public health, including improving early-warning systems. The article designs a novel hybrid deep learning framework FPHFA (FPHFA is the abbreviation of the title of this paper) for PM2.5 concentration forecasting is proposed, which learns spatially correlated features and long-term dependencies of time series data related to PM2.5. Owing to the complex nonlinear dynamic and spatial features of pollutant data, the FPHFFA model combines multi-channel one-dimensional convolutional neural networks, bi-directional long short-term memory neural networks, and attention mechanisms for the first time. Multi-channel 1D CNNs are applied to capture trend features between some sites and overall spatial characteristics of PM2.5 concentration, Bi LSTMs are used to learn the temporal correlation of PM2.5 concentration, and the attention mechanism is used to focus more effective information at different moments. We carried out experimental evaluations using the Beijing dataset, and the outcomes show that our proposed model can effectively handle PM2.5 concentration prediction with satisfactory accuracy. For the prediction task from 1 to 12 h, our proposed prediction model performs well. The FPHFA also achieves satisfactory results for prediction tasks from 13 to 96 h. [ABSTRACT FROM AUTHOR]

Air pollution is a growing problem and poses a challenge to people's healthy lives. Accurate prediction of air pollutant concentrations is considered the key to air pollution warning and management. In this paper, a novel PM2.5 concentration prediction model, CBAM-CNN-Bi LSTM, is constructed by deep learning techniques based on the principles related to spatial big data. This model consists of the convolutional block attention module (CBAM), the convolutional neural network (CNN), and the bi-directional long short-term memory neural network (Bi LSTM). CBAM is applied to the extraction of feature relationships between pollutant data and meteorological data and assists in deeply obtaining the spatial distribution characteristics of PM2.5 concentrations. As the output layer, Bi LSTM obtains the variation pattern of PM2.5 concentrations from spatial data, overcomes the problem of long-term dependence on PM2.5 concentrations, and achieves the task of accurately forecasting PM2.5 concentrations at multiple sites. Based on real datasets, we perform an experimental evaluation and the results show that, in comparison to other models, CBAM-CNN-Bi LSTM improves the accuracy of PM2.5 concentration prediction. For the prediction tasks from 1 to 12 h, our proposed prediction model performs well. For the 13 to 48 h prediction task, the CBAM-CNN-Bi LSTM also achieves satisfactory results. [ABSTRACT FROM AUTHOR]

In the curling sport, the coefficient of friction between the curling stone and pebbled ice is crucial to predict the motion trajectory. However, the theoretical and experimental investigations on stone–ice friction are limited, mainly due to the limitations of the field measurement techniques and the inadequacy of the experimental data from professional curling rinks. In this paper, on-site measurement of the stone–ice friction coefficient in a prefabricated ice rink for the Beijing Winter Olympics curling event was carried out based on computer vision technology. Firstly, a procedure to determine the location of the curling stone was proposed using YOLO-V3 (You Only Look Once, Version 3) deep neural networks and the CSRT Object tracking algorithm. Video data was recorded during the curling stone throwing experiments, and the friction coefficient was extracted. Furthermore, the influence of the sliding velocity on the friction coefficient was discussed. Comparison with published experimental data and models and verification of the obtained results, using a sensor-based method, were conducted. Results show that the coefficient of friction (ranging from 0.006 to 0.016) decreased with increasing sliding velocity, due to the presence of a liquid-like layer. Our obtained results were consistent with the literature data and the friction model of Lozowski. In addition, the experimental results of the computer vision technique method and the accelerometer sensor method showed remarkable agreement, supporting the accuracy and reliability of our proposed measurement procedure based on deep learning. [ABSTRACT FROM AUTHOR]

House price is closely associated with the development of the national economy and people's daily life. Understanding the spatial distribution characteristics and influencing factors of the house price is of great practical significance. Although a lot of attention has been paid to modeling the house price from structure and location attributes, limited work has considered the impact of visual attributes. Intuitively, a better visual environment may raise the surrounding house price. When aggregating multiple factors that influence house price, the multiscale geographically weighted regression (MGWR) provides a suitable solution. Specifically, the MGWR assigns each factor a bandwidth to model the spatial heterogeneity, e.g., a factor may have different influences at different places. In this paper, we introduce the visual environment factors into the MGWR method. In detail, we extract ten visual elements, e.g., sky, vegetation, road, from the Baidu street view (BSV) images, using a deep learning framework. We further define six visual environment factors to investigate their influence on house price. Based on the data from two representative Chinese cities, i.e., Beijing and Chongqing, we reveal the influence degree and spatial scale difference of six visual indexes on the house price in two cities. Results show that: (1) the influence intensity of our proposed six visual environment factors on the house price in different regions of the city can be identified, and the green view index (GVI) is the most important visual environmental factor; and (2) the influence of these view indexes changes significantly or even reversely depends on different areas. [ABSTRACT FROM AUTHOR]

Ultra-short-term load demand forecasting is significant to the rapid response and real-time dispatching of the power demand side. Considering too many random factors that affect the load, this paper combines convolution, long short-term memory (LSTM), and gated recurrent unit (GRU) algorithms to propose an ultra-short-term load forecasting model based on deep learning. Firstly, more than 100,000 pieces of historical load and meteorological data from Beijing in the three years from 2016 to 2018 were collected, and the meteorological data were divided into 18 types considering the actual meteorological characteristics of Beijing. Secondly, after the standardized processing of the time-series samples, the convolution filter was used to extract the features of the high-order samples to reduce the number of training parameters. On this basis, the LSTM layer and GRU layer were used for modeling based on time series. A dropout layer was introduced after each layer to reduce the risk of overfitting. Finally, load prediction results were output as a dense layer. In the model training process, the mean square error (MSE) was used as the objective optimization function to train the deep learning model and find the optimal super parameter. In addition, based on the average training time, training error, and prediction error, this paper verifies the effectiveness and practicability of the load prediction model proposed under the deep learning structure in this paper by comparing it with four other models including GRU, LSTM, Conv-GRU, and Conv-LSTM. [ABSTRACT FROM AUTHOR]

Based on a set of deep learning and mode decomposition methods, a short-term prediction model for PM2.5 concentration for Beijing city is established in this paper. An ensemble empirical mode decomposition (EEMD) algorithm is first used to decompose the original PM2.5 timeseries to several high- to low-frequency intrinsic mode functions (IMFs). Each IMF component is then trained and predicted by a combination of three neural networks: back propagation network (BP), long short-term memory network (LSTM), and a hybrid network of a convolutional neural network (CNN) + LSTM. The results showed that both BP and LSTM are able to fit the low-frequency IMFs very well, and the total prediction errors of the summation of all IMFs are remarkably reduced from 21 g/m3 in the single BP model to 4.8 g/m3 in the EEMD + BP model. Spatial information from 143 stations surrounding Beijing city is extracted by CNN, which is then used to train the CNN+LSTM. It is found that, under extreme weather conditions of PM2.5 < 35 g/m3 and PM2.5 > 150 g/m3, the prediction errors of the CNN + LSTM model are improved by ~30% compared to the single LSTM model. However, the prediction of the very high-frequency IMF mode (IMF-1) remains a challenge for all neural networks, which might be due to microphysical turbulences and chaotic processes that cannot be resolved by the above-mentioned neural networks based on variable–variable relationship. [ABSTRACT FROM AUTHOR]

As a symbol of Chinese culture, Chinese traditional-style architecture defines the unique characteristics of Chinese cities. The visual qualities and spatial distribution of architecture represent the image of a city, which affects the psychological states of the residents and can induce positive or negative social outcomes. Hence, it is important to study the visual perception of Chinese traditional-style buildings in China. Previous works have been restricted by the lack of data sources and techniques, which were not quantitative and comprehensive. In this paper, we proposed a deep learning model for automatically predicting the presence of Chinese traditional-style buildings and developed two view indicators to quantify the pedestrians' visual perceptions of buildings. Using this model, Chinese traditional-style buildings were automatically segmented in streetscape images within the Fifth Ring Road of Beijing and then the perception of Chinese traditional-style buildings was quantified with two view indictors. This model can also help to automatically predict the perception of Chinese traditional-style buildings for new urban regions in China, and more importantly, the two view indicators provide a new quantitative method for measuring the urban visual perception in street level, which is of great significance for the quantitative research of tourism route and urban planning. [ABSTRACT FROM AUTHOR]

Existing water gauge reading approaches based on image analysis have problems such as poor scene adaptability and weak robustness. Here, we proposed a novel water level measurement method based on deep learning (YOLOv5s, convolutional neural network) to overcome these problems. The proposed method uses the YOLOv5s to extract the water gauge area and all scale character areas in the original video image, uses image processing technology to identify the position of the water surface line, and then calculates the actual water level elevation. The proposed method is validated with a video monitoring station on a river in Beijing, and the results show that the systematic error of the proposed method is only 7.7 mm, the error is within 1 cm/the error is between 1 cm and 3 cm, and the proportion of the number of images is 95%/5% (daylight), 98%/2% (infrared lighting at night), 97%/2% (strong light), 45%/44% (transparent water body), 91%/9% (rainfall), and 90%/10% (water gauge is slightly dirty). The results demonstrate that the proposed method shows good performance in different scenes, and its effectiveness has been confirmed. At the same time, it has a strong robustness and provides a certain reference for the application of deep learning in the field of hydrological monitoring. [ABSTRACT FROM AUTHOR]

The fast and accurate extraction of plastic greenhouses over large areas is important for environmental and agricultural management. Traditional spectral index methods and object-based methods can suffer from poor transferability or high computational costs. Current deep learning-based algorithms are seldom specifically aimed at extracting plastic greenhouses at large scales. To extract plastic greenhouses at large scales with high accuracy, this study proposed a new deep learning-based network, U-FDS Net, specifically for plastic greenhouse extraction over large areas. U-FDS Net combines full-scale dense connections and adaptive deep supervision and has strong future fusion capabilities, allowing more accurate extraction results. To test the extraction accuracy, this study compiled new greenhouse datasets covering Beijing and Shandong with a total number of more than 12,000 image samples. The results showed that the proposed U-FDS net is particularly suitable for complex backgrounds and reducing false positive conditions for nongreenhouse ground objects, with the highest mIoU (mean intersection over union) an increase of ~2%. This study provides a high-performance method for plastic greenhouse extraction to enable environmental management, pollution control and agricultural plans. [ABSTRACT FROM AUTHOR]

Groundwater depletion is adversely affecting Beijing's ecology and environment. However, the effective execution of the South-to-North Water Diversion Project's middle route (SNDWP-MR) is anticipated to mitigate Beijing's groundwater depletion. Here, we propose a robust hybrid statistical downscaling method aimed at enhancing the capability of the Gravity Recovery and Climate Experiment (GRACE) to detect the small-scale groundwater storage anomaly (GWSA) in Beijing. We used three deep learning (DL) methods to reconstruct the 0.5° × 0.5° terrestrial water storage anomaly (TWSA) between 2004 and 2021. Moreover, multiple processing strategies were used to downscale the GWSA to 0.25° from 2004 to 2021 by integrating wells and GRACE/GRACE follow-on data from the optimal DL model. Additionally, we analyzed the spatiotemporal evolution trends of GW in Beijing before and after the implementation of the SNDWP-MR. The results show that the long short-term memory model delivers optimal performance in the TWSA reconstruction of Beijing, with the correlation coefficient (CC), Nash–Sutcliffe coefficient (NSE), and root mean square error (RMSE) being 0.98, 0.96, and 10.19 mm, respectively. The GWSA before and after downscaling is basically consistent with wells data, but the CC and RMSE of downscaling the GWSA from 2004 to 2021 are improving by 34% and 31%, respectively. Before the SNDWP-MR (2004–2014), the trend of GWSA in Beijing was − 17.68 ± 4.46 mm/y, with a human contribution of 69.30%. After SNDWP-MR (2015–2021), GWSA gradually increased by 10.00 mm per year, with the SNDWP-MR accounting for 18.30%. This study delivers a technical innovation reference for dynamically monitoring a small-scale GWSA from GRACE/GRACE-FO data. [ABSTRACT FROM AUTHOR]

In this paper, we consider water surface object detection in natural scenes. Generally, background subtraction and image segmentation are the classical object detection methods. The former is highly susceptible to variable scenes, so its accuracy will be greatly reduced when detecting water surface objects due to the changing of the sunlight and waves. The latter is more sensitive to the selection of object features, which will lead to poor generalization as a result, so it cannot be applied widely. Consequently, methods based on deep learning have recently been proposed. The River Chief System has been implemented in China recently, and one of the important requirements is to detect and deal with the water surface floats in a timely fashion. In response to this case, we propose a real-time water surface object detection method in this paper which is based on the Faster R-CNN. The proposed network model includes two modules and integrates low-level features with high-level features to improve detection accuracy. Moreover, we propose to set the different scales and aspect ratios of anchors by analyzing the distribution of object scales in our dataset, so our method has good robustness and high detection accuracy for multi-scale objects in complex natural scenes. We utilized the proposed method to detect the floats on the water surface via a three-day video surveillance stream of the North Canal in Beijing, and validated its performance. The experiments show that the mean average precision (MAP) of the proposed method was 83.7%, and the detection speed was 13 frames per second. Therefore, our method can be applied in complex natural scenes and mostly meets the requirements of accuracy and speed of water surface object detection online. [ABSTRACT FROM AUTHOR]

The concentration of PM2.5 is an important index to measure the degree of air pollution. When it exceeds the standard value, it is considered to cause pollution and lower the air quality, which is harmful to human health and can cause a variety of diseases, i.e., asthma, chronic bronchitis, etc. Therefore, the prediction of PM2.5 concentration is helpful to reduce its harm. In this paper, a hybrid model called CNN-BiLSTM-Attention is proposed to predict the PM2.5 concentration over the next two days. First, we select the PM2.5 concentration data in hours from January 2013 to February 2017 of Shunyi District, Beijing. The auxiliary data includes air quality data and meteorological data. We use the sliding window method for preprocessing and dividing the corresponding data into a training set, a validation set, and a test set. Second, CNN-BiLSTM-Attention is composed of the convolutional neural network, bidirectional long short-term memory neural network, and attention mechanism. The parameters of this network structure are determined by the minimum error in the training process, including the size of the convolution kernel, activation function, batch size, dropout rate, learning rate, etc. We determine the feature size of the input and output by evaluating the performance of the model, finding out the best output for the next 48 h. Third, in the experimental part, we use the test set to check the performance of the proposed CNN-BiLSTM-Attention on PM2.5 prediction, which is compared by other comparison models, i.e., lasso regression, ridge regression, XGBOOST, SVR, CNN-LSTM, and CNN-BiLSTM. We conduct short-term prediction (48 h) and long-term prediction (72 h, 96 h, 120 h, 144 h), respectively. The results demonstrate that even the predictions of the next 144 h with CNN-BiLSTM-Attention is better than the predictions of the next 48 h with the comparison models in terms of mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R2). [ABSTRACT FROM AUTHOR]

Urban road intersection bottleneck has become an important factor in causing traffic delay and restricting traffic efficiency. It is essential to explore the prediction of the operating performance at intersections in real-time and formulate corresponding strategies to alleviate intersection delay. However, because of the sophisticated intersection traffic condition, it is difficult to capture the intersection traffic Spatio-temporal features by the traditional data and prediction methods. The development of big data technology and the deep learning model provides us a good chance to address this challenge. Therefore, this paper proposes a multi-task fusion deep learning (MFDL) model based on massive floating car data to effectively predict the passing time and speed at intersections over different estimation time granularity. Moreover, the grid model and the fuzzy C-means (FCM) clustering method are developed to identify the intersection area and derive a set of key Spatio-temporal traffic parameters from floating car data. In order to validate the effectiveness of the proposed model, the floating car data from ten intersections of Beijing with a sampling rate of 3s are adopted for the training and test process. The experiment result shows that the MFDL model enables us to capture the Spatio-temporal and topology feature of the traffic state efficiently. Compared with the traditional prediction method, the proposed model has the best prediction performance. The interplay between these two targeted prediction variables can significantly improve prediction accuracy and efficiency. Thereby, this method predicts the intersection operation performance in real-time and can provide valuable insights for traffic managers to improve the intersection's operation efficiency. [ABSTRACT FROM AUTHOR]

Deep learning provides a promising approach for air pollution prediction. The existing deep learning-based predicted models generally consider either the temporal correlations of air quality monitoring stations or the nonlinear relationship between the PM2.5 (particulate matter with an aerodynamic diameter of less than 2.5 μm) concentrations and explanatory variables. Spatial correlation has not been effectively incorporated into prediction models, therefore exhibiting poor performance in PM2.5 prediction tasks. Additionally, determining the manner by which to expand longer-term prediction tasks is still challenging. In this paper, to allow for spatiotemporal correlations, a spatiotemporal convolutional recursive long short-term memory (CR-LSTM) neural network model is proposed for predicting the PM2.5 concentrations in long-term prediction tasks by combining a convolutional long short-term memory (ConvLSTM) neural network and a recursive strategy. Herein, the ConvLSTM network was used to capture the complex spatiotemporal correlations and to predict the future PM2.5 concentrations; the recursive strategy was used for expanding the long-term prediction tasks. The CR-LSTM model was used to realize the prediction of the future 24 h of PM2.5 concentrations for 12 air quality monitoring stations in Beijing by configuring both the appropriate time lag derived from the temporal correlations and the spatial neighborhood, including the hourly historical PM2.5 concentrations, the daily mean meteorological data, and the annual nighttime light and normalized difference vegetation index (NDVI). The results showed that the proposed CR-LSTM model achieved better performance (coefficient of determination (R2) = 0.74; root mean square error (RMSE) = 18.96 μg/m3) than other common models, such as multiple linear regression (MLR), support vector regression (SVR), the conventional LSTM model, the LSTM extended (LSTME) model, and the temporal sliding LSTM extended (TS-LSTME) model. The proposed CR-LSTM model, implementing a combination of geographical rules, recursive strategy, and deep learning, shows improved performance in longer-term prediction tasks. [ABSTRACT FROM AUTHOR]

The physical presence of a street, called the "street view", is a medium through which people perceive the urban form. A street's spatial ratio is the main feature of the street view, and its measurement and quality are the core issues in the field of urban design. The traditional method of studying urban aspect ratios is manual on-site observation, which is inefficient, incomplete and inaccurate, making it difficult to reveal overall patterns and influencing factors. Street view images (SVI) provide large-scale urban data that, combined with deep learning algorithms, allow for studying street spatial ratios from a broader space-time perspective. This approach can reveal an urban forms' aesthetics, spatial quality, and evolution process. However, current streetscape research mainly focuses on the creation and maintenance of spatial data infrastructure, street greening, street safety, urban vitality, etc. In this study, quantitative research of the Beijing street spatial ratio was carried out using street view images, a convolution neural network algorithm, and the classical street spatial ratio theory of urban morphology. Using the DenseNet model, the quantitative measurement of Beijing's urban street location, street aspect ratio, and the street symmetry was realized. According to the model identification results, the law of the gradual transition of the street spatial ratio was depicted (from the open and balanced type to the canyon type and from the historical to the modern). Changes in the streets' spatiotemporal characteristics in the central area of Beijing were revealed. Based on this, the clustering and distribution phenomena of four street aspect ratio types in Beijing are discussed and the relationship between the street aspect ratio type and symmetry is summarized, selecting a typical lot for empirical research. The classical theory of street spatial proportion has limitations under the conditions of high-density development in modern cities, and the traditional urban morphology theory, combined with new technical methods such as streetscape images and deep learning algorithms, can provide new ideas for the study of urban space morphology. [ABSTRACT FROM AUTHOR]

As an essential material carrier of cultural heritage, the accurate identification and effective monitoring of buildings in traditional Chinese villages are of great significance to the sustainable development of villages. However, along with rapid urbanization in recent years, many towns have experienced problems such as private construction, hollowing out, and land abuse, destroying the traditional appearance of villages. This study combines deep learning technology and UAV remote sensing to propose a high-precision extraction method for conventional village architecture. Firstly, this study constructs the first sample database of traditional village architecture based on UAV remote sensing orthophotos of eight representative villages in Beijing, combined with fine classification; secondly, in the face of the diversity and complexity of the built environment in traditional villages, we use the Mask R-CNN instance segmentation model as the basis and Path Aggregate Feature Pyramid Network (PAFPN) and Atlas Space Pyramid Pool (ASPP) as the main strategies to enhance the backbone model for multi-scale feature extraction and fusion, using data increment and migration learning as auxiliary means to overcome the shortage of labeled data. The results showed that some categories could achieve more than 91% accuracy, with average precision, recall, F1-score, and Intersection over Union (IoU) values reaching 71.3% (+7.8%), 81.9% (+4.6%), 75.7% (+6.0%), and 69.4% (+8.5%), respectively. The application practice in Hexi village shows that the method has good generalization ability and robustness, and has good application prospects for future traditional village conservation. [ABSTRACT FROM AUTHOR]

Trend prediction based on sensor data in a multi-sensor system is an important topic. As the number of sensors increases, we can measure and store more and more data. However, the increase in data has not effectively improved prediction performance. This paper focuses on this problem and presents a distributed predictor that can overcome unrelated data and sensor noise: First, we define the causality entropy to calculate the measurement's causality. Then, the series causality coefficient (SCC) is proposed to select the high causal measurement as the input data. To overcome the traditional deep learning network's over-fitting to the sensor noise, the Bayesian method is used to obtain the weight distribution characteristics of the sub-predictor network. A multi-layer perceptron (MLP) is constructed as the fusion layer to fuse the results from different sub-predictors. The experiments were implemented to verify the effectiveness of the proposed method by meteorological data from Beijing. The results show that the proposed predictor can effectively model the multi-sensor system's big measurement data to improve prediction performance. [ABSTRACT FROM AUTHOR]

Many Chinese cities have severe air pollution due to the rapid development of the Chinese economy, urbanization, and industrialization. Particulate matter (PM2.5) is a significant component of air pollutants. It is related to cardiopulmonary and other systemic diseases because of its ability to penetrate the human respiratory system. Forecasting air PM2.5 is a critical task that helps governments and local authorities to make necessary plans and actions. Thus, in the current study, we develop a new deep learning approach to forecast the concentration of PM2.5 in three major cities in China, Beijing, Shijiazhuang, and Wuhan. The developed model is based on the Informer architecture, where the attention distillation block is improved with a residual block-inspired structure from efficient networks, and we named the model ResInformer. We use air quality index datasets that cover 98 months collected from 1 January 2014 to 17 February 2022 to train and test the model. We also test the proposed model for 20 months. The evaluation outcomes show that the ResInformer and ResInformerStack perform better than the original model and yield better forecasting results. This study's methodology is easily adapted for similar efforts of fast computational modeling. [ABSTRACT FROM AUTHOR]

The segmentation of high-resolution (HR) remote sensing images is very important in modern society, especially in the fields of industry, agriculture and urban modelling. Through the neural network, the machine can effectively and accurately extract the surface feature information. However, using the traditional deep learning methods requires plentiful efforts in order to find a robust architecture. In this paper, we introduce a neural network architecture search (NAS) method, called NAS-HRIS, which can automatically search neural network architecture on the dataset. The proposed method embeds a directed acyclic graph (DAG) into the search space and designs the differentiable searching process, which enables it to learn an end-to-end searching rule by using gradient descent optimization. It uses the Gumbel-Max trick to provide an efficient way when drawing samples from a non-continuous probability distribution, and it improves the efficiency of searching and reduces the memory consumption. Compared with other NAS, NAS-HRIS consumes less GPU memory without reducing the accuracy, which corresponds to a large amount of HR remote sensing imagery data. We have carried out experiments on the WHUBuilding dataset and achieved 90.44% MIoU. In order to fully demonstrate the feasibility of the method, we made a new urban Beijing Building dataset, and conducted experiments on satellite images and non-single source images, achieving better results than SegNet, U-Net and Deeplab v3+ models, while the computational complexity of our network architecture is much smaller. [ABSTRACT FROM AUTHOR]

The efficient and accurate application of deep learning in the remote sensing field largely depends on the pre-processing technology of remote sensing images. Particularly, image fusion is the essential way to achieve the complementarity of the panchromatic band and multispectral bands in high spatial resolution remote sensing images. In this paper, we not only pay attention to the visual effect of fused images, but also focus on the subsequent application effectiveness of information extraction and feature recognition based on fused images. Based on the WorldView-3 images of Tongzhou District of Beijing, we apply the fusion results to conduct the experiments of object recognition of typical urban features based on deep learning. Furthermore, we perform a quantitative analysis for the existing pixel-based mainstream fusion methods of IHS (Intensity-Hue Saturation), PCS (Principal Component Substitution), GS (Gram Schmidt), ELS (Ehlers), HPF (High-Pass Filtering), and HCS (Hyper spherical Color Space) from the perspectives of spectrum, geometric features, and recognition accuracy. The results show that there are apparent differences in visual effect and quantitative index among different fusion methods, and the PCS fusion method has the most satisfying comprehensive effectiveness in the object recognition of land cover (features) based on deep learning. [ABSTRACT FROM AUTHOR]

Accurate mapping of dams can provide useful information about geographical locations and boundaries and can help improve public dam datasets. However, when applied to disaster emergency management, it is often difficult to completely determine the distribution of dams due to the incompleteness of the available data. Thus, we propose an automatic and intelligent extraction method that combines location with post-segmentation for dam detection. First, we constructed a dataset named RSDams and proposed an object detection model, YOLOv5s-ViT-BiFPN (You Only Look Once version 5s-Vision Transformer-Bi-Directional Feature Pyramid Network), with a training method using deep transfer learning to generate graphical locations for dams. After retraining the model on the RSDams dataset, its precision for dam detection reached 88.2% and showed a 3.4% improvement over learning from scratch. Second, based on the graphical locations, we utilized an improved Morphological Building Index (MBI) algorithm for dam segmentation to derive dam masks. The average overall accuracy and Kappa coefficient of the model applied to 100 images reached 97.4% and 0.7, respectively. Finally, we applied the dam extraction method to two study areas, namely, Yangbi County of Yunnan Province and Changping District of Beijing in China, and the recall rates reached 69.2% and 81.5%, respectively. The results show that our method has high accuracy and good potential to serve as an automatic and intelligent method for the establishment of a public dam dataset on a regional or national scale. [ABSTRACT FROM AUTHOR]

Pollen allergies are seasonal epidemic diseases that are accompanied by high incidence rates, especially in Beijing, China. With the development of deep learning, key progress has been made in the task of automatic pollen grain classification, which could replace the time-consuming and laborious manual identification process using a microscope. In China, few pioneering works have made significant progress in automatic pollen grain classification. Therefore, we first constructed a multi-class and large-scale pollen grain dataset for the Beijing area in preparation for the task of pollen classification. Then, a deblurring pipeline was designed to enhance the quality of the pollen grain images selectively. Moreover, as pollen grains vary greatly in size and shape, we proposed an easy-to-implement and efficient multi-scale deep learning architecture. Our experimental results showed that our architecture achieved a 97.7% accuracy, based on the Resnet-50 backbone network, which proved that the proposed method could be applied successfully to the automatic identification of pollen grains in Beijing. [ABSTRACT FROM AUTHOR]

Quickly and accurately extracting buildings from remote sensing images is essential for urban planning, change detection, and disaster management applications. In particular, extracting buildings that cannot be sheltered in emergency shelters can help establish and improve a city's overall disaster prevention system. However, small building extraction often involves problems, such as integrity, missed and false detection, and blurred boundaries. In this study, EfficientUNet+, an improved building extraction method from remote sensing images based on the UNet model, is proposed. This method uses EfficientNet-b0 as the encoder and embeds the spatial and channel squeeze and excitation (scSE) in the decoder to realize forward correction of features and improve the accuracy and speed of model extraction. Next, for the problem of blurred boundaries, we propose a joint loss function of building boundary-weighted cross-entropy and Dice loss to enforce constraints on building boundaries. Finally, model pretraining is performed using the WHU aerial building dataset with a large amount of data. The transfer learning method is used to complete the high-precision extraction of buildings with few training samples in specific scenarios. We created a Google building image dataset of emergency shelters within the Fifth Ring Road of Beijing and conducted experiments to verify the effectiveness of the method in this study. The proposed method is compared with the state-of-the-art methods, namely, DeepLabv3+, PSPNet, ResUNet, and HRNet. The results show that the EfficientUNet+ method is superior in terms of Precision, Recall, F1-Score, and mean intersection over union (mIoU). The accuracy of the EfficientUNet+ method for each index is the highest, reaching 93.01%, 89.17%, 91.05%, and 90.97%, respectively. This indicates that the method proposed in this study can effectively extract buildings in emergency shelters and has an important reference value for guiding urban emergency evacuation. [ABSTRACT FROM AUTHOR]

Urban street shadows can provide essential information for many applications, such as the assessment and protection of ecology and environment, livability evaluation, etc. In this research, we propose an effective and rapid method to quantify the diurnal and spatial changes of urban street shadows, by taking Beijing city as an example. In the method, we explore a novel way of transferring street characteristics to semantically segment street-level panoramic images of Beijing by using DeepLabv3+. Based on the segmentation results, the shading situation is further estimated by projecting the path of the sun in a day onto the semantically segmented fisheye photos and applying our firstly defined shading coverage index formula. Experimental results show that in several randomly selected sampling regions in Beijing, our method can successfully detect more than 83% of the shading changes compared to the ground truth. The results of this method contribute to the study of urban livability and the evaluation of human life comfort. The quantitative evaluation method of the shading coverage index proposed in this research has certain promotion significance and can be applied to shading-related research in other cities. [ABSTRACT FROM AUTHOR]

Accurate forecasting of future meteorological elements is critical and has profoundly affected human life in many aspects from rainstorm warning to flight safety. The conventional numerical weather prediction (NWP) sometimes leads to unsatisfactory performance due to inappropriate initial state settings. In this paper, a short-term weather forecasting model based on wavelet packet denoising and Catboost is proposed, which takes advantage of the fusion information combining the historical observation data with the prior knowledge from NWP. The feature selection and spatiotemporal feather addition are also explored to further improve performance. The proposed method is evaluated on the datasets provided by Beijing weather stations. Experimental results demonstrate that compared with many deep-learning or machine-learning methods such as LSTM, Seq2Seq, and random forest, the proposed Catboost model incorporated with wavelet packet denoising can achieve shorter convergence time and higher prediction accuracy. [ABSTRACT FROM AUTHOR]

Heat waves may negatively impact the economy and human life under global warming. The use of air conditioners can reduce the vulnerability of humans to heat wave disasters. However, air conditioner usage has been not clear until now. Traditional registration investigation methods are cumbersome and require expensive labor and time. This study used a Labelme image tagging tool and an available street view images database to firstly establish a monographic dataset to detect external air conditioner unit features and proposed two deep learning algorithms of Mask-RCNN and YOLOv5 to automatically retrieve air conditioners. The training dataset used street view images in the 2nd Ring Road area of downtown Beijing. The model evaluation mAP of Mask-RCNN and YOLOv5 reached 0.99 and 0.9428. In comparison, the performance of YOLOv5 was superior, which is attributed to the YOLOv5 model being better at detecting smaller target entities equipped with a lighter network structure and an enhanced feature extraction network. We demonstrated the feasibility of using street view images to retrieve air conditioners and showed their great potential to detect air conditioners in the future. [ABSTRACT FROM AUTHOR]

The urban environment has a great impact on the wellbeing of citizens and it is of great significance to understand how citizens perceive and evaluate places in a large scale urban region and to provide scientific evidence to support human-centered urban planning with a better urban environment. Existing studies for assessing urban perception have primarily relied on low efficiency methods, which also result in low evaluation accuracy. Furthermore, there lacks a sophisticated understanding on how to correlate the urban perception with the built environment and other socio-economic data, which limits their applications in supporting urban planning. In this study, a new data-enabled intelligence framework for evaluating human perceptions of urban space is proposed. Specifically, a novel classification-then-regression strategy based on a deep convolutional neural network and a random-forest algorithm is proposed. The proposed approach has been applied to evaluate the perceptions of Beijing and Chengdu against six perceptual criteria. Meanwhile, multi-source data were employed to investigate the associations between human perceptions and the indicators for the built environment and socio-economic data including visual elements, facility attributes and socio-economic indicators. Experimental results show that the proposed framework can effectively evaluate urban perceptions. The associations between urban perceptions and the visual elements, facility attributes and a socio-economic dimension have also been identified, which can provide substantial inputs to guide the urban planning for a better urban space. [ABSTRACT FROM AUTHOR]

Transport emissions and street dust are important sources of summertime air pollution in urban centers. Street greening and buildings have an influence on the diffusion of air pollution from streets. For field measurements, many studies have analyzed the effect of street green space arrangement on the diffusion of air pollution, but these studies have neglected the patterns at the landscape scale. Other studies have analyzed the effects of the large scale of green space on air pollution, but the vertical distribution of street buildings and greening has rarely been considered. In this study, we analyzed the impact of the vertical distribution of urban street green space on summertime air pollution in urban centers on the urban scale for the first time by using a deep-learning method to extract the vertical distribution of street greening and buildings from street view image data. A total of 687,354 street view images were collected. The green index and building index were proposed to quantify the street greening and street buildings. The multilevel regression method was used to analyze the association between the street green index, building index and air pollution indexes. For the cases in this study, including the central urban areas of Beijing, Shanghai and Nanjing, our multilevel regressions results suggested that, in the central area of the city, the vertical distribution of street greening and buildings within a certain range of the monitoring site is association with the summertime air pollution index of the monitoring site. There was a significant negative association between the street greening and air pollution indexes (radius = 1–2 km, NO2, p = 0.042; radius = 3–4 km, AQI, p = 0.034; PM10, p = 0.028). The street length within a certain range of the monitoring site has a positive association with the air pollution indexes (radius = 1–2 km, AQI, p = 0.072; PM10, p = 0.062). With the increase of the distance between streets and the monitoring sites, the association between streets and air pollution indexes decreases. Our findings on the association between the vertical structure of street greening, street buildings and summertime air pollution in urban centers can support urban street planning. [ABSTRACT FROM AUTHOR]

Accurate estimation of fine particulate matter with diameter ≤2.5 μm (PM2.5) at a high spatiotemporal resolution is crucial for the evaluation of its health effects. Previous studies face multiple challenges including limited ground measurements and availability of spatiotemporal covariates. Although the multiangle implementation of atmospheric correction (MAIAC) retrieves satellite aerosol optical depth (AOD) at a high spatiotemporal resolution, massive non-random missingness considerably limits its application in PM2.5 estimation. Here, a deep learning approach, i.e., bootstrap aggregating (bagging) of autoencoder-based residual deep networks, was developed to make robust imputation of MAIAC AOD and further estimate PM2.5 at a high spatial (1 km) and temporal (daily) resolution. The base model consisted of autoencoder-based residual networks where residual connections were introduced to improve learning performance. Bagging of residual networks was used to generate ensemble predictions for better accuracy and uncertainty estimates. As a case study, the proposed approach was applied to impute daily satellite AOD and subsequently estimate daily PM2.5 in the Jing-Jin-Ji metropolitan region of China in 2015. The presented approach achieved competitive performance in AOD imputation (mean test R2: 0.96; mean test RMSE: 0.06) and PM2.5 estimation (test R2: 0.90; test RMSE: 22.3 μg/m3). In the additional independent tests using ground AERONET AOD and PM2.5 measurements at the monitoring station of the U.S. Embassy in Beijing, this approach achieved high R2 (0.82–0.97). Compared with the state-of-the-art machine learning method, XGBoost, the proposed approach generated more reasonable spatial variation for predicted PM2.5 surfaces. Publically available covariates used included meteorology, MERRA2 PBLH and AOD, coordinates, and elevation. Other covariates such as cloud fractions or land-use were not used due to unavailability. The results of validation and independent testing demonstrate the usefulness of the proposed approach in exposure assessment of PM2.5 using satellite AOD having massive missing values. [ABSTRACT FROM AUTHOR]

Class imbalance is a key issue for the application of deep learning for remote sensing image classification because a model generated by imbalanced samples training has low classification accuracy for minority classes. In this study, an accurate classification approach using the multistage sampling method and deep neural networks was proposed to classify imbalanced data. We first balance samples by multistage sampling to obtain the training sets. Then, a state-of-the-art model is adopted by combining the advantages of atrous spatial pyramid pooling (ASPP) and Encoder-Decoder for pixel-wise classification, which are two different types of fully convolutional networks (FCNs) that can obtain contextual information of multiple levels in the Encoder stage. The details and spatial dimensions of targets are restored using such information during the Decoder stage. We employ four deep learning-based classification algorithms (basic FCN, FCN-8S, ASPP, and Encoder-Decoder with ASPP of our approach) on multistage training sets (original, MUS1, and MUS2) of WorldView-3 images in southeastern Qinghai-Tibet Plateau and GF-2 images in northeastern Beijing for comparison. The experiments show that, compared with existing sets (original, MUS1, and identical) and existing method (cost weighting), the MUS2 training set of multistage sampling significantly enhance the classification performance for minority classes. Our approach shows distinct advantages for imbalanced data. [ABSTRACT FROM AUTHOR]