结构增强型生成对抗网络SAR图像超分辨率重建

针对利用生成对抗网络模型(Generative Adversarial Network,GAN)重建SAR(Synthetic Aperture Radar)图像存在边缘细节信息不足和“伪影”(artifacts)现象,该文基于增强型超分辨率生成对抗网络(Enhanced Super-Resolution Generative Adversarial Networks,ESRGAN)光学模型,重新设计生成网络上采样重建模块和结构损失函数,提出一种结构增强型生成对抗网络SAR图像超分辨率重建算法,包括特征提取、特征增强和上采样重建3个模块:在特征提取模块采用小尺度卷积层对输入SAR图像进行低层次特征提取;在特征增强模块采用多个级联残差密集块(Residual-in-Residual Dense Block,RRDB)和卷积层提取输入特征;在上采样重建模块交替使用最近邻插值(Nearest Neighbor Interpolation,NNI)和亚像素卷积(Sub-Pixel Convolution,SPC)对特征进行放大重建,使特征信息交互融合。与传统插值算法和经典深度学习重建算法相比,该算法在视觉效果和定量评价方面均有显著提升,能够在保持原网络模型重建图像内容信息不丢失的基础上,增强重建图像边缘细节信息和减缓“伪影”现象,有利于后续目标识别和灾害监测等工作开展。     

A lightweight ensemble spatiotemporal interpolation model for geospatial data

ABSTRACT

Missing data is a common problem in the analysis of geospatial information. Existing methods introduce spatiotemporal dependencies to reduce imputing errors yet ignore ease of use in practice. Classical interpolation models are easy to build and apply; however, their imputation accuracy is limited due to their inability to capture spatiotemporal characteristics of geospatial data. Consequently, a lightweight ensemble model was constructed by modelling the spatiotemporal dependencies in a classical interpolation model. Temporally, the average correlation coefficients were introduced into a simple exponential smoothing model to automatically select the time window which ensured that the sample data had the strongest correlation to missing data. Spatially, the Gaussian equivalent and correlation distances were introduced in an inverse distance-weighting model, to assign weights to each spatial neighbor and sufficiently reflect changes in the spatiotemporal pattern. Finally, estimations of the missing values from temporal and spatial were aggregated into the final results with an extreme learning machine. Compared to existing models, the proposed model achieves higher imputation accuracy by lowering the mean absolute error by 10.93 to 52.48% in the road network dataset and by 23.35 to 72.18% in the air quality station dataset and exhibits robust performance in spatiotemporal mutations.     

Deep spatio-temporal residual neural networks for road-network-based data modeling

Recently, researchers have introduced deep learning methods such as convolutional neural networks (CNN) to model spatio-temporal data and achieved better results than those with conventional methods. However, these CNN-based models employ a grid map to represent spatial data, which is unsuitable for road-network-based data. To address this problem, we propose a deep spatio-temporal residual neural network for road-network-based data modeling (DSTR-RNet). The proposed model constructs locally-connected neural network layers (LCNR) to model road network topology and integrates residual learning to model the spatio-temporal dependency. We test the DSTR-RNet by predicting the traffic flow of Didi cab service, in an 8-km² region with 2,616 road segments in Chengdu, China. The results demonstrate that the DSTR-RNet maintains the spatial precision and topology of the road network as well as improves the prediction accuracy. We discuss the prediction errors and compare the prediction results to those of grid-based CNN models. We also explore the sensitivity of the model to its parameters; this will aid the application of this model to network-based data modeling.     

Regularities in Spatial Information Processing: Implications for Modeling Destination Choice

This paper investigates the determinants of spatial knowledge and how our knowledge of space varies according to geographic location. By using data on U.S. city names recalled at 22 test locations, a multivariate model of the information surface specific to each test location is calibrated. This model links the probability of a city being recalled from memory to its distance from the test site, its population size, its location with respect to other cities, and whether or not it is a state capital. The paper then suggests that these recall data provide information on spatial knowledge surfaces from which large-scale spatial choices, such as migration destinations, are made. Results from the analysis lend further evidence to the idea that spatial knowledge is stored and processed hierarchically and that individuals underrepresent information in large clusters. Consequently, the results have important implications for modeling any spatial behavior based on individuals' spatial information surfaces. In particular, the results cast further doubt on the validity of traditional large-scale spatial choice frameworks and lend support to the competing destinations hypothesis.     

Advanced spatial statistics for analysing and visualizing geo-referenced data

Abstract

Spatial statistics supplies advanced methods for analysing environmental data, and copes with observational interdependencies similar to the way principal components analysis treats multicollinearity. The U.S. Environmental Protection Agency’s Environmental Monitoring and Assessment Program (EMAP) utilizes kriging from geostatistics for mapping and visualizing environmental data. A conceptual framework is articulated between the interpolation problem in kriging and the missing data problem in spatial statistics, with special reference to relations between the exponential semi-variogram and the conditional autoregressive models. Supercomputing experiments are summarized that simplify numerically the probability density function normalizing factor, which is of particular relevance to estimation tasks for the EMAP project.     

Leveraging correlation across space and time to interpolate geophysical data via CoKriging

Managing geophysical data generated by emerging spatiotemporal data sources (e.g. geosensor networks) presents a growing challenge to Geographic Information System science. The presence of correlation poses difficulties with respect to traditional spatial data analysis. This paper describes a novel spatiotemporal analytical scheme that allows us to yield a characterization of correlation in geophysical data along the spatial and temporal dimensions. We resort to a multivariate statistical model, namely CoKriging, in order to derive accurate spatiotemporal interpolation models. These predict unknown data by utilizing not only their own geosensor values at the same time, but also information from near past data. We use a window-based computation methodology that leverages the power of temporal correlation in a spatial modeling phase. This is done by also fitting the computed interpolation model to data which may change over time. In an assessment, using various geophysical data sets, we show that the presented algorithm is often able to deal with both spatial and temporal correlations. This helps to gain accuracy during the interpolation phase, compared to spatial and spatiotemporal competitors. Specifically, we evaluate the efficacy of the interpolation phase by using established machine-learning metrics (i.e. root mean squared error, Akaike information criterion and computation time).     

Automated terrain feature identification from remote sensing imagery: a deep learning approach

ABSTRACT

Terrain feature detection is a fundamental task in terrain analysis and landscape scene interpretation. Discovering where a specific feature (i.e. sand dune, crater, etc.) is located and how it evolves over time is essential for understanding landform processes and their impacts on the environment, ecosystem, and human population. Traditional induction-based approaches are challenged by their inefficiency for generalizing diverse and complex terrain features as well as their performance for scalable processing of the massive geospatial data available. This paper presents a new deep learning (DL) approach to support automatic detection of terrain features from remotely sensed images. The novelty of this work lies in: (1) a terrain feature database containing 12,000 remotely sensed images (1,000 original images and 11,000 derived images from data augmentation) that supports data-driven model training and new discovery; (2) a DL-based object detection network empowered by ensemble learning and deep and deeper convolutional neural networks to achieve high-accuracy object detection; and (3) fine-tuning the model’s characteristics and behaviors to identify the best combination of hyperparameters and other network factors. The introduction of DL into geospatial applications is expected to contribute significantly to intelligent terrain analysis, landscape scene interpretation, and the maturation of spatial data science.     

A human-machine adversarial scoring framework for urban perception assessment using street-view images

ABSTRACT

Though global-coverage urban perception datasets have been recently created using machine learning, their efficacy in accurately assessing local urban perceptions for other countries and regions remains a problem. Here we describe a human-machine adversarial scoring framework using a methodology that incorporates deep learning and iterative feedback with recommendation scores, which allows for the rapid and cost-effective assessment of the local urban perceptions for Chinese cities. Using the state-of-the-art Fully Convolutional Network (FCN) and Random Forest (RF) algorithms, the proposed method provides perception estimations with errors less than 10%. The driving factor analysis from both the visual and urban functional aspects demonstrated its feasibility in facilitating local urban perception derivations. With high-throughput and high-accuracy scorings, the proposed human-machine adversarial framework offers an affordable and rapid solution for urban planners and researchers to conduct local urban perception assessments.     

Spatial interpolation using MLP–RBFN hybrid networks

It is easy for a multi-layered perception (MLP) to fit a stratified spatial interpolation pattern whose form is close to open surface; while it is easy for a radial basis function network (RBFN) to fit a pocket (radial) spatial interpolation pattern whose form is close to closed surface. However, in the real world, the spatial interpolation pattern may consist of stratified and pocket patterns. Neither MLP nor RBFN can fit the pattern easily. To combine their advantages to fit the complex hybrid spatial interpolation patterns, in this article we propose a novel neural network, MLP–RBFN hybrid network (MRHN), whose hidden layer contains sigmoid and Gaussian units at the same time. Although there are two kinds of processing units in MRHN, in this study we used the principle of minimizing the error sum of squares to derive the supervised learning rules for all the network parameters. This research took rainfall distribution in Taiwan as a case study. The results show that (1) the prediction error of the testing dataset outside the training dataset demonstrated that MRHN was the most accurate among the three networks, RBFN was the next best, and MLP was the worst; (2) the MLP model seriously underestimated the values of high observed rainfall; (3) over-learning may be a serious shortcoming of using RBFN in spatial interpolation applications; (4) MRHN may have better generalization learning capacity than RBFN in spatial interpolation applications.     

A treatise on the use of geostatistics for the characterization of nonrenewable resources

Five decades of geostatistical development are reviewed to summarize the state of the art for spatial interpolation vis-à-vis kriging or a form thereof. Although a search of the literature reveals a variety of kriging methods, there are but two infrastructures for geostatistical interpolation: simple cokriging, for estimating a single variable using two variables, and generalized cokriging, for estimating one or more variables using the same number of variables that are estimated. The many forms of kriging are varieties of these two interpolation infrastructures. This notion is emphasized to aid the selection of an appropriate interpolation model for a nonrenewable resource. These models are discussed, and literature for the models and for applicable software is cited. Additionally, all aspects of spatial interpolation are discussed, including the adequacy of spatial sampling, distribution characteristics of spatial samples, semivariograms, search parameters, and selection of interpolation models in conformance with spatial data characteristics. Finally, the relationship between interpolation and raster-based geographic information systems is emphasized.