DeepIndoorCrowd: Predicting crowd flow in indoor shopping malls with an interpretable transformer network

Accurate and interpretable prediction of crowd flow would benefit business management and public security. The existing studies are challenged to adapt to the indoor environment due to its complex and dynamic spatial interaction patterns. In this study, we propose a crowd flow predicting method for indoor shopping malls, which simultaneously features temporal variables and semantic factors to suit the shopping mall environment. A deep learning model named DeepIndoorCrowd is presented. The model aims at capturing temporal dependencies and the semantic pattern in crowd flow to generate an accurate multi-horizon prediction. With a multi-term temporal dependency capturing structure, the model is effective in learning both daily and weekly patterns of the indoor crowd flow in a shopping mall and is able to provide the temporal interpretation of the prediction result. Moreover, a semantic-temporal fusion module is introduced to utilize the semantic information of stores in prediction, which has proved to be effective in enhancing the model's ability to learn temporal patterns. Experiments were conducted on a real-world dataset to verify the proposed approach. The ablation study demonstrates that the DeepIndoorCrowd can effectively improve the efficiency and accuracy of the prediction up to 18.7%. In addition, some interesting indoor crowd flow patterns were discovered by analyzing the model's interpretation of the prediction result. The proposed prediction method provides an intuitive way of modeling indoor crowd flow, and the experiment's outcome can help indoor managers better understand stores' flow traffic.     

基于RNN的空气污染时空预报模型研究

针对空气污染物时间序列中包含缺失值以及现有时间序列预报模型缺乏对时序特征状态建模的问题,该文构建了基于缺失值处理算法和RNN(循环神经网络)的时空预报框架。对空气污染物时序数据设计了3种缺失值处理算法(前向递补、均值替代和权重衰减),用缺失标签和缺失时长对缺失值建模,并在此基础上搭建含有全连接层与LSTM层的深度循环神经网络(DRNN)用于时空预报。使用深度全连接神经网络(DFNN)作为DRNN的对照,用京津冀区域的空气质量和气象数据训练模型,并比较不同模型的预测精度。通过实验,比较了3种缺失值处理方法的效果,结果表明,LSTM在空气污染时空序列预测上的表现优于传统的全连接神经网络层,证实了提出的基于深度学习的时空预报框架的有效性。     

几种具有编解码结构的深度学习模型在建筑物综合中的应用与比较

深度学习技术促使诸多领域研究取得突破性进展, 基于深度神经网络的地图综合研究备受期待。将建筑物综合过程抽象解释为编解码过程, 系统地研究基于编解码结构的深度神经网络在建筑物综合中的应用。首先, 利用空间划分与矢量-栅格数据转换相结合的方式构建样本和样本集; 然后, 利用样本集训练基于编解码结构的深度神经网络, 实现建筑物综合学习泛化并测试、评估其效果; 最后, 搭建5种代表性的基于编解码结构的深度神经网络, 分析比较各模型在建筑物综合中的应用效果。实验结果表明, 基于编解码结构的深度神经网络能够从建筑物综合样本中学习或推理出部分建筑物综合知识和综合操作, 且5种模型中Pix2Pix更适用于建筑物综合的学习模拟。     

A multi‐label cellular automata model for land change simulation

The use of cellular automata (CA) has for some time been considered among the most appropriate approaches for modeling land‐use changes. Each cell in a traditional CA model has a state that evolves according to transition rules, taking into consideration its own and its neighbors’ states and characteristics. Here, we present a multi‐label CA model in which a cell may simultaneously have more than one state. The model uses a multi‐label learning method—a multi‐label support vector machine, Rank‐SVM—to define the transition rules. The model was used with a multi‐label land‐use dataset for Luxembourg, built from vector‐based land‐use data using a method presented here. The proposed multi‐label CA model showed promising performance in terms of its ability to capture and model the details and complexities of changes in land‐use patterns. Applied to historical land use data, the proposed model estimated the land use change with an accuracy of 87.2% exact matching and 98.84% when including cells with a misclassification of a single label, which is comparably better than a classical multi‐class model that achieved 83.6%. The multi‐label cellular automata outperformed a model combining CA and artificial neural networks. All model goodness‐of‐fit comparisons were quantified using various performance metrics for predictive models.     

An Algebra for Spatiotemporal Data: From Observations to Events

Recent technological advances in geospatial data gathering have created massive data sets with better spatial and temporal resolution than ever before. These large spatiotemporal data sets have motivated a challenge for Geoinformatics: how to model changes and design good quality software. Many existing spatiotemporal data models represent how objects and fields evolve over time. However, to properly capture changes, it is also necessary to describe events. As a contribution to this research, this article presents an algebra for spatiotemporal data. Algebras give formal specifications at a high‐level abstraction, independently of programming languages. This helps to develop reliable and expressive applications. Our algebra specifies three data types as generic abstractions built on real‐world observations: time series, trajectory and coverage. Based on these abstractions, it defines object and event types. The proposed data types and functions can model and capture changes in a large range of applications, including location‐based services, environmental monitoring, public health, and natural disasters.     

Mapping fine‐scale urban housing prices by fusing remotely sensed imagery and social media data

The accurate mapping of urban housing prices at a fine scale is essential to policymaking and urban studies, such as adjusting economic factors and determining reasonable levels of residential subsidies. Previous studies focus mainly on housing price analysis at a macro scale, without fine‐scale study due to a lack of available data and effective models. By integrating a convolutional neural network for united mining (UMCNN) and random forest (RF), this study proposes an effective deep‐learning‐based framework for fusing multi‐source geospatial data, including high spatial resolution (HSR) remotely sensed imagery and several types of social media data, and maps urban housing prices at a very fine scale. With the collected housing price data from China's biggest online real estate market, we produced the spatial distribution of housing prices at a spatial resolution of 5 m in Shenzhen, China. By comparing with eight other multi‐source data mining techniques, the UMCNN obtained the highest housing price simulation accuracy (Pearson R = 0.922, OA = 85.82%). The results also demonstrated a complex spatial heterogeneity inside Shenzhen's housing price distribution. In future studies, we will work continuously on housing price policymaking and residential issues by including additional sources of spatial data.     

Fuzzy Neural Network Models for Supervised Classification: Multispectral Image Analysis

Abstract

It has been well established that neural networks provide a reasonable and powerful alternative to conventional classifiers. During the past few years there has been a large and energetic upswing in research efforts aimed at synthesizing fuzzy logic with neural networks. This combination of fuzzy logic and neural networks seems natural because two approaches generally attack the design of “intelligent” systems from quite different angles. Neural networks provide algorithms for learning, classification, and optimization whereas fuzzy logic deals with issues such as reasoning on a higher (semantic or linguistic) level. Consequently the two technologies complement each other. In this paper we propose two novel fuzzy‐neural network models for supervised learning. The first model consists of three layers, and the second model consists of four layers. In both models, the first two layers implement fuzzy membership functions and the remaining layers implement the inference engine. Both models use the gradient decent technique for learning. As an illustration, we have analyzed two Thematic mapper images using these models. Results are presented in the paper.     

基于模糊神经网络的模糊空间对象生成方法研究

随着对GIS中的空间对象模型和自然地理特征表达的研究深入,模糊空间对象被提出。针对模糊空间对象表达的特点,提出了一种基于模糊神经网络的模糊空间对象生成方法。该方法将模糊技术与神经网络相结合,利用神经网络的学习能力调整模糊隶属函数和模糊规则,使系统具备自适应的特性。实验表明,这种基于模糊神经网络的生成模糊空间对象的方法比传统方法大大的提高了成果的精度。     

基于模糊神经网络的模糊空间对象生成方法研究

随着对GIS中的空间对象模型和自然地理特征表达的研究深入,模糊空间对象被提出。针对模糊空间对象表达的特点,提出了一种基于模糊神经网络的模糊空间对象生成方法。该方法将模糊技术与神经网络相结合,利用神经网络的学习能力调整模糊隶属函数和模糊规则,使系统具备自适应的特性。实验表明,这种基于模糊神经网络的生成模糊空间对象的方法比传统方法大大的提高了成果的精度。     

事件驱动的城市公共交通时空数据模型研究

以模型驱动架构为理论基础,以事件驱动的方法进行面向对象的时空数据建模研究,建立城市公共交通领域的概念模型、静态结构模型、动态行为模型和时态行为模型,并结合软件模型实现时空数据面向对象的组织与操作。事件驱动的面向对象时空数据模型实现了关系数据库和对象关系型GIS平台下面向对象的数据建模,扩展空间对象的动态行为与时态行为特征,实现对复杂空间对象的操作并保证了其空间关系及时态一致性。本研究为行业空间数据建模提供理论和方法上的参考,通过在广州市公共交通规划管理地理信息系统中的应用对模型的实用性进行验证。     

'Situating' Simulation to Model Human Spatio-Temporal Interactions: An Example Using Crime Events

Many social phenomena have a spatio‐temporal dimension and involve dynamic decisions made by individuals. In the past, researchers have often turned to geographic information systems (GIS) to model these interactions. Although GIS provide a powerful tool for examining the spatial aspects of these interactions, they are unable to model the dynamic, individual‐level interactions across time and space. In an attempt to address these issues, some researchers have begun to use simulation models. But these models rely on artificial landscapes that do not take into account the environment in which humans move and interact. This research presents the methodology for ‘situating’ simulation through the use of a new modeling tool, Agent Analyst, which integrates agent‐based modeling (ABM) and GIS. Three versions of a model of street robbery are presented to illustrate the importance of using ‘real’ data to inform agent activity spaces and movement. The successful implementation of this model demonstrates that: (1) agents can move along existing street networks; (2) land use patterns can be used to realistically distribute agent's homes and activities across a city; and (3) the incidence and pattern of street robberies is significantly different when ‘real’ data are used.     

Spatial Approaches to Modeling Dispersion of Communicable Diseases – A Review

The dispersion of communicable diseases in a population is intrinsically spatial. In the last several decades, a range of spatial approaches has been devised to model epidemiological processes; and they differ significantly from each other. A review of spatially oriented epidemiological models is necessary to assess advances in spatial approaches to modeling disease dispersion and to help identify those most appropriate for specific research goals. The most notable difference in the design of these spatially oriented models is the scale and mobility of the modeling unit. Using two criteria, this review identifies six types of spatially oriented models. These include: (1) population‐based wave models, (2) sub‐population models, (3) individual‐based cellular automata models, (4) mobile sub‐population models, (5) individual‐based spatially implicit models, and (6) individual‐based mobile models. Each model type is evaluated in terms of its design principles, assumptions, and intended applications. For the evaluation of design, four aspects of design principles are discussed: the modeling unit, the interaction between the modeling units, the spatial process, and the temporal process utilized in a design. Insights gained from this review can be useful for devising much‐needed spatially and temporally oriented strategies to forecast, prevent, and control communicable diseases.     

Estimation of soil moisture using deep learning based on satellite data: a case study of South Korea

The Korea Meteorological Administration uses soil moisture (SM) observed by the Advanced Microwave Scanning Radiometer-2 (AMSR2) to monitor drought. However, it may not be appropriate for monitoring drought in South Korea due to significant underestimation of SM. In this study, we used a deep learning method that performs better than traditional statistical and physical models for reliable estimation of SM based on remotely sensed satellite data. For estimating SM, we carefully selected input variables that exhibit a feedback loop with SM. To build an effective deep learning model, we examined the influences of sampling criteria and input parameters as well as the accuracy of several deep neural networks. The selected model was cross-validated to determine its stability. The estimated SM using deep learning had a high correlation coefficient (R) of 0.89 and a low root mean square error (RMSE; 3.825%) and bias (?0.039%) compared to in-situ measurements. A time series analysis using dynamic time warping was conducted which showed that the estimated SM was almost similar to the in-situ SM. In order to investigate the improvement in SM estimation using our method, it was compared with the Global Land Data Assimilation System and AMSR2. Significant improvements in R and a reduction in error values by more than half were achieved using our method. The estimated SM has finer spatial resolution at 4 km, and it can be rapidly produced, which will be useful for drought monitoring over the Korean Peninsula in near-real-time.     

基于深度学习的航空遥感影像密集匹配

本文探讨了深度学习在航空影像密集匹配中的性能,并与经典方法进行了比较,对模型泛化能力进行了评估。首先,实现了MC-CNN(matching cost convolutional neural network)、GC-Net(geometry and context network)、DispNet(disparity estimation network)3种代表性卷积神经元网络在航空立体像对上的训练和测试,并与传统方法SGM(semi-global matching)和商业软件SURE进行了比较。其次,利用直接迁移学习方法,评估了各模型在不同数据集间的泛化能力。最后,利用预训练模型和少量目标数据集样本,评估了模型微调的效果。试验包含3套航空影像、2套开源街景影像。试验表明:①与传统的遥感影像密集匹配方法相比,目前深度学习方法略有优势;②GC-Net与MC-CNN表现了良好的泛化能力,在开源数据集上训练的模型可以直接应用于遥感影像,且3PE(3-pixel-error)精度没有明显下降;③在训练样本不足时,利用预训练模型做初值并进行参数微调可以得到比直接训练更好的结果。     

多源数据融合的高时空分辨率植被指数生成

高时空分辨率的植被指数VI(Vegetation Index)数据是农业和生态研究的重要基础数据集,目前常用的VI数据的时空分辨率存在不可调和矛盾。考虑VI时序变化对数据融合的影响,提出一种新的VI数据时空融合模型VISTFM(Vegetation Index Spatial and Temporal Fusion Model),VISTFM采用模糊C聚类算法,对存量时序VI数据按土地利用类型划分为若干子类,从高低分辨率影像中随土地覆被类的变化规律提取子类,结合低分辨率影像提取的土地覆被类变化规律融合生成高时空分辨率的VI数据。用常用的Landsat和MODIS数据验证该算法,测试表明,VISTFM能够较好的捕获VI的中间变化过程,与常用的基于线性混合模型的模型和时空自适应反射率融合模型及其改进模型相比,利用VISTFM获得的植被指数数据集具有更高的时空分辨率。     

An Experimental Performance Evaluation of Spatio-Temporal Join Strategies

Many applications capture, or make use of, spatial data that changes over time. This requirement for effective and efficient spatio‐temporal data management has given rise to a range of research activities relating to spatio‐temporal data management. Such work has sought to understand, for example, the requirements of different categories of application, and the modelling facilities that are most effective for these applications. However, at present, there are few systems with fully integrated support for spatio‐temporal data, and thus developers must often construct custom solutions for their applications. Developers of both bespoke solutions and of generic spatio‐temporal platforms will often need to support the fusion of large spatio‐temporal data sets. Supporting such requests in a database setting involves the use of join operations with both spatial and temporal conditions – spatio‐temporal joins. However, there has been little work to date on spatio‐temporal join algorithms or their evaluation. This paper presents an evaluation of several approaches to the implementation of spatio‐temporal joins that build upon widely available indexing techniques. The evaluation explores how several algorithms perform for databases with different spatial and temporal characteristics, with a view to helping developers of generic infrastructures or custom solutions in the selection and development of appropriate spatio‐temporal join strategies.     

A reliable traffic prediction approach for bike‐sharing system by exploiting rich information with temporal link prediction strategy

Bike‐sharing systems have been widely used in major cities across the world. As bike borrowing and return at different stations in different periods are not balanced, the bikes in a bike‐sharing system need to be redistributed frequently to rebalance the system. Therefore, traffic flow forecasting of the bike‐sharing system is an important issue, as this is conducive to achieving rebalancing of the bike system. In this article, we present a new traffic flow prediction approach based on the temporal links in dynamic traffic flow networks. A station clustering algorithm is first introduced to cluster stations into groups. A temporal link prediction method based on the dynamic traffic flow network method (STW+M) is then proposed to predict the traffic flow between stations. In our method, the non‐negative tensor decomposition and time‐series analysis model capture the rich information (temporal variabilities, spatial characteristics, and weather information) of the across‐clusters transition. Then, a temporal link prediction strategy is used to forecast potential links and weights in the traffic flow network by investigating both the network structure and the results of tensor computations. In order to assess the methods proposed in this article, we have used the data of bike‐sharing systems in New York and Washington, DC to conduct bike traffic prediction and the experimental results have shown that our method produces the lowest root mean square error (RMSE) and mean square error (MSE). Compared to four prediction methods from the literature, our RMSE and MSE of the two datasets have been lowered by an average of 2.55 (Washington, DC) and 2.41 (New York) and 3.35 (Washington, DC) and 2.96 (New York), respectively. The results show that the proposed approach improves predictions of traffic flow.     

Spatial Modeling of the Black Death in Sweden

The purpose of this work is to determine whether spatial modeling can be used to model the spread of the Black Death. The study is limited to models for the propagation of the disease in Sweden in 1350. Geographic data of Swedish water bodies and medieval road networks, historical data on the population in Swedish parishes, including their medieval boundaries, along with historical notes and disease characteristics, were used to build alternative models for spatial distribution. Three different models are presented: one radial, one cost‐based and one combining network analysis and radial propagation. Simulations were made to depict different scenarios on the spread of the disease, as well as the drastic changes in the overall population of Sweden, over a couple of hundred years. For purpose of validation the population decrease estimated in each parish is compared with independent historical documents. Results from model scenarios are visualized in maps of propagation, animated video sequences and a web map service. Our analyses clearly demonstrate the power of spatial analysis and geographic information systems to describe, model and visualize epidemiologic processes in space and time.     

A spatial‐temporal‐semantic approach for detecting local events using geo‐social media data

Social media networks allow users to post what they are involved in with location information in a real‐time manner. It is therefore possible to collect large amounts of information related to local events from existing social networks. Mining this abundant information can feed users and organizations with situational awareness to make responsive plans for ongoing events. Despite the fact that a number of studies have been conducted to detect local events using social media data, the event content is not efficiently summarized and/or the correlation between abnormal neighboring regions is not investigated. This article presents a spatial‐temporal‐semantic approach to local event detection using geo‐social media data. Geographical regularities are first measured to extract spatio‐temporal outliers, of which the corresponding tweet content is automatically summarized using the topic modeling method. The correlation between outliers is subsequently examined by investigating their spatial adjacency and semantic similarity. A case study on the 2014 Toronto International Film Festival (TIFF) is conducted using Twitter data to evaluate our approach. This reveals that up to 87% of the events detected are correctly identified compared with the official TIFF schedule. This work is beneficial for authorities to keep track of urban dynamics and helps build smart cities by providing new ways of detecting what is happening in them.     

SPAWNN: A Toolkit for SPatial Analysis With Self‐Organizing Neural Networks

This article introduces the SPAWNN toolkit, an innovative toolkit for spatial analysis with self‐organizing neural networks, which is published as free and open‐source software ( http://www.spawnn.org ). It extends existing toolkits in three important ways. First, the SPAWNN toolkit distinguishes between self‐organizing neural networks and spatial context models with which the networks can be combined to incorporate spatial dependence and provides implementations for both. This distinction maintains modularity and enables a multitude of useful combinations for analyzing spatial data with self‐organizing neural networks. Second, SPAWNN interactively links different self‐organizing networks and data visualizations in an intuitive manner to facilitate explorative data analysis. Third, it implements cutting‐edge clustering algorithms for identifying clusters in the trained networks. Toolkits such as SPAWNN are particularly needed when researchers and practitioners are confronted with large amounts of complex and high‐dimensional data. The computational performance of the implemented algorithms is empirically demonstrated using high‐dimensional synthetic data sets, while the practical functionality highlighting the distinctive features of the toolkit is illustrated with a case study using socioeconomic data of the city of Philadelphia, Pennsylvania.