Proceedings of the 5th ACM SIGSPATIAL International Workshop on AI for Geographic Knowledge Discovery最新文献

{"title":"Incorporating spatial context for post-OCR in map images","authors":"M. Namgung, Yao-Yi Chiang","doi":"10.1145/3557918.3565864","DOIUrl":"https://doi.org/10.1145/3557918.3565864","url":null,"abstract":"Extracting text from historical maps using Optical Character Recognition (OCR) engines often results in partially or incorrectly recognized words due to complex map content. Previous work utilizes lexical-based approaches with linguistic context or applies language models to correct OCR results for documents. However, these post-OCR methods cannot directly consider spatial relations of map text for correction. For example, \"Mississippi\" and \"River\" constitute the place phrase \"Mississippi River\" (linguistic relation), and near \"highway\", there are likely to exist intersected \"road\" to enter the \"highway\" (spatial relation). This paper presents a novel approach that exploits the spatial arrangement of map text using a contextual language model, BART [6] for post-processing of map text from OCR. The approach first structures word-level map text into sentences based on their spatial arrangement while preserving the spatial location of words constituting a place name and corrects imperfect OCR text using neighboring information. To train BART for capturing spatial relations in map text, we automatically generate large numbers of synthetic maps to fine-tune BART with location names and their spatial context. We conduct experiments on synthetic and real-world historical maps of various map styles and scales and show that the proposed method can achieve significant improvement over the commonly used lexical approach.","PeriodicalId":428859,"journal":{"name":"Proceedings of the 5th ACM SIGSPATIAL International Workshop on AI for Geographic Knowledge Discovery","volume":"34 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129706635","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"highway2vec: representing OpenStreetMap microregions with respect to their road network characteristics","authors":"Kacper Leśniara, Piotr Szymański","doi":"10.1145/3557918.3565865","DOIUrl":"https://doi.org/10.1145/3557918.3565865","url":null,"abstract":"Recent years brought advancements in using neural networks for representation learning of various language or visual phenomena. New methods freed data scientists from hand-crafting features for common tasks. Similarly, problems that require considering the spatial variable can benefit from pretrained map region representations instead of manually creating feature tables that one needs to prepare to solve a task. However, very few methods for map area representation exist, especially with respect to road network characteristics. In this paper, we propose a method for generating microregions' embeddings with respect to their road infrastructure characteristics. We base our representations on OpenStreetMap road networks in a selection of cities and use the H3 spatial index to allow reproducible and scalable representation learning. We obtained vector representations that detect how similar map hexagons are in the road networks they contain. Additionally, we observe that embeddings yield a latent space with meaningful arithmetic operations. Finally, clustering methods allowed us to draft a high-level typology of obtained representations. We are confident that this contribution will aid data scientists working on infrastructure-related prediction tasks with spatial variables.","PeriodicalId":428859,"journal":{"name":"Proceedings of the 5th ACM SIGSPATIAL International Workshop on AI for Geographic Knowledge Discovery","volume":"71 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131464565","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Remote sensing visual question answering with a self-attention multi-modal encoder","authors":"João Daniel Silva, João Magalhães, D. Tuia, Bruno Martins","doi":"10.1145/3557918.3565874","DOIUrl":"https://doi.org/10.1145/3557918.3565874","url":null,"abstract":"Visual Question Answering (VQA) on remote sensing imagery can help non-expert users in extracting information from Earth observation data. Current approaches follow a neural encoder-decoder design, combining convolutional and recurrent encoders together with cross-modal fusion components. However, in other VQA application domains, the current state-of-the-art methods rely on self-attention, employing multi-modal encoders based on the Transformer architecture. In this work, we assess the degree to which a model based on self-attention can bring improvements over previous methods for remote sensing VQA. We specifically present results with an extended version of a previous model named MM-BERT, originally proposed for medical VQA and which does not require the extraction of region features from the images, or model pre-training with extensive amounts of data. Experiments show that the proposed method can improve results over previous approaches. Even without in-domain pre-training or specific adaptations to the remote sensing domain, and using as input low-resolution versions of the images, we can achieve a high accuracy over three different datasets extensively used in previous studies.","PeriodicalId":428859,"journal":{"name":"Proceedings of the 5th ACM SIGSPATIAL International Workshop on AI for Geographic Knowledge Discovery","volume":"17 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126642904","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Fine-grained location prediction of non geo-tagged tweets: a multi-view learning approach","authors":"Mohammad Abboud, K. Zeitouni, Y. Taher","doi":"10.1145/3557918.3565875","DOIUrl":"https://doi.org/10.1145/3557918.3565875","url":null,"abstract":"Geotagged Social Media (GTSM) data, especially geotagged tweets are valuable sources of information for many important applications. Only small portions of geotagged tweets are available (less than 3%). Identifying tweet location is a challenging problem that has attracted the interest of both academic and industry fields. Existing approaches have satisfactory accuracy at country and city level, but fail in locating more precisely the tweets. This paper presents FLAIR, an approach for geolocating tweets at finer granularities. Our objective is to predict the tweet location in a well-known and pre-defined area, that is to reduce the distance error between the predicted and real locations. In this work, we propose a location prediction model leveraging spatial model for POIs extracted from a text from one hand, and textual model comparing text similarity between geotagged and non-geotagged tweets, from another hand. We adopt a multi-view learning approach to combine the results of both predictions. Experimental results show that our proposed model outperforms the existing solutions.","PeriodicalId":428859,"journal":{"name":"Proceedings of the 5th ACM SIGSPATIAL International Workshop on AI for Geographic Knowledge Discovery","volume":"36 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122744762","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Density-based cluster detection at multiple spatial scales via kullback-leibler divergence of reachability profiles","authors":"Orhun Aydin, C. Osorio-Murillo, Cheng-Chia Huang","doi":"10.1145/3557918.3565870","DOIUrl":"https://doi.org/10.1145/3557918.3565870","url":null,"abstract":"Density-based clustering methods are frequently used to define spatial clusters and outliers (noise) for location-only data. Different algorithms for solving this problem emerged over the past few decades, with their main difference being the numerical representation of the spatial density. A problem not addressed by conventional density-based clustering methods is defining alternate spatial cluster maps at statistically significant spatial scales. This problem differs from conventional clustering, as the goal of finding alternate clusters is to define different spatial cluster maps for all statistically significant spatial scales. Knowledge of distinct spatial scales pertinent to clustering is important for understanding various scales underlying the data. In addition, alternate clusters with different spatial scales can inform decisions that require to be made at different spatial granularity. In this paper, we introduce a statistical test that uses Kullback-Leibler (KL) divergence loss between different spatial density profiles to identify all statistically significant spatial scales at which clustering occurs. The proposed method defines different clustering maps that reflect different scales at which spatial clusters occur. We define the divergence on a 1-D representation of cluster density, the reachability profile, to cluster spatial units with varying spatial scales. We illustrate the use of multiple spatial clustering at different scales by comparing the proposed method to the state-of-the-art for defining a single map of multiscale clusters, HDBScan. We conclude the paper by applying the proposed method to physical and human geography problems, area of interest delineation, and wildfire cluster modeling, respectively.","PeriodicalId":428859,"journal":{"name":"Proceedings of the 5th ACM SIGSPATIAL International Workshop on AI for Geographic Knowledge Discovery","volume":"37 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125068230","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"SPEMI","authors":"Jun Tang, Haoxiang Zhang, Binjie Zhang, Jiahui Jin, Y. Lyu","doi":"10.1145/3557918.3565873","DOIUrl":"https://doi.org/10.1145/3557918.3565873","url":null,"abstract":"Region embedding is a primary task for a wide variety of urban-related downstream applications. However, many existing embedding techniques neglected the fact that the regions in a city have been developed differently by many factors such as planning policies, economic, and population mitigation. Such a spatial imbalance problem may result in a quite different region embedding to distinguish differences between regions, even though the regions could be similar in terms of the certain application tasks. In this paper, we propose a SPatial EMInence (SPEMI) model that normalizes region embeddings to mitigate the effects from spatial imbalance. In particular, we present a context-aware spatial feature, called spatial eminence, that measures a region's importance to its spatial context. The experimental results of store placement recommendation using real-world urban data show that SPEMI improves the performance of citywide region embeddings by up to 27.92%.","PeriodicalId":428859,"journal":{"name":"Proceedings of the 5th ACM SIGSPATIAL International Workshop on AI for Geographic Knowledge Discovery","volume":"2 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130750574","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Unsupervised historical map registration by a deformation neural network","authors":"Sidi Wu, R. Schnürer, M. Heitzler, L. Hurni","doi":"10.1145/3557918.3565871","DOIUrl":"https://doi.org/10.1145/3557918.3565871","url":null,"abstract":"Image registration that aligns multi-temporal or multi-source images is vital for tasks like change detection and image fusion. Thanks to the advance and large-scale practice of modern surveying methods, multi-temporal historical maps can be unlocked and combined to trace object changes in the past, potentially supporting research in environmental science, ecology and urban planning, etc. Even when maps are geo-referenced, the contained geographical features can be misaligned due to surveying, painting, map generalization, and production bias. In our work, we adapt an end-to-end unsupervised deformation network that couples rigid and non-rigid transformations to align scanned historical map sheets at different time stamps. To the best of our knowledge, we are the first to use unsupervised deep learning to register map images. We address the sparsity of map features by introducing a loss based on distance fields. When aligning the displaced landmark locations by our proposed method, the results are promising both quantitatively and qualitatively. The generated smooth deformation grid can be applied to vector features directly to align them from the source map sheet to the target map sheet.","PeriodicalId":428859,"journal":{"name":"Proceedings of the 5th ACM SIGSPATIAL International Workshop on AI for Geographic Knowledge Discovery","volume":"48 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116630455","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Real-time GeoAI for high-resolution mapping and segmentation of arctic permafrost features: the case of ice-wedge polygons","authors":"Wenwen Li, Chia-Yu Hsu, Sizhe Wang, C. Witharana, A. Liljedahl","doi":"10.1145/3557918.3565869","DOIUrl":"https://doi.org/10.1145/3557918.3565869","url":null,"abstract":"This paper introduces a real-time GeoAI workflow for large-scale image analysis and the segmentation of Arctic permafrost features at a fine-granularity. Very high-resolution (0.5m) commercial imagery is used in this analysis. To achieve real-time prediction, our workflow employs a lightweight, deep learning-based instance segmentation model, SparseInst, which introduces and uses Instance Activation Maps to accurately locate the position of objects within the image scene. Experimental results show that the model can achieve better accuracy of prediction at a much faster inference speed than the popular Mask-RCNN model.","PeriodicalId":428859,"journal":{"name":"Proceedings of the 5th ACM SIGSPATIAL International Workshop on AI for Geographic Knowledge Discovery","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130530307","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"SHGCN: a hypergraph-based deep learning model for spatiotemporal traffic flow prediction","authors":"Yi Wang, Di Zhu","doi":"10.1145/3557918.3565866","DOIUrl":"https://doi.org/10.1145/3557918.3565866","url":null,"abstract":"Traffic flow prediction, as one of the prominent tasks in intelligent transportation systems, is challenging due to underlying complex spatiotemporal characteristics. Consideration of historical spatial and temporal dependencies is essential for the traffic prediction of a geographic unit for a future time period. Existing works mainly adopted graphs to represent the irregular layout of spatial units, where nodes are signal of spatial units and edges are link strengths between units. For contemporary deep learning based spatiotemporal prediction tasks, the temporal dependence can be well modeled via convolution neural network or recurrent neural network, and spatial dependence features are commonly captured using graph convolution networks. However, classic graph structures cannot fully represent the complex nature of spatial relationships in transportation networks, because the spatial pattern of a location might be influenced by multiple sets of contextual information simultaneously, while a graph edge can only describe the linkage between two nodes. In addition, most existing models ignore the synchronous dependence between temporal and spatial features, leading to a mismatch between the temporal and spatial features of a location. Based on such problems, a hypergraph-based deep learning model, namely synchronous hypergraph convolutional network (SHGCN), is proposed to better capture the complex relationships between spatial and temporal knowledge. A novel synchronous hypergraph cell (SH-Cell) is designed based on LSTM cells integrated in the form of a Seq2seq architecture. Then, we construct dynamic hypergraphs to capture the synchronous spatiotemporal dependence adaptively using SH-Cells. Experimental results demonstrate the superiority of SHGCN over well-known benchmarks on two real-world publicly-available traffic datasets. This research provides new insights for improving the traffic flow prediction accuracy and understanding complex spatiotemporal relationships towards a more reliable urban traffic management.","PeriodicalId":428859,"journal":{"name":"Proceedings of the 5th ACM SIGSPATIAL International Workshop on AI for Geographic Knowledge Discovery","volume":"44 2 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123372097","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Towards the intelligent era of spatial analysis and modeling","authors":"Di Zhu, Song Gao, Guofeng Cao","doi":"10.1145/3557918.3565863","DOIUrl":"https://doi.org/10.1145/3557918.3565863","url":null,"abstract":"Geographic phenomena are considered complex due to the heterogeneous nature of spatial dependencies. It is impossible to specify a universal law described in statistical or physical languages that can perfectly characterize a real-world geographic process and explain how it forms certain observed patterns. Traditional spatial analytics based on strict statistical principles, strong assumptions, or classic computation workflows are facing great challenges and opportunities when embracing the explosive growth of geospatial data and recent technical innovations. Here, we highlight the promises of Intelligent Spatial Analytics (ISA), a new set of spatial analytical approaches based on spatially explicit deep neural networks with more flexible data representation, modules for complex spatial dependence, weaker model prior assumptions, and hence the enhanced ability to predict/explain unknowns. Three essential topics in spatial analysis, i.e., geostatistics, spatial econometrics, and flow analytics are elaborated as examples in the vision of ISA. We also discuss challenging issues of ISA as an invitation to explore deeper linkages between machine/deep learning and spatial analysis at the frontier of Geospatial Artificial Intelligence.","PeriodicalId":428859,"journal":{"name":"Proceedings of the 5th ACM SIGSPATIAL International Workshop on AI for Geographic Knowledge Discovery","volume":"49 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116295253","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}