Asanobu Kitamoto

Spatio-Temporal Data Mining for Typhoon Image Collection

Our research aims at discovering useful knowledge from the large collection of satellite images of typhoons using data mining approaches. We first introduce the creation of the typhoon image collection that consists of around 34,000 typhoon images for the northern and southern hemisphere, providing the medium-sized, richly-variational and quality-controlled data collection suitable for spatio-temporal data mining research. Next we apply several data mining approaches for this image collection. We start with spatial data mining, where principal component analysis is used for extracting basic components and reducing dimensionality, and it revealed that the major principal components describe latitudinal structures and spiral bands. Moreover, clustering procedures give the “birds-eye-view” visualization of typhoon cloud patterns. We then turn to temporal data mining, including state transition rules, but we demonstrate that it involves intrinsic difficulty associated with the nonlinear dynamics of the atmosphere, or chaos. Finally we briefly introduce our system IMET (Image Mining Environment for Typhoon analysis and prediction), which is designed for the intelligent and efficient searching and browsing of the typhoon image collection.

Image Classification Using Probabilistic Models that Reflect the Internal Structure of Mixels

The purpose of this paper is to establish an image classification method which properly considers the spatial quantisation effect of digital imagery and its inevitable consequence, the presence of mixels. To achieve this goal, we propose two new probabilistic models, namely the area proportion distribution and mixel distribution. The former probabilistic model serves as the prior distribution of area proportions that reflect the internal structure of mixels, and Beta distribution is proposed as the general model of the area proportion distribution. On the other hand, the latter probabilistic model is a unique model both in concept and shape, and its uniqueness is the source of its effectiveness against an image histogram which can be represented by a set of trough and peak regions by means of the mixel distributions and pure pixel distributions, respectively. Moreover, the expected area proportion is proposed for computing the area proportions of mixels. Finally, experimental results on satellite image classification are analysed to validate the effectiveness of our proposed probabilistic models. By comparing the mixture density model with our proposed models to those without them, we conclude that, both in terms of quantitative and qualitative evaluation, our probabilistic models work effectively for images with the presence of mixels.

Similarity retrieval of NOAA satellite imagery by graph matching

An attributed relational graph (ARG) is introduced into our NOAA satellite image database system. The node and the branch of an ARG denotes a classified region and a spatial relationship between adjacent regions, respectively. Furthermore, a few attributes of a node/branch help to express numerical shape features of regions. Similarity retrieval thereby turns out to be equivalent to graph matching. The similarity retrieval process of the system is as follows: (1) select a visual example image as a query and generate its graph structure, (2) calculate an optimal graph matching cost between a query graph and an archived graph in the database, utilizing algorithm A* with heuristic information, (3) choose archived images in the ascending order of a corresponding matching cost.

TSum4act: A Framework for Retrieving and Summarizing Actionable Tweets During a Disaster for Reaction

Social networks (e.g. Twitter) have been proved to be an almost real-time mean of information spread, thus they can be exploited as a valuable channel of information for emergencies (e.g. disasters) during which people need updated information for suitable reactions. In this paper, we present TSum4act, a framework designed to tackle the challenges of tweets (e.g. diversity, large volume, and noise) for disaster responses. The objective of the framework is to retrieve actionable tweets (e.g. casualties, cautions, and donations) that were posted during disasters. For this purpose, the framework first identifies informative tweets to remove noise; then assigns informative tweets into topics to preserve the diversity; next summarizes the topics to be compact; and finally ranks the results for user’s faster scan. In order to improve the performance, we proposed to incorporate event extraction for enriching the semantics of tweets. TSum4act has been successfully tested on Joplin tornado dataset of 230.535 tweets and the completeness of 0.58 outperformed 17%, of the retweet baseline’s.

Mapping urban and peri-urban breeding habitats of Aedes mosquitoes using a fuzzy analytical hierarchical process based on climatic and physical parameters

The spread of dengue fever depends mainly on the availability of favourable breeding sites for its mosquito vectors around human dwellings. To investigate if the various factors influencing breeding habitats can be mapped from space, dengue indices, such as the container index, the house index and the Breteau index, were calculated from Ministry of Public health data collected three times annually in Phitsanulok, Thailand between 2009 and 2011. The most influential factors were found to be temperature, humidity, rainfall, population density, elevation and land cover. Models were worked out using parameters mostly derived from freely available satellite images and fuzzy logic software with parameter synchronisation and a predication algorithm based on data mining and the Decision Tree method. The models developed were found to be sufficiently flexible to accommodate additional parameters and sampling data that might improve prediction of favourable breeding hotspots. The algorithm applied can not only be used for the prediction of near real-time scenarios with respect to dengue, but can also be applied for monitoring other diseases influenced by environmental and climatic factors. The multi-criteria model presented is a cost-effective way of identifying outbreak hotspots and early warning systems lend themselves for development based on this strategy. The proposed approach demonstrates the successful utilisation of remotely sensed images to map mosquito breeding habitats.

Toponym-based geotagging for observing precipitation from social and scientific data streams

Weather is a typical topic of daily conversations, so it is a natural idea to use social data to observe weather. Geotagging is a key to use social data for weather applications because weather is a highly localized phenomenon on the earth. Hence we developed software called GeoNLP for toponym-based geotagging, and applied it to Twitter data stream to find toponyms (place names) from Japanese tweets talking about precipitation events. We observed that less than 10 percent of the tweets contain toponym information, but it can capture precipitation events for each place. We also show temporal relationship between rain events and tweets. A case study shows that the relative number of tweets about rain and snow indicates the status of weather. In a few months, we collected almost million tweets about precipitation events with toponyms, but bias of tweets toward highly populated area is a big problem for applying the method to rural areas. The result indicates that social data streams can be used as complementary data source to scientific data streams.

Mixture density estimation under the existence of mixels

The paper focuses on a mixed pixel or mixel. In the mixture density estimation problem, the authors include implicit distributions produced by mixels, in addition to the explicit distributions produced by pure pixels. NOAA-AVHRR channel 5 images are classified into several categories including mixel categories based on a Bayesian decision rule, and three optimization schemes are compared.

Near real-time characterisation of urban environments: a holistic approach for monitoring dengue fever risk areas

Despite frequent use of digital devices in everyday life, cost-effective measurement of public health issues in urban areas is still challenging. This study was, therefore, planned to extract land-use types using object-based and spatial metric approaches to explore the dengue incidence in relation to the surrounding environment in near real-time using Google and Advanced Land Observation Satellite images. The characterised image showed useful classification of an urban area with 77% accuracy and 0.68 kappa. Geospatial analysis on public health data indicated that most of the dengue cases were found in densely populated areas surrounded by dense vegetation. People living in independent houses having sparsely vegetated surroundings were found to be less vulnerable. Disease incidence was more prevalent in people of 5–24 years of age (67%); while in terms of occupation, mostly students, the unemployed, labourers and farmers (88%) were affected. In general, males were affected slightly more (10%) than females. Proximity analyses indicated that most of the dengue cases were around institutions (40%), religious places (18%) and markets (15%). Thus, usage of Digital Earth scalable tools for monitoring health issues would open new ways for maintaining a healthy and sustainable society in the years ahead.

3D reconstruction of a collapsed historical site from sparse set of photographs and photogrammetric map

This paper deals with the challenge of city-scale 3D reconstruction using computer vision techniques. Our method combines the photogrammetric map created from aerial photographs with photographs taken by the general public. The former gives the surface, while the latter gives the texture, and we make a 3D model step-by-step based on a semiautomatic process. We applied this method to the 3D reconstruction of the citadel of Bam, which is a collapsed historical site by the earthquake. Available photographs are limited because new images cannot be captured after the collapse, but we successfully produced a 3D model of the site with texture taken from the photograph. Our system is based on 3ds Max software with several MAXScript tools, such as automatic tools for generating mesh surface from wireframe by assuming walls, slopes and grounds, and assistance tools for a semi-automatic process of estimating camera parameters and transformation matrix.

Constructing Multilingual Preterminological Graphs using various online-community resources

We are describe the concept of dedicated Multilingual Preterminological Graphs MPGs, and some automatic approaches for constructing them by analyzing the behavior of online community users. A Multilingual Preterminological Graph is a special lexical resource that contains massive amount of terms related to a special domain, and can be used as raw material to later build a standardized terminological repository. Building such a graph is difficult using traditional approaches, as it needs huge efforts by domain specialists and terminologists. In our approach, we build such a graph by analyzing the access log files of the website of the community, and by finding the important terms that have been used to search in that website, and their association with each other. We aim at making this graph as a seed repository so multilingual volunteers can contribute. We are experimenting this approach with the Digital Silk Road Project. We have used its access log files since its beginning in 2003, and obtained an initial graph of around 116000 terms. As an application, we used this graph to obtain a preterminological multilingual database that is serving a CLIR system for the DSR project.