Diferenças

Aqui você vê as diferenças entre duas revisões dessa página.

--- thales:geodma_paper [2010/08/03 20:17] – tkorting
+++ thales:geodma_paper [2010/09/06 20:40] (atual) – removed tkorting
@@ Linha 1: / Linha 1: @@
-====== Interpreting Images with GeoDMA ======
-===== Abstract =====
-Object oriented analysis offers effective tools to represent the knowledge in a scene. Knowledge-based interpretation arises as an effective way to interpret remote sensing imagery. In this approach, human’s expertise is organized in a knowledge base to be used as input of automated interpretation processes, thus enhancing performance and accuracy, and reducing at the same time the subjectivity in the interpretation process. Some systems such as Definiens, ENVI-FX, and more recently, InterIMAGE, have incorporated useful tools to aid the object-oriented classification. In this context, this work presents a system for object image analysis, called Geographical Data Mining Analyst (GeoDMA), which implements several mechanisms for automatic image interpretation. It performs image segmentation, objects feature extraction, supervised and unsupervised classification with raster, shape and cellular data sets. GeoDMA has been used for intra-urban land cover classification of high spatial resolution images and also to detect deforestation patterns in the Brazilian Amazon. Such different applications warrant the generalist behavior of the system. In this article, we discuss each module implemented in the system, the integration of objects obtained by segmentation to extract texture features and landscape metrics. We also show the main tools for data preprocessing, sample selection and feature visualization. Finally, one application of automatic urban interpretation is shown.
-===== Introduction =====
-Remotely sensed data, combined with additional data from ecosystem models, offers an unprecedented opportunity for predicting and understanding the behavior of the Earth's ecosystem. Due to the large amount of available data, novel techniques are needed to facilitate the automatic extraction and analysis of Earth Science patterns. However, the detection of these patterns is a difficult task due to the spatio-temporal nature of the data [tan2001finding].
-Data Mining is a technique commonly defined as the extraction of patterns or models from observed data, usually as part of a more general process of extracting high-level potentially useful knowledge, from low-level data. Such process is generally known as Knowledge Discovery in Databases (KDD). The automated process of KDD can be linked to visual tools, with interactive data presentation and query resources, allowing domain experts to quickly examine ``what if'' scenarios. According to [oliveira2003visual], visual data exploration is a completely human guided process, whereas straight data mining algorithms analyze data sets automatically. Therefore a stronger strategy would be to tightly couple visualization and analytical processes into one KDD tool.
-Algorithms for data mining provide the analysis of large sets of observational images to find (un)suspected relationships, and to summarize the data in novel ways that are both understandable and useful to stakeholders. Classification results in remote sensing are defined as objects discernable in images, which can be either crisp or fuzzy and vague. Crisp objects are well-defined entities with sharp boundaries. Vague objects are either objects with a poor definition, or objects with a vague boundary. Landscape examples of vague objects are a mountain with no apparent support, or a city having gradual transition zones to the rural land and a contaminated river [stein2009handling].
-Remote sensing satellites are currently the most significant source of new data about the planet, and satellite image databases are the fastest growing archives of spatial information. For instance, the Brazil's National Institute for Space Research (INPE) holds more than 130 terabytes of image data, covering 30 years of remote sensing activities which are available on a database with free online access [rahman2009data]. According to [rushing2005adam], the sheer volume of geoscience data makes it a perfect case for application of data mining techniques. Such techniques applied to remote sensing imagery deals specifically with the challenge of capturing patterns, processes, and agents present in the geographic space, in order to extract specific knowledge to understand or to make decisions related to a set of relevant topics, including land change, climate variations and biodiversity studies.
-In the area of analysis tools for remote sensing data, many Free and Open Source Geographical Information Systems (FOSGIS) initiatives are becoming available in the recent years. They aim to cover several aspects of geographical applications, such as visualization and analysis of spatial data, or analysis of raster data, such as images or digital terrain models [camara2009free]. In this work we describe a novel software for image analysis based on data mining techniques, called Geographical Data Mining Analyst -- GeoDMA. It has been developed by the Image Processing Division at INPE (Brazil's National Institute for Space Research).
-===== Related Work =====
-Here is a description of some free softwares used in the scenario of geographical data classification.
-The InterIMAGE system aims at developing an open-source, knowledge-based framework for automatic interpretation of remote sensing data. It integrates image processing operators in the interpretation process, and provides access to external programs by its control mechanism, that can be coded in any computer language [costa2010knowledge]. The system is founded on geoAIDA [buckner2001geoaida], inheriting from that system its interpretation engine and some of its basic knowledge representation structures.
-The central goal of InterIMAGE is to aid the user in the scene interpretation, providing a symbolic meaning to the different regions that compose the scene. InterIMAGE integrates different georreferenced data covering the region of interest, such as a collection of images acquired at different points in time or acquired by different sensors, or even vector data.
-Proposed by Anselin //et.al.// [anselin2006gis], //GeoDa software for geodata analysis// is a free software designed to perform spatial analysis. GeoDa is driven by a point and click interface and does not need any programming expertise. The application includes functionalities like simple mapping, exploratory data analysis, visualization of global and local spatial autocorrelation, and spatial regression. According to the authors, the system provides the user a natural path through an empirical spatial data analysis exercise, through functions available in the software.
-GeoDa provides data analysis with several statistical inferences, and creating of variables using mathematical functions, algebra, and rate smoothing. On the other hand, it does not allow the generation of new variables like those based on shape or spectral analysis. One drawback is that the source-code is not open, and thus it does not allow the users with computational knowledge to contribute with new functionalities. There are currently two versions of the system. One is named //Legacy GeoDa//, and is not fully and open source project, since it relies into proprietary ESRI's MapObjects library [marshall2002developing]. The other version, //OpenGeoDa//, is defined as a cross-platform, open source version of the program. OpenGeoDa runs on Windows, Mac OS, and Linux.
-Bação //et.al.// [baccao2004gso] describe Geo-SOM, a variant of the well known Self-Organizing Maps (SOM). It enables the user to interact and to explore spatial census data, underlying its fuzzy nature and it is adapted to deal with specific features of spatial data, such as geographic location. It was implemented in Matlab, and the classification algorithm is based in the idea of creating a bridge between geographic and attributes space.
-Geo-SOM was already used to perform spatial census data classification, however it does not provide access to other types of data, such as spectral values, found in satellite imagery, or to assemble one cellular space with the data.
-Takatsuka and Gahegan [takatsuka2002geovista] describes the GeoVISTA Studio, which was built to support complex geoscientific datasets analysis as collaboration between computationally based and human-based expertise. GeoVISTA performs data analysis by connecting components developed using Java programming language. Every component contains mechanisms for easy and consistent interaction, whose behavior and appearance can be customized. After creating a diagram connecting all components, GeoVISTA Studio produces standalone applications, or applets.
-GeoVISTA Studio provides an interesting way to perform data mining in geospatial datasets. However, this interactivity of knowledge discovery from databases is slow. For every change made in the application, the components connection has to be edited to generate a new strategy for the applet.
-Openshaw [openshaw1999gdm] analyze some data mining tools, such as Intelligent Data Miner, MineSet, PASW Modeler, and STATISTICA. However, they were not designed to take into account geographical features. Differently, they were developed to analyze large commercial databases to model and predict customer-buying behavior. Thus, there is a need for user-friendly systems to analyze geographical data so they can be exploited by domain scientists [miller2001gdm]. These systems must offer tools for data visualization to stimulate the imagination and to help users on decision-making as suggested by Pelekis [pelekis2005lrs].
-//Falta descrever o sistema ADaM e a proposta do Marcelino//
-===== System Description =====
-==== Input data ====
-==== Attributes extraction ====
-==== Exploratory analysis ====
-==== Classification ====
-===== Experiments in GeoDMA =====
-==== Detecting deforestation changes in the Amazon forest ====
-Trabalho da Érika
-==== Deforestation Example ====
-Exemplo do Artigo do Marcelino
-==== Intra-urban classification ====
-Trabalho da Carolina
-===== Conclusions =====