Object-Oriented Analysis On Urban Spatial Information Using Remote Sensing Data

The majority of human activities occur in the cities,towns,and metropolitan areas.The issues related to urban regions are always active research areas for geographers.With the extensive applications of Geospatial Information Technologies,includingGeographical Information System (GIS),Remote Sensing (RS),and Global PositionSystem (GPS),the methods adopted in the research on the urban issues are morequantitative.RS is one of the most important and major sources for urban spatial data.The traditional way to analyze the RS data is pixel-based classification,which hasbeen proved sufferring from several drawbacks,such as low classification accuracy,very limited spatial information to be derived,and salt-and-pepper effects.Theparcel-based and object-based approaches have been proposed to improve theclassification processes,and many studies have reported better classification resultsthan per-pixel methods.Since the concept of object-orientation was first introducedinto Geographical Information Science (GIScience) community in the late 1980s,ithas been employed in the development of GIS software,construction of spatialdatabases,and representations of geospatial features.As a representation,modeling,and abstraction formalism of the reality,object-oriented paradigm has beendemonstrated to be an efficient approach to the problems of the advanced spatialfunctions,including environmental modeling,decision support,and knowledgediscovery.However,the application of object-oriented analysis for urban spatialinformation is still in infancy with the following limitations:1) object-orientedmethod was primarily applied to the urban land cover classification,little research hasbeen done on the utilization of object-oriented concepts in other aspects of urbananalysis.2) No much research efforts have been directed to the theoretical andconceptual foundation for object-oriented urban analysis.3) Some of fundamentalalgorithms and software tools are not available for supporting object-oriented analysisfor urban applications.This research represents an effort to explore conceptual basis,methods,algorithms and applications of object-oriented analysis for urban spatialinformation.The main scientific findings of this research include:1)The concept of object-orientation is not only a programming technique but also representation formalism for the human cognition of real world.In the field ofGISience,object-orientation is one fundamental model for representing geographicalfeatures,beyond its common applications in remote sensing image classification.Thebasic concepts of the abstraction,inheritance,polymorphism and encapsulationmechanism in objected-oriented analysis are found very useful for representing,organizing and analyzing geospatial features.2)This research examines the spatial representation of two different types ofgeospatial features in terms of human spatial cognition:continuous field and discreteobject.Object-oriented representation is more appropriate for representing discreteobject-like geospatial features.The representation of geospatial features as discreteobjects is more consistent with human cognition process,and object-orientedrepresentation is a higher level of human cognition and knowledge acquisition.3)This research examines three basic concepts:image object,geospatial object,andurban spatial object.The dissimilarities and relations among them are elaboratedclearly.Three criteria - Boundary,Attributes,and Scale - are proposed to identify thegeospatial objects.This research reveals each geospatial or urban spatial object iscomprised of three compontents:identity,state and behavior.These three componentsare represented by unique name (or identification number),attributes,andrelationships correspondingly.In addition,the information associated with an objecthas three dimensions:theme,space and time.The thematic,spatial,temporalattributes and relationships are the information underpinnings of geospatial and urbanspatial object.As a significant result of this dissertation,a general framework forobject-oriend analysis of urban spatial information is established to support casestudies in this dissertation research.The author believes this framework is applicableto other studies in the field.4)A series of algorithms and methods have been developed and implemented foridentifying objects,deriving attributes and determining relationships.Thesealgorithms and methods automate all the steps from the data preprocessing to theinformation derivation.The implemented algorithms include recursive connectedcomponent identification and indexing,morphological operations (closing,filling,andtrim operation),boundary tracing algorithms for inner,ourter and extended borders, 2D and 3D morphological attributes computation algorithms,determination ofthematic,spatial (topological,quasi-topological,proximal,directional) and temporalrelations between objects.This research also proposes a new term"Tightness",whichis defined as the ratio of the common boundary length between two objects to theperimeter of one of them as a reference object.The Tightness shows how interrelatedbetween two adjacent objects.This research also developed a new algorithm toquantify the proximity between two areal objects.Furthermore,10 object changescenarios and their corresponding detection are proposed in this research,which havebeen proven useful for the urban object change analysis.5)The conceptual framework and methods presented in this research have beensuccessfully applied to two application examples.In the first one,by exploiting highresolution airborne LiDAR data along with color infrared aerial photographs,a newobject-oriented two-stage method is proposed to quantify the detailed information ofurban landscape components through a case study of downtown Houston,Texas,USA.The urban landscape components are identified and classified by integrating spectralinformation,2D and 3D morphological attributes for urban objects.In the secondapplication,a new object-oriened method for representing and analyzing spatialpatterns of urban growth is put forward.The attributes of individual objects andrelationships between them are incorporated to determine the urban growth types.Theobject-oriented method has been implemented using VB.Net and C++ programminglanguages as an ArcGIS extension module and successfully applied to the analysis ofurban growth pattern of College Station,Texas,USA,during 1992-2001.Theseexamples demonstrate that the object-oriented method renders a powerful tool for theurban spatial information analysis.