Knowledge Acquisition for the Interpretation of Spatial Data

Document Sample
scope of work template
							Knowledge Acquisition for the
Interpretation of Spatial Data


          Monika Sester
                   Abstract
   在review後,提出符合「半自動模式獲得系統」
    (semi-automatic model acquisition system)的
    方法論,並證明其在某些應用上是具有潛力的,
    亦即建立了找尋data中 所具有的概念或知識的
    一個方法,並對這個方法作評價,且說明其優
    缺點,最後將之利用於空間資料的詮釋上,及
    略述了這個方法在資料詮釋工作上的可利用性。
                       Review
   Mohan & Nevatia (1989)
   Haala (1996)
   Strat、Bhanu & Poggio (1994)
   Wiston (1977)
   Connell & Bardy (1985)
   Quinlan (1990)、Wrobel (1994) 、Pearce & Caelli (1997)
   Koperski & Han (1995)
   Mackaness et al. (1997)
   Weibel et al. (1995)
     Principle of Human Learning
   Seeing and finding relations to already existing
    knowledge (Anderson 1980)
   Be analysed based on known object,features and
    relation
       Approach and Principle of
          Machine Learning
   Learning classification tree with ID3
   Object-oriented learning and interpretation
    procedure
   Object-oriented data structure
     1. Properties of objects and relation
     2. Basic structuring methods
    Learning the Structure of Amp
               Object
   Example for learning object descriptions
   Example for learning relations
   Example of suburban area
   Summary and evaluation of the approach
           Further Applications
   Derivation of models for image interpretation
   Map matching---data integration
   Map generalization and database generalization
報告完畢

謝   謝
     Learning Classification Tree
              With ID3
   Quinlan (1986):藉已分類完成的樣本所形成的
    決策樹,來對未知物做分類
   優點
   缺點
   Example
Input
Output
    Object-oriented Learning and
      Interpretation Procedure
   原     理:每一個物件具一個屬性,並與其他
    物件共享某種關係
   假     設:幾何特性、2D物件
   訓練原則:概念由學習而建立;訓練者負責組
    織結構
   學習系統
Interaction of Learning and
       Interpretation
Shell of Object clarres
Representation of a Drawing of
         an Airplane
           Properties of Objects
   Contains/contained
   Object-elongation
   Object-lq
   Object-size
   Polytype
   Object-form
   Num-point、num-right、num-parallel…
   Junction-ell、junction-tee、junction-frk….
            Relations of Objects
   Connection
   Inside,enclose
   Are-parallel,are-orthogonal
   Common-ell、common-tee、common-arw…
   Size-diff
   Distance
   Left-position,top-position
   Common-sides
   Same-polytype
                     Example
   A:[ contains: yes ,contained: no ,object-elongation
       :1.4 ,object-elongation: 2 ,…..
   rel(A,B)=[ connection: no ,inside:no ,encloses:yes
               ,are-parallel:no ,….
      Basic Structuring Methods
   Recognize-object
   Part-of
Learning Environment(left)with
 Scene to Be Interpreted(right)
判斷物體類別標準之學習
系統自動建立物件類別
街道與軌道間關連性的學習
學習後的結果
完整語意網模式的形成
Example of Suburban Area
分類結果
Summary and Evaluation of the
         Approach
   可由程式語言來支援學習及詮釋環境物件
   兼顧詮釋data的豐富性及演算法的複雜性
   技術簡單
   概念的學習是有效的
   結構僅能由要素組成
   所有的物件由某種物件分出來,其屬性相同
   學習是歸納的過程,不能證明這的概括化的正
    確性
   樣本的選取需小心

						
Related docs