A Comparative Study of Three Image Representations for Population Estimation Mining Using Remote Sensing Imagery

详细信息    查看全文

• 作者：Kwankamon Dittakan (25)
  Frans Coenen (25)
  Rob Christley (26)
  Maya Wardeh (26)
  
• 关键词：Satellite Image Analysis and Mining ; Data Mining Applications ; Population Estimation Mining
• 刊名：Lecture Notes in Computer Science
• 出版年：2013
• 出版时间：2013
• 年：2013
• 卷：8346
• 期：1
• 页码：265-276
• 全文大小：597KB
• 作者单位：Kwankamon Dittakan (25)
  Frans Coenen (25)
  Rob Christley (26)
  Maya Wardeh (26)
  
  25. Department of Computer Science, University of Liverpool, Liverpool, L69 3BX, United Kingdom
  26. Institute of Infection and Global Health, University of Liverpool, Leahurst Campus, Chester High Road, CH64 7TE, Neston, Cheshire, United Kingdom
  
• ISSN：1611-3349

文摘

Census information regarding populations is important with respect to many governmental activities such as commercial, health, communication, social and infrastructure planning and development. However, the traditional census collection methods (ground surveys) are resource intensive in terms of both cost and time. The resources required for census collection activities are particularly high in rural areas which feature poor communication and transport networks. In this paper the interpretation of high-resolution satellite imagery is proposed as a low cost (but less accurate) alternative to obtaining census data. The fundamental idea is to build a classifier that can label households according to “Family size” which can then be used to generate a census estimation. The challenge is how best to translate the raw satellite data into a form that captures key information in such a way that an appropriate classifier can still be built. Three different representations are considered: (i) Colour Histogram, (ii) Local Binary Pattern and (iii) Graph-based. The representations were evaluated by generating census information using test data collected from a rural area to the northwest of Addis Ababa in Ethiopia.