Please wait a minute...


Remote Sensing Technology and Application  2018, Vol. 33 Issue (3): 408-417    DOI: 10.11873/j.issn.1004-0323.2018.3.0408
A Novel Method to Make Feature Space Adequately Complex:A Case Study on the Classification of Underlying Surfacein Urban Area from Remote Sensing Data
Ren Zhehao,Zhou Jianhua
(School of Geographic Sciences,East China Normal University,Shanghai 200241,China)
Download:  PDF (11730KB) 
Export:  BibTeX | EndNote (RIS)      
Abstract  An adequately complex feature space is indispensible as classifying underlying surface in urban area from remote sensing data.Therefore,more independent descriptors are required to increase dimension of the feature space.However,pervasive basic descriptors,as we all know,are usually not enough to construct the feature space.Three novel and pervasive approaches to getting new descriptors by extending these basic descriptors are explored in this paper.They are introduced as follows.1) Take standard deviation of neighbourhood elements in a basic descriptor as weight to indicate neighbourhood\|based multi\|scale information for the center pixel and name the approach as NMIS.The NMIS\|extended value of the center pixel is summed from several layers.These layers are different from each other only in the size of neighbourhood in which the standard deviation is calculated.2) Form multi\|scale texture layers by using a set of size\|given structure elements and name this approach as STIM.Each layer is a STIM\|extended descriptor and serves as an independent descriptor in the feature space.With a set of STIM\|extended descriptors having a basic texture descriptor as their common source,the difference in coarseness between classes can be identified.3) The third extending approach knows as polymorphic density dimension.The density dimension (De) is an algorithm for combining multiple basic features into a single descriptor to indicate geographical distribution of neighborhood elements carrying these features.Compared with previous De,a descriptor of the polymorphic De also combines multiple basic features but allows these features in different types (e.g.being spectrum and texture ones).The extending descriptor is independent from anyone of these combined features and able to be added into the feature space including these features.Accuracy assessment indicated that the average overall accuracy of classification with an extended\|descriptor\|involved input feature vector is 7.86% better than that with only basic\|descriptor\|involved one.
Key words:  Urban underlying surface; Neighbourhood\      based; Multi-scale; Polymorphic density dimension; Remote sensing     
Received:  01 August 2017      Published:  04 July 2018
E-mail this article
Add to my bookshelf
Add to citation manager
E-mail Alert
Articles by authors
Ren Zhehao
Zhou Jianhua

Cite this article: 

Ren Zhehao,Zhou Jianhua. A Novel Method to Make Feature Space Adequately Complex:A Case Study on the Classification of Underlying Surfacein Urban Area from Remote Sensing Data. Remote Sensing Technology and Application, 2018, 33(3): 408-417.

URL:     OR

No Suggested Reading articles found!