H-Spar: Framework Model for Efficient Land-Use Classification using Hybrid Features and Coarse to Fine Sparselets
P. Dolphin Devi1, K. Chitra2
1P. Dolphin Devi, Vice Principal, Kalvi Matriculation Higher Secondary School, Oddanchatram, Dindigul (Tamil Nadu), India.
2Dr. K. Chitra, Professor, Department of Computer Science, Government Arts College Melur, Madurai (Tamil Nadu), India.
Manuscript received on 27 March 2019 | Revised Manuscript received on 06 April 2019 | Manuscript Published on 27 April 2019 | PP: 785-790 | Volume-7 Issue-6S2 April 2019 | Retrieval Number: F11030476S219/2019©BEIESP
Open Access | Editorial and Publishing Policies | Cite | Mendeley | Indexing and Abstracting
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC-BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Abstract: Land-use (LU) classification is a challenging field in Remote Sensing (RS) and in quantitatively estimating the impact of human involvement on natural resources. This LU classification covers wide range of application in natural resource management and ability to understand the human modification. Defining and extraction of feature and classification approaches may pose new challenges for researchers. Thus, this paper proposed hybrid feature extraction namely Sparse Principal Component Analysis with Pyramid Histogram of Oriented Gradients (SPCA-PHOG) for land use classification. Furthermore spatial pattern based classification is introduced than even before. In this research paper, an innovative technique of “sparselets” with hybrid feature extraction (H-Spar) is presented which is based on efficient midlevel visual elements for land-use classification. This element is used to represents an image with huge number of part detectors rather than low level image attributes. If the part detectors increases with images, it will leads to computational complexity problem. To solve is issue, novel training framework is introduced, is represented by with huge number of part detectors namely “sparselets”. This proposed hybrid feature extraction and sparselets based classification estimated on datasets and compared with existing approaches.
Keywords: Land Use, Partlets, Sparseletes, SPCA, HOG.
Scope of the Article: Patterns and Frameworks