LAND USE MAPPING OF BERGAMA TEST AREA USING HEADWALL HYPERSPEC VNIR IMAGES


Tunç Görmüş E., AKAR Ö.

INTERNATIONAL SYMPOSIUM ON APPLIED GEOINFORMATICS, İstanbul, Turkey, 7 September - 09 November 2019, pp.1-8

  • Publication Type: Conference Paper / Full Text
  • City: İstanbul
  • Country: Turkey
  • Page Numbers: pp.1-8
  • Erzincan Binali Yildirim University Affiliated: Yes

Abstract

The most significant information that distinguish hyperspectral mages from other optical images is that these images are formed from bands of many very narrow and contiguous wavelengths. As each object has its own spectral signature, it is possible to extract information which is not visible in other optical images, but in hyperspectral images. In this study hyperspectral images were provided from General Directorate of Mapping of Turkey. These images are belong to Bergama province of İzmir and acquired from 2100m of average height by Headwall Hyperspec VNIR camera in 2017. The aim of this study is to generate the land use map of Bergama region with high accuracy, by using hyperspectral images. This study has been achieved in 5 steps.  In the first step aerial hyperspectral images were corrected atmospherically and radiometrically. Then dimensionality of these images were reduced to 20 band from 400 in order to get rid of the noisy and correlated bands. In the third step reduced data was classified by Random Forest classification method to generate the classes that will form the land use map. In order to increase the obtained classification accuracy, 8 of co-occurrence texture information, namely; mean, variance, homogeneity, contrast, dissimilarity, entropy, second moment and correlation, added to the reduced images and then classified again by Random Forest method for comparison reasons. In the last step where the classification results were discussed, it was seen that texture information have increased the classification accuracy of hyperspectral images 4%, compared to classification of hyperspectral image without texture information. This shows that texture information increases the classification accuracy of hyperspectral images. By using hyperspectral images and texture information together, it is achieved to classify land classes that have similar spectral properties where otherwise would not be classified truly.   Therefore, land use map was prepared from the classification result of hyperspectral image with texture information. As a result, this study uses the educational high resolution aerial hyperspectral data to produce land use maps both from hyperspectral data and together with texture information. The potential of generating land use maps from these data set has not been analyzed earlier to the best of authors’ knowledge.  Therefore, this study will guide and encourage researchers to use this data and enhance the application variability.

Keywords: Headwall Hyperspec, Hyperspectral Images, Land Use maps, Random Forest Classification.