A study on the classification of urban region using Hyper-spectrum data at AVIRIS

Abstract

The purpose of this study is to improve accuracy of land cover classification in urban area. This study used Quick bird and airborne hyper spectrum sensor AVIRIS. Land cover survey was attempted at residential area. There were houses, road, pond, park and vegetations. There were many kinds of vegetations, broad leave trees as Ash, Elm, Willow, Maple, Cotton tree, Needle leave trees as Pine, Tsuga, Spruce. In the study in the past, about Quick Bird analysis, supervised classification (Maximum likelihood algorithm) was executed with some supervisors obtained from field study. 6categories were classified. The results show that with Quick Bird analysis, distribution of vegetation is comprehended well, but timber species are not comprehended well. About AVIRIS analysis, SAM (Spectral Angle Mapper classification) was executed with some supervisors obtained from field study. About broad leave trees, there were five species which spectrums were classified to categories. About grass, 3 categories were classified with conditions of land coverage condition. The accuracies of classification were 26% to 84%. It is confirmed that better results were obtained with AVIRIS. In this study, the classification of the material was tried. There were many kinds of roof materials at residential area, for example, Onix black, Shasta white, Desert tan, Siera gray, Terra cotta, Tile, Wood, Concrete, Asphalt. About AVIRIS analysis using measured spectrums, 7categories were classified. Target area was 2residential areas and Site of university. Target area size was 360m by 360m. The accuracies of classification were 44.4% to 100%. It is confirmed that better results were obtained with AVIRIS.

Vegetation Cover Estimation Using Convolutional Neural Networks

“…in 10-year intervals [3], especially for the purpose of city planning to grasp land cover and vegetation in urban areas. Recently, remote-sensing techniques [10] have shown to be powerful in understanding land and vegetation covering in the territories, in addition to monitoring climate changes. Significant research efforts have been devoted to develop techniques for mapping and estimating forest cover and carbon emissions using remote sensing technology over large scales.…”