Application of Short Wave Infrared Hyperspectral Airborne Image Library for Quality Improvement of Land Cover Classification

Recent research studies on land cover classification using hyperspectral imagery have focused on diversifying classification classes of land cover as well as improving the accuracy of classification by using the spectral information of each pixel. Conventional hyperspectral images contain wavelengths up to the visible near-infrared (VNIR) wavelength range or low-spatial -resolution images. This has made it difficult to obtain the various types of information of each pixel during land cover classification, and each class could not be assigned individual characteristics owing to the low level of distinction in the information of each pixel. To address this issue, in this study, we acquired images by airborne hyperspectral imaging, with the aim of improving the accuracy of land cover classification by using hyperspectral imagery with a high spatial resolution and spectral resolution resulting from having a wavelength range of 380-2400 nm, which includes the short wave infrared (SWIR) wavelength range. In addition, a spectral library was set up as a means to perform land cover classification using hyperspectral images, and a correlation analysis was carried out to assess the objectivity and accuracy of the spectral library. Moreover, the spectral library was used as a training sample during land cover classification, thereby enhancing accuracy. To assess the accuracy of the spectral library that was set up, the correlation between the spectral library in question with the image spectral library of hyperspectral images was analyzed, and the results showed a high correlation between 0.81 and 0.99. In the process of constructing the spectral library, the spectral library was corrected to the extent that the information in the hyperspectral image would not be lost, and it was constructed in a manner that would increase the accuracy of land cover classification. As a result of applying the finally constructed spectral library to the hyperspectral image land cover classification, a high classification accuracy of 92.9% was obtained.