This study was conducted to improve the accuracy of classification by applying the spectral library of hyperspectral images to the Land Cover Classification using the wavelength range of 380 to 2400 nm. It was judged that a spectral library built to meet the current situation in Korea is less likely to be used as a spectral library that is either manufactured indoors or does not contain SWIR (Short Wave Infra-Red) wavelength areas. To improve this, hyperspectral images including SWIR wavelengths were utilized and the spectral library was constructed based on actual field acquired spectral data. In addition, the accuracy assessment of the spectral library itself was conducted through the calibration process of the hyperspectral image and the spectral library, and 92.9% of the classification of high accuracy was obtained from the Land Cover Classification. The above results have the following process.

First, in the process of building a ground-spectral library, the reflectivity is corrected by wavelengths that are close to 100% reflectivity by utilizing White Reference (WR) when data is acquired. Next, Band Matching is performed to match data from ground spectrometer with 1nm of spectral resolution to hyperspectral image. The statistical interpolation was used to match the data from ground spectrometer with 2,151 bands to the number of bands in hyperspectral images with 448 bands. And if the WR calibration task is not close to 100% reflectivity, or the wavelength of a spark-like band is considered noise and is performed by removing the wavelength area. Complete the above tasks for each class to build a ground spectral library for each class.

Second, for the accuracy evaluation of the ground spectral library, a image spectral library of hyperspectral images containing the SWIR spectroscopy area was produced to perform the correlation assessment. If the ground spectral library is calibrated to match the band number and width of the hyperspectral image before proceeding, the image library must also remove the noise wavelength area of the ground spectral library to assess the correlation. In addition, even if the pre-treatment of hyperspectral aerial images before noise wavelength removal is performed, sparking of wavelength curves may occur depending on weather or humidity at that time. To address this, it is necessary to utilize the "spectral smoothing" function to correct the correlation assessment with the ground spectral library. Finally, the video library was conducted based on the GPS location data obtained from the pre-site survey by class to conduct a correlation assessment with the ground spectral library. As a result, the coefficient of determination for all classes showed a high correlation between 0.84 and 0.99. Due to this, it is believed that accuracy and objectivity of spectral libraries that are constructed are high.

Third, it was carried out by utilizing a spectral library that was built on the Land Cover Classification of hyperspectral images that included actual SWIR spectral areas. While Land Cover Classification was conducted, A spectral library that built a training sample was used and the Reference data selection was selected based on field survey and GPS location data. As a result, the kappa coefficient is 0.92, which indicates a higher accuracy than the 0.89 standard for the Land Cover Classification, which makes the land cover classification good. Also, given the high number of producer and user accuracy for each class, the Land Cover Classification can also have high confidence.