Classification of coniferous tree species using aerial hyper spectral observation

Aerial data of hyper spectral image are acquired using AISA system on 11th August, 2007 and the data have 190 bands from visible to short wave infrared. Using the data, we would like to make high accurate classification map of tree species. We study about an effectiveness of hyper spectral image data that have high resolution of wavelengths. The survey area is located at the foot of mountain. The place is consisted of cedar, larch, pine, natural broadleaf tree, bare soil and grass field. We perform the processing by using the free software of "Multi Spec" that is suitable for conducting hyper spectral data. For processing of classification, we must decrease 190 bands to suitable amount of bands. The declining method of bands is processed by two methods. One is Feature Extraction Method that is the extraction using separation level and another is a method that selects the resemble bands of an existed sensor of satellite. Using spectrum reflect characteristic on wavelengths of the bands, we evaluate the two declining methods. The processing of classification of tree species is a supervised method using suitable bands and field survey data for the training data. We compare about the classification maps. We classify mainly coniferous tree species and the accuracy of the classification map is very high. Suitable bands by Feature Extraction are picked a band from wavelengths of red and two bands from each wavelengths of near infrared red, short wave infrared red and longer near infrared red. In wavelength that selects suitable bands by Feature Extraction, there is difference about reflection strength of each object. Using aerial hyper spectral data, we make a classification map of tree species with high accuracy.