In this study, the purpose was present the possibility of systematic and efiicient vegetation survey through accuracy assessment from supervised classification using the aerial imagery and satellite images(GeoEye-1) of high resolution. There were appeared 4 species for conifes and 10 species for broadleaved trees in the field survey.
Aerial images and satellite images have high spatial resolution 0.5m, 0.12m respectively. Also, there were including each of the four bands(Band1, 2, 3, 4). However, the wavelength range of the band with a images of the both are different and spectral information(DNs: Digital Numbers) are also different. Therefore, classification was performed by changed the size of the training area, also analyzed affect classification accuracy by spatial and spectral resolution difference.
Band4 was the most useful band among the four bands through analyzed by tree species DNs characteristics. As a result, there were low accuracy with satellite and aerial images. The values were 34.7%, 27.8% by error matrix and 0.3, 0.23 without probability of casual coincidence(Kappa).
In the case of the multi-spectral images(Band1+2+3+4), training sample area was set (3*3), (5*5), (7*7) with satellite images, meanwhile, aerial image was set (5*5), (12*12), (17*17). As the result, overall accuracy were appeared 84.7%, 68.7%, 58.7% with satellite images, 81%, 70.2%, 64.2% with aerial images training sample numerical order, 0.84, 0.67, 0.56, 0.8, 0.68, 0.61 without probability of casual coincidence(Kappa) for only training samples. As a result of taking into account the distribution rario of the tree species from field survey, it was determined the overall accuracy assessment of randomly selected 200 pixel. These were 64% with satellite images and 57.5% with aerial images. Also overall accuracy were 0.61 with satellite images and 0.54 with aerial images in consideration of the Kappa coefficient. Therefore, in the case of overall accuracy was found to be aerial image is high. It is determined that the classification accuracy is high satellite image of accuracy assessment when considering only training area is compared to the aerial image.
It is considered that it is possible to obtain a classification result of high reliability when it is set smaller the size of the training area. Further, if the spatial resolution match between satellite image and aerial image, the result will be more objective and accurate.