CAJ | 학술논문

对光谱角制图分类法进行改进,主要方法是在原影像的6个波段基础上,加入了K-L变换的前3个波段、K-T变换的3个波段(亮度、绿度和湿度)以及NDVI(归一化植被指数)信息将图像波段数增加至13个,增强了地物之间的光谱区分度(即类间方差);得到光谱角度图像后,提出一种无需阈值的地类确定方法,减少了人工参与,提高了判别精度和处理效率.
대광보각제도분류법진행개진,주요방법시재원영상적6개파단기출상,가입료K-L변환적전3개파단、K-T변환적3개파단(량도、록도화습도)이급NDVI(귀일화식피지수)신식장도상파단수증가지13개,증강료지물지간적광보구분도(즉류간방차);득도광보각도도상후,제출일충무수역치적지류학정방법,감소료인공삼여,제고료판별정도화처리효솔.
Land use/cover change has been the main factor of environmental change.Analyzing the LUCC of Yellow River Delta plays a great role in its sustainable development.With the aid of remote sensing technology and remote sensing images,this paper aims to build region classification map by discriminanting ground features' categories using SAM.The principle of SAM is to confirm each type of surface features' attribution by comparing the angle formed by the surface features vector and reference vector features in the spectrum space.The traditional SAM only used the angle parameter,which could attain higher classification accuracy only when the variance within the classes of pixels to be recognized was small and the between-class variance is large.In consideration of its shortage,the improved SAM was used,for this method the image bands were added to 13 based on the original 6 bands by K-L transformation and K-L transformation and NDVI,which enhanced the spectral differentiation between ground features,and got spectral curves of all kinds of ground features.Several bands had obvious difference in the curve graph,which had active effect on feature discrimination.And then it used a processing mode which did not set the threshold value under ERDAS platform;after the processing, it got a series of spectral angle images,then judged the similarity degree of each pixel with the reference classes.The pixel class should be vested in the class where the largest pixel value existed,the rest could be done in the same manner.In the last,the classification map of the study area was attained.This method reduced the manual involvement and improved the discrimination accuracy and treatment efficiency greatly.