CAJ | 학술논문

在进行地物分类、目标识别等应用研究之前，一般都要在高维数据中（例如高光谱图像数据）对典型的地物进行光谱分析、特征波段的提取，以便提取出最佳的有效信息即异常特征，剔除无用的干扰信息。本文利用小波变换具有空频局部化和多分辨率的特点，在小波变换域分析并提取光谱曲线。在小波变换各个高频分量中利用相邻的正负极值点找到原始光谱曲线上每个吸收带的左右边界；利用局部过零点，就可以比较准确的提取出各个吸收带的中心波长，通过定量频率计算可以得到分析目标吸收带的最佳小波分解层数。该方法与传统的光谱特征提取方法相比更加简单，有效，进一步实验也证明该方法是比较理想的光谱特征提取方法。
재진행지물분류、목표식별등응용연구지전，일반도요재고유수거중（례여고광보도상수거）대전형적지물진행광보분석、특정파단적제취，이편제취출최가적유효신식즉이상특정，척제무용적간우신식。본문이용소파변환구유공빈국부화화다분변솔적특점，재소파변환역분석병제취광보곡선。재소파변환각개고빈분량중이용상린적정부겁치점조도원시광보곡선상매개흡수대적좌우변계；이용국부과영점，취가이비교준학적제취출각개흡수대적중심파장，통과정량빈솔계산가이득도분석목표흡수대적최가소파분해층수。해방법여전통적광보특정제취방법상비경가간단，유효，진일보실험야증명해방법시비교이상적광보특정제취방법。
Band Selection and spectral feature analysis from a dataset with high dimension is a common step prior to some typical hyperspectral image application, such as classification, target detection. Such works are usually trying to not only preserve opitmal information but remove disturbance features. By exploiting characteristic of wavelet transforming, we proposed a method to extract anomaly features in hyperspectral image. In our method, both boundary and central wavelength of absorption peaks are determinated by wavelet coefficients. Adjoin extreme value with different sign in high frequency sub-band in wavelet coefficients are used to calculate lower and upper boundary;and zero-point of those extreme value is considered as position of central wavelength. Based on those features, we conduct optimal layers needed to decompose during wavelet transforming. Experiment results show that the proposed approach is simpler and effective than others.