地物光谱不确定性及光谱角制图算法适用性研究

地物光谱不确定往往使同种地物光谱之间存在一定程度的差异，影响了地物的识别精度，对光谱角制图算法的地物识别效果也会产生一定的影响。光谱角制图算法(spectral angle mapper，SAM) 是基于光谱曲线整体相似性的一种算法，在高光谱遥感信息分类中应用广泛，但在计算两条地物光谱曲线的相似性时并没有考虑地物光谱不确定性的影响，因此往往会不能正确识别出目标地物。针对地物光谱的不确定性，研究了光谱角制图算法的适用性，并对光谱角制图算法进行了改进。改进的基本思路为：设置测试光谱与参考光谱各波段的光谱差异量为一相同值，并根据同种地物光谱向量之间夹角最小的原则，利用求导的方法求出光谱差异量，以克服地物光谱不确定性的影响。为了验证改进效果，利用USGS的五种高岭石标准光谱，在考虑地物光谱不确定性的情况下，分别选择局部波段和全局波段计算高岭石标准光谱之间的光谱角，并对光谱角计算结果和光谱角制图算法的适用性进行了分析。通过USGS标准矿物光谱数据的实验证明：改进的光谱角度制图算法利用光谱差异量可以有效表征同种地物光谱的差异，能够克服地物光谱不确定性的影响并提高地物识别的精度，对地物光谱不确定性具有更好的适用性，并对基本符合光谱差异向量各维值相等的局部波段组合具有更好的效果。

Spectral Uncertainty of Terrestrial Objects and the Applicability of Spectral Angle Mapper Algorithm

The spectral uncertainty of terrestrial objects causes a certain degree of spectral differences among feature spectra, which affects the accuracy of object recognition and also impacts the object recognition of spectral angle mapper algorithm (SAM). The spectral angle mapper algorithm is based on the overall similarity of the spectral curves, which was widely used in the classification of hyperspectral remotely sensed information. The spectral angle mapper algorithm does not take the spectral uncertainty of terrestrial objects into account while calculating the spectral angle between the spectral curves, and therefore does not tend to correctly identify the target objects. The applicability of the spectral angle mapper algorithm is studied for the spectral uncertainty of terrestrial objects and a modified SAM is proposed in this paper．In order to overcome the influence of the spectral uncertainty，the basic idea is to set a spectral difference value for the test spectra and the reference spectra and to calculate the spectral difference value based on derivation method according to the principle of minimum angle between the test spectra and the reference spectra. By considering the impact of the spectral uncertainty of terrestrial objects, this paper uses five kaolinite mineral spectra of USGS to calculate the spectral angle between the five kalinite mineral spectra by using local band combination and all bands to verify the improved algorithm. The calculation results and the applicability of the spectral angle mapper algorithm were analyzed. The results obtained from the experiments based on USGS mineral spectral data indicate that the modified SAM is not only helpful in characterizing and overcoming the impact of the spectral uncertainty but it can also improve the accuracy of object recognition to certain extent especially for selecting local band combination and has better applicability for the spectral uncertainty of terrestrial objects.