论文部分内容阅读
利用高分辨率光谱仪在实地测得的光谱数据来识别美国加州的６种主要针叶树种。树冠阴面和阳面的高光谱数据分别在１９９６年夏、秋测得。首先对原始光谱数据作简单处理，然后进行６种数据变换：对数变换、一阶微分变换、对数变换后一阶微分变换、归一化变换、归一化变换后一阶微分变换及归一化后对数变换。采用相邻窄波段逐步加宽的办法，测试不同波段宽度对树种识别精度的影响。所有的变换方法及波段宽度试验最后均由神经元网络算法产生的树种分类精度来评价。试验结果表明对数变换后一阶微分和归一化变换后一阶微分能够获得高于９４％的平均精度；归一化变换和微分处理能够限制阴影的影响；２０ｎｍ的波段宽度用于识别此６种针叶树种是较为理想的。我们发现太阳高度角变化对树种识别影响不大。
Six major conifers in California, USA, are identified using spectroscopic data measured on a high-resolution spectrometer. The canopy shade and sun hyperspectral data were measured in the summer of 1996 and autumn respectively. First of all, the original spectral data are simply processed and then six kinds of data transformations are performed: logarithmic transformation, first-order differential transformation, first-order differential transformation after logarithmic transformation, normalized transformation, first-order differential transformation after normalized transformation and After a change of logarithm. The effect of width of different bands on the recognition accuracy of tree species was tested by gradually widening the adjacent narrow bands. All transformation methods and band width experiments are finally evaluated by the classification accuracy of the tree species generated by the neural network algorithm. The experimental results show that the first-order differential after logarithmic transformation and the first-order differential after normalized transformation can achieve an average precision higher than 94%; the normalized transformation and differential processing can limit the influence of shadows; the 20 nm band width is used to identify this 6 kinds of conifer species is more ideal. We found that the change of solar altitude has little effect on tree species identification.