基于色素含量的针叶树种敏感波段提取研究

Sensitive band range extraction research for coniferous species based on plant pigment content

通过分析马尾松、杉木主要色素和冠层光谱数据的相关关系提取敏感波段,然后利用7种分类算法对所提取波段进行分类,最后对高斯合并后的光谱数据进行分类,用以测试提取波段的可推广性。结果表明:马尾松和杉木的差异主要是受叶绿素的影响,并且2种针叶树种的敏感波段位于401～504 nm和659～686 nm;用于区分2种针叶树种高光谱数据的最佳分类方法为Fisher分类法,最高分类精度达到了100%;模拟成像光谱数据的高斯合并数据抑制了高频噪声,但也过滤掉了2种针叶树种光谱数据的细微差异,分类精度降低,为70%～80%,而叶绿素所提取波段仍然优于其它色素提取的波段,这说明401～504 nm和659～686 nm波段具有可推广和进一步研究的价值。

Through analyzing the relation between the spectral reflectance of canopy and the pigment content, the sensitive band ranges of Cunninghamia lanceolata and Pinus massoniana were extracted. Then the data of band ranges selected were classified by seven classification algorithms including Support Vector Machine （SVM）-Radial Basis Function （RBF）, BP neural network, Mahalanobis Distance, Bayes, Fisher, Support Vector Machine （SVM）-Linear, and Spectral Angle Mapping （SAM）. In order to test the dependability and popularization of bands selected, the data after Gauss merge process were classified. The results show that the difference of C. lanceolata and P massoniana was largely influenced by chlorophyll. The sensitive band ranges for two conifers situated at 401~504 nrn and 659-686 nm. By comparing seven methods, Fisher classification method have best performance, their maximum precision of classification were 100%. The data after Gauss merge process that modeled the imaging spectrometer data suppressed the high-frequency noise impact, but the subtle differences of two conifers were filtered out, so the precision of classification came down to 70% - 80%. The performance of chlorophyll could be better than other pigment. It is proved that the band ranges of 401-504 nm and 659-686 nm had good ~eneralizabilitv and further research value.