| 可见/近红外光谱技术识别树叶树种的研究 |
| |
| 引用本文： | 汪紫阳,尹世逵,李春旭,李耀翔.可见/近红外光谱技术识别树叶树种的研究[J].西北林学院学报,2019(1):229-236,260. |
| |
| 作者姓名： | 汪紫阳 尹世逵 李春旭 李耀翔 |
| |
| 作者单位： | 东北林业大学 |
| |
| 基金项目： | 林业公益性行业科研专项(201504508);十三五国家重点研发计划项目(2017YFC0504103) |
| |
| 摘 要： | 探索使用可见/近红外光谱技术识别树叶树种的可行性,为野外可见/近红外光谱技术用于树种识别提供方法。本试验识别了9个树种,测试了光谱预处理方法、识别方法对可见/近红外光谱识别的准确率的影响。对9种阔叶树种共46棵树,分别采用距离法和PLS-DA建立识别模型,比较不同波段和导数预处理方法对模型预测效果的影响。结果表明,使用距离法对原始光谱进行识别时,识别准确率<50%,不能够有效识别树叶树种。使用距离法对预处理后的光谱进行识别时,识别准确率为近红外350～2 500nm(99.16%)>350～1 000nm(88.05%)>1 000～2 500nm(81.24%),且任意单个树种的识别准确率都>98%,能够有效识别树叶树种。使用偏最小二乘法(PLS-DA)结合单列识别变量矩阵时,识别准确率高达100%,识别模型的相关系数为0.993 6,RMSEC为0.120,RMSEP为0.144,但只能成功识别4种树叶树种,当树叶种数>4时,预测模型的识别准确率陡降。使用偏最小二乘法(PLS-DA)结合多列识别变量矩阵对9种树叶的识别准确率高达99.58%,识别模型的相关系数为0.888 6～0.956 9,RMSEC为0.084 5～0.15,RMSEP为0.088 7～0.155。本试验为可见/近红外光谱技术快速识别树种提供了一种新的方法和思路。
|
| 关 键 词： | 可见/近红外光谱 树种识别 树叶 偏最小二乘法 |
Identification of Tree Leaf and Species by Vis/NIR Spectroscopy |
| |
| WANG Zi-yang,YIN Shi-kui,LI Chun-xu,LI Yao-xiang.Identification of Tree Leaf and Species by Vis/NIR Spectroscopy[J].Journal of Northwest Forestry University,2019(1):229-236,260. |
| |
| Authors: | WANG Zi-yang YIN Shi-kui LI Chun-xu LI Yao-xiang |
| |
| Institution: | (Northeast Forestry University,Harbin 150040,Heilongjiang,China) |
| |
| Abstract: | A feasibility study was carried out to identify tree leaf and species by visible and near infrared spectroscopy (Vis/NIRS) technique to provide a practical method for the identification of tree species in the field.Forty-six broad-leaved trees of 9 species were sampled to examine the influences of different spectral pretreatments and identification methods on the rate of identification accuracy were tested.Different wavelengths and identification methods (distance method and PLS-DA) were compared over the effect of model prediction.The results showed that by using distance method to identify the raw spectrum,the rates of identification accuracy of all species were under 50%,which could not effectively identify the tree species.However,when the distance method was used to identify the pretreated spectra,the rates of identification accuracy were 99.16%(350-2 500 nm ),88.05%(350-1 000 nm) and 81.24%(1 000-2 500 nm),respectively.With the wavelength of 350-2 500 nm,the accuracy rates for all the individual tree species achieved over 98%.When the PLS-DA method combined with single column identification variable matrix was used,the rates of identification accuracy were 100%,with the correlation coefficient of 0.993 6,RMSEC 0.120,and RMSEP 0.144.With this method,the maximum number of identifying tree species was 4,over 4,the rate of identification accuracy decreased significantly.When the PLS-DA method combined with multiple column identification variable matrix was used to identify the 1 st and smoothing spectrum of 9 species,the rate of identification accuracy was 99.58%,with correlation coefficients of 0.888 6-0.956 9,RMSEC 0.084 5 -0.15,and RMSEP 0.088 7-0.155.The results of this study would provide a new method and way in rapid identification of common tree species. |
| |
| Keywords: | Vis/NIR spectroscopy Tree identification Leaf PLS-DA |
| 本文献已被 CNKI 维普 等数据库收录！ |
|
