中文摘要：

提出了一种利用可见-近红外反射光谱技术对婆婆纳、波斯婆婆纳、直立婆婆纳等3种入侵植物和本地杂草宝盖草的植物辨别方法,可以对外表相似度极高的这4种植物进行有效鉴别。研究在对光谱曲线进行预处理和聚类分析后,随机采用30×4个样本作为建模样本,其余的20×4个样本作为预测样本,应用独立软模式法SIMCA（soft independent models of class analogy）进行分类,在显著性水平为5%下,其预测分辨率为78.75%,去除婆婆纳后的预测分辨率为90%。根据变量建模能力（modeling power）值,找到敏感波段496～521,589～626和789～926nm,并将相应的波段的光谱值作为最小二乘的支持向量机LS-SVM（least squares support vector machine）的输入,进行建模预测,并以预测结果作为目标函数值,进行遗传算法GA（genetic algorithm）优化,结果发现,预测分辨率达95.35%,辨识效果好,能快速正确区分外来入侵植物。

英文摘要：

The feasibility of visible and short-wave near-infrared spectroscopy （VIS/WNIR） techniques as means for the nondestructive and fast detection of alien invasive weeds was evaluated. Selected sensitive bands were found validated. In the present study,3 kinds of alien invasive weeds,Veronica persica,Veronica polita,and Veronica arvensis Linn,and one kind of local weed,Lamiaceae amplexicaule Linn,were employed. The results showed that visible and NIR （Vis/NIR） technology could be introduced in classification of the alien invasive weeds or local weed with the similar outline. Thirty×4 weeds samples were randomly selected for the calibration set,while the remaining 20×4 samples for the prediction set. Smoothing methods of moving average and standard normal variate （SNV） were used to pretreat spectra data. Based on principal components analysis,soft independent models of class analogy （SIMCA） were applied to make the model. Four frontal principal components of each catalogues were applied as the input of SIMCA,and with a significance level of 0.05,recognition ratio of 78.75% was obtained. The average prediction result is 90% except for Veronica polita. According to the modeling power of each spectra data in SIMCA,some possible sensitive bands,496-521,589-626 and 789-926 nm,were founded. By using these possible sensitive bands as the inputs of least squares support vector machine （LS-SVM）,and setting the result of LS-SVM as the object function value of genetic algorithm （GA）,mutational rate,crossover rate and population size were set up as 0.9,0.5 and 50 respectively. Finally,recognition ratio of 95.63% was obtained. The prediction results of 95.63% indicated that the selected wavelengths reflected the main characteristics of the four weeds,which proposed a new way to accelerate the research on cataloguing alien invasive weeds.