单 位：
吉林大学生物与农业工程学院;吉林农业大学工程技术学院
关键词：
叶绿素荧光光谱;光能利用效率;主成分分析;支持向量机
摘 要：
基于激光诱导叶绿素荧光光谱分析技术,提出了一种基于支持向量机理论的光能利用效率预测方法。同步采集黄瓜叶片的叶绿素荧光光谱、净光合速率和光合有效辐射,选取500~800 nm波段的叶绿素荧光光谱作为研究对象,首先对原光谱进行SG-FDT预处理;其次对预处理的光谱采用PCA方法提取特征值,根据累计贡献率选取前10个主成分代替原光谱信息;最后通过支持向量机建立光能利用效率预测模型。通过对惩罚系数C和核函数参数g的大量测试,最终确定C为0.031 25、g为1,并利用60个训练样本对模型进行训练。10个测试样本的预测结果表明,测试样本的平均误差为8.94%,具有很好的预测效果。
译 名：
Predicting Light Use Efficiency with Chlorophyll Fluorescence Spectra Based on SVM
作 者：
Ren Shun;Yu Haiye;Zhou Li'na;School of Biological and Agricultural Engineering,Jilin University;College of Engineering and Technology,Jilin Agricultural University;
关键词：
Chlorophyll fluorescence spectra Light use efficiency Principal component analysis Support vector machine
摘 要：
Light use efficiency is an important parameter of plant productivity model.It is an evaluation index for plant to turn the solar energy into dry matter efficiency.Taking cucumbers as the study objects,a method for light use efficiency prediction was proposed with the help of analysis technique of laserinduced chlorophyll fluorescence spectra based on the theory of support vector machine(SVM).Chlorophyll fluorescence spectra,net photosynthetic rate and photosynthetic active radiation of cucumber leaves were synchronously acquired,and the 500 ~ 800 nm band of chlorophyll fluorescence spectrum was selected as study objects.Firstly,the original spectra was pretreated by SG-FDT method.Secondly,the characteristic values of pretreated spectra were extracted by using principal component analysis(PCA)method,the first ten principal components whose cumulative contribution rate was 93.49% were selected instead of the original spectral information in the study.Finally,the prediction model of light use efficiency was established through the SVM with the radial basis function.The penalty parameter C and kernel function parameter g were ultimately determined as C = 0.031 25,g = 1 by carrying out a large number of tests,and then 60 training samples were combined to train the model.Ten testing samples were used to test the established model,and the results showed that the average error of the testing samples was 8.94%,which indicated a good predictive power.
