光谱学与光谱分析, 2019, 39 (9): 2862, 网络出版: 2019-09-28  

多分类器融合提取土壤养分特征波长

Extracting Characteristic Wavelength of Soil Nutrients Based on Multi-Classifier Fusion
李雪莹 1,2,3,*范萍萍 1,2,3刘岩 1,2,3王茜 1,2,3吕美蓉 1,2,3
作者单位
1 齐鲁工业大学(山东省科学院), 山东省科学院海洋仪器仪表研究所, 山东 青岛 266061
2 山东省海洋环境监测技术重点实验室, 山东 青岛 266061
3 国家海洋监测设备工程技术研究中心, 山东 青岛 266061
摘要
光谱已经应用于土壤养分速测的分析, 但是如何寻找土壤光谱特征波段, 尽最大可能避免无用信息干扰、 保留有用信息, 建立准确度高、 预测效果好的模型仍是一个亟需解决的问题。 以青岛三个不同地区土壤样品为例, 测定土壤样品的紫外-可见-近红外光谱及其总碳(TC)、 总氮(TN)、 总磷(TP)含量; 分别采用连续投影算法(SPA)、 无信息变量消除法(UVE)、 遗传算法(GA)、 相关系数法(CC)四种算法(四种单分类器)对土壤光谱提取特征波长; 再引入投票法和加权投票法的多分类器融合方法将四种算法融合得到特征波长; 以偏最小二乘回归(PLSR)建立各土壤养分含量的模型, 通过对模型效果的评价标准(建模集绝对系数R2c、 校正均方根误差RMSEC、 检验集绝对系数R2p、 预测均方根误差RMSEP和相对分析误差RPD值)来判别各单分类器算法和多分类器融合算法对土壤养分含量特征波长的提取效果。 分别对四种算法、 筛选其中三种算法、 最优二种算法进行融合, 分析融合后模型效果和特征波长个数, 结果表明: 将四种单分类器经投票法融合后, 其模型效果大部分不如单分类器, 且相对好的模型特征波长个数较多; 相较于投票法多分类器融合, 四种单分类器经加权投票法融合模型效果有了一定的提高, TC和TN都能够在较少的波长中获得较好的预测效果, 但仅TN经融合后, 模型效果优于每个单分类器; TC, TN, TP分别在取SPA+UVE+GA, SPA+UVE+GA(或SPA+GA+CC)、 SPA+UVE+GA三种单分类器进行加权投票法融合后, 均能获得最优模型效果, 且明显优于每个单分类器, 模型效果有了显著提高; 各土壤养分含量经两个最优单分类器加权投票法融合后, 仍能得到好于最优单分类器的建模效果, TC和TP建模效果略差于三个单分类器融合结果, TN建模效果与三个单分类器融合结果相同。 因此, 在筛选三种算法融合, 且其中包含最优两种算法的情况下, 能够以较少的特征波长个数获得明显高于单分类器的建模效果。 该方法为寻找土壤养分以及其他复杂物质成分的光谱特征波段提供了新方法, 也为多种算法的综合运用提供了新思路。
Abstract
Although spectral technology has been applied to the rapid detection of soil nutrient, how to find the spectral characteristic bands of soil, to avoid useless information and to keep useful information, and to establish a model with high accuracy and good predictive effect is still an urgent problem to be solved. Taking soil samples from three different regions in Qingdao as an example, the ultraviolet-visible-near-infrared spectra and total carbon (TC), total nitrogen (TN) and total phosphorus (TP) content of soil samples were determined. Successive Projections Algorithm (SPA), Uninformative Variable Elimination (UVE), Genetic Algorithm (GA) and Correlation Coefficient Method (CC) four kinds of algorithms (four single classifiers) were used to extract the characteristic wavelength of the soil spectra. The multi-classifier fusion of the voting method and the weighted voting method were used to obtain the characteristic wavelength. The soil nutrient content models were established by the partial least squares regression (PLSR). Through theresult of these models (the determination coefficient of calibration set R2c, the corrected root mean square error RMSEC, the determination coefficient of test set R2p, the predicted root mean square error RMSEP and residual predictive deviation RPD), we evaluated the effect of extracting the characteristic wavelength of soil nutrient content among each single classifier algorithm and multiple-classifier fusion algorithm. In this paper, the multi-classifier fusion of four algorithms, three algorithms and optimal two algorithms were analyzed. The results showed that, after merging four kinds of single classifier by voting method, the model effect was mostly inferior to each single classifier, and there were many characteristic wavelengths in the relative good model. The model effect of four single classifier by weighted voting method had been improved compared with that by voting method. TC and TN could achieve better prediction effect in less wavelength, but only after TN fusion, the model effect was better than each single classifier. TC, TN and TP were fused by weighted voting method with SPA+UVE+GA, SPA+UVE+GA (or SPA+GA+CC) and SPA+UVE+GA three kinds of single classifier, and the optimal model effect was obtained, which was superior to each single classifier. The soil nutrient content was fused by weighted voting method with two optimal single classifier, the modeling effect was better than that of the optimal single classifier, the results of TC and TP modeling were slightly worse than those of three single classifiers, and TN modeling effect was the same as that of three single classifiers. So TC, TN and TP could obtain higher results than single classifier in case of selecting three kinds of algorithms and including the optimal two algorithms. It provides a new method for finding spectral characteristic bands of soil nutrients and other complex substances, and also provides a new idea for the comprehensive application of various algorithms.

李雪莹, 范萍萍, 刘岩, 王茜, 吕美蓉. 多分类器融合提取土壤养分特征波长[J]. 光谱学与光谱分析, 2019, 39(9): 2862. LI Xue-ying, FAN Ping-ping, LIU Yan, WANG Qian, L Mei-rong. Extracting Characteristic Wavelength of Soil Nutrients Based on Multi-Classifier Fusion[J]. Spectroscopy and Spectral Analysis, 2019, 39(9): 2862.

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!