光学精密工程
光學精密工程
광학정밀공정
OPTICS AND PRECISION ENGINEERING
2010年
3期
586-592
,共7页
黄富荣%潘涛%张甘霖%潘贤章%刘登飞
黃富榮%潘濤%張甘霖%潘賢章%劉登飛
황부영%반도%장감림%반현장%류등비
土壤%锌%近红外漫反射光谱%偏最小二乘法%波段选择
土壤%鋅%近紅外漫反射光譜%偏最小二乘法%波段選擇
토양%자%근홍외만반사광보%편최소이승법%파단선택
soil%Zn%near-infrared diffuse reflectance spectroscopy%partial least square method%wave band choice
采用近红外漫反射光谱和偏最小二乘法(PLS)建立了土壤锌快速分析的定量模型,并进行了波段优选.首先,基于单波长模型预测效果将全体样品划分为定标集和预测集;然后,采用多元散射校正(MSC)和Savitzky-Golay(SG)平滑方法对光谱进行预处理.选取全谱400~2 500 nm,400~1 100 nm,1 100~1 900 nm,1 900~2 500 nm,580~900 nm等5个波段,每个波段分别采用原谱、一阶导数谱、二阶导数谱,共建立了15个定标模型.同时调整SG平滑点数和PLS因子数,每个模型分别进行PLS数值实验,按照预测效果进行优选.结果显示,采用1 900~2 500 nm波段一阶导数谱的模型效果最好,预测相关系数(RP)、RMSEP、RRMSEP分别为0.806,31.0 mg/kg和19.96%.这些结果表明,1 900~2500 nm波段可以代替全谱波段得到更好的预测效果,可为设计专用土壤近红外光谱仪提供依据.
採用近紅外漫反射光譜和偏最小二乘法(PLS)建立瞭土壤鋅快速分析的定量模型,併進行瞭波段優選.首先,基于單波長模型預測效果將全體樣品劃分為定標集和預測集;然後,採用多元散射校正(MSC)和Savitzky-Golay(SG)平滑方法對光譜進行預處理.選取全譜400~2 500 nm,400~1 100 nm,1 100~1 900 nm,1 900~2 500 nm,580~900 nm等5箇波段,每箇波段分彆採用原譜、一階導數譜、二階導數譜,共建立瞭15箇定標模型.同時調整SG平滑點數和PLS因子數,每箇模型分彆進行PLS數值實驗,按照預測效果進行優選.結果顯示,採用1 900~2 500 nm波段一階導數譜的模型效果最好,預測相關繫數(RP)、RMSEP、RRMSEP分彆為0.806,31.0 mg/kg和19.96%.這些結果錶明,1 900~2500 nm波段可以代替全譜波段得到更好的預測效果,可為設計專用土壤近紅外光譜儀提供依據.
채용근홍외만반사광보화편최소이승법(PLS)건립료토양자쾌속분석적정량모형,병진행료파단우선.수선,기우단파장모형예측효과장전체양품화분위정표집화예측집;연후,채용다원산사교정(MSC)화Savitzky-Golay(SG)평활방법대광보진행예처리.선취전보400~2 500 nm,400~1 100 nm,1 100~1 900 nm,1 900~2 500 nm,580~900 nm등5개파단,매개파단분별채용원보、일계도수보、이계도수보,공건립료15개정표모형.동시조정SG평활점수화PLS인자수,매개모형분별진행PLS수치실험,안조예측효과진행우선.결과현시,채용1 900~2 500 nm파단일계도수보적모형효과최호,예측상관계수(RP)、RMSEP、RRMSEP분별위0.806,31.0 mg/kg화19.96%.저사결과표명,1 900~2500 nm파단가이대체전보파단득도경호적예측효과,가위설계전용토양근홍외광보의제공의거.
The quantification model of zinc rapid analysis in the soils was established by using near-infrared diffuse reflectance spectroscopy and a Partial Least Square (PLS) method, and the optimal measured wave band was also selected. All the samples were divided into calibration set and prediction set based on the prediction effect of single wavelength model, and then the spectra were pretreated by Multiplicative Scatter Correction (MSC) and Savitzky-Golay (SG) smoothing methods. The following five wave bands, the whole region 400-2 500 nm,400-1 100 nm,1 100-1 900 nm,1 900-2 500 nm,580-900 nm were selected, and 15 calibration models were constructed for each band selected by the original spectra, the first derivative spectra and the second derivative spectra respectively. By simultaneous adjusting the numbers of SG smoothing points and PLS factors,the PLS computational experiments for each model were carried out and the best model was selected according to the prediction effect. The results indicate that the prediction effect of the model using the first derivative spectra in 1 900-2 500 nm is the best, and its prediction correlation coefficient (RP), RMSEP and RRMSEP are 0.806, 31.0 mg/kg, and 19.96% respectively. Obtained results show that 1 900-2 500 nm can replace the whole band to get a better calibration effect, and can provide a basis for the design of special soil near-infrared spectroscopy instruments.