光谱学与光谱分析
光譜學與光譜分析
광보학여광보분석
SPECTROSCOPY AND SPECTRAL ANALYSIS
2015年
8期
2154-2158
,共5页
高俊峰%张初%谢传奇%朱逢乐%郭振豪%何勇
高俊峰%張初%謝傳奇%硃逢樂%郭振豪%何勇
고준봉%장초%사전기%주봉악%곽진호%하용
高光谱图像%甘蔗%可溶性固形物%模型%纹理特征
高光譜圖像%甘蔗%可溶性固形物%模型%紋理特徵
고광보도상%감자%가용성고형물%모형%문리특정
Hyperspectral image%Sugarcane%Soluble solid content%Model%Textural features
为了探究应用近红外高光谱成像技术对甘蔗内部可溶性固形物(SSC )预测的可行性,试验样本选择三种不同品种中的240个甘蔗节作为研究对象。通过高光谱成像系统获取甘蔗节的近红外光谱信息和图像信息,并分别探讨了光谱信息和图像纹理信息对甘蔗可溶性固形物预测的可行性。采用最小二乘回归(PLSR),最小二乘支持向量机(LS‐SVM )及主成分回归(PCR)建模方法构建甘蔗可溶性固形物的预测模型。比较了连续投影算法(SPA)、无信息变量消除算法(UVE)及区间偏最小二乘(iPLS)特征提取方法对预测结果的影响。实验结果表明:基于甘蔗的光谱信息能实现可溶性固形物的预测,其中偏最小二乘回归模型的建模集和预测集的相关系数分别为0.879和0.843,均方根误差分别为0.644和0.742。通过U V E算法提取105个有效波长所建立的PLSR模型的建模集及预测集相关系数分别为0.860和0.813,均方根误差分别为0.693和0.810。
為瞭探究應用近紅外高光譜成像技術對甘蔗內部可溶性固形物(SSC )預測的可行性,試驗樣本選擇三種不同品種中的240箇甘蔗節作為研究對象。通過高光譜成像繫統穫取甘蔗節的近紅外光譜信息和圖像信息,併分彆探討瞭光譜信息和圖像紋理信息對甘蔗可溶性固形物預測的可行性。採用最小二乘迴歸(PLSR),最小二乘支持嚮量機(LS‐SVM )及主成分迴歸(PCR)建模方法構建甘蔗可溶性固形物的預測模型。比較瞭連續投影算法(SPA)、無信息變量消除算法(UVE)及區間偏最小二乘(iPLS)特徵提取方法對預測結果的影響。實驗結果錶明:基于甘蔗的光譜信息能實現可溶性固形物的預測,其中偏最小二乘迴歸模型的建模集和預測集的相關繫數分彆為0.879和0.843,均方根誤差分彆為0.644和0.742。通過U V E算法提取105箇有效波長所建立的PLSR模型的建模集及預測集相關繫數分彆為0.860和0.813,均方根誤差分彆為0.693和0.810。
위료탐구응용근홍외고광보성상기술대감자내부가용성고형물(SSC )예측적가행성,시험양본선택삼충불동품충중적240개감자절작위연구대상。통과고광보성상계통획취감자절적근홍외광보신식화도상신식,병분별탐토료광보신식화도상문리신식대감자가용성고형물예측적가행성。채용최소이승회귀(PLSR),최소이승지지향량궤(LS‐SVM )급주성분회귀(PCR)건모방법구건감자가용성고형물적예측모형。비교료련속투영산법(SPA)、무신식변량소제산법(UVE)급구간편최소이승(iPLS)특정제취방법대예측결과적영향。실험결과표명:기우감자적광보신식능실현가용성고형물적예측,기중편최소이승회귀모형적건모집화예측집적상관계수분별위0.879화0.843,균방근오차분별위0.644화0.742。통과U V E산법제취105개유효파장소건립적PLSR모형적건모집급예측집상관계수분별위0.860화0.813,균방근오차분별위0.693화0.810。
In order to explore the feasibility of prediction soluble solid contents (SSC) in sugarcane stalks by using near infrared hyperspectral imaging techniques ,two hundred and forty sugarcane stalks which come from three different varieties were stud‐ied .After obtaining the raw hyperspectral images of sugarcane stalks ,the spectral information and textural features were dis‐cussed respectively .The prediction models were established by using partial least squares regression (PLSR) ,principal compo‐nents regression (PCR) and least squares support vector machines (LS‐SVM ) algorithms .Besides ,three different selected wavelengths algorithms such as successive projection (SPA ) algorithms ,intervals partial least squares (iPLS ) algorithms and uninformation variables elimination (UVE) algorithm were analyzed after building partial least squares regression model .The re‐sults indicate that partial least squares regression model based on spectral features can be an steady model to predict SSC and the correlation coefficient (R2 ) of calibration sets and prediction sets are 0.879 ,0.843 .The root mean square errors of calibration sets and prediction sets are 0.644 ,0.742 respectively .The obtained 105 wavelengths which were selected by UVE algorithm are effective spectral features .The R2 results of calibration sets and prediction sets of its PLSR model are 0.860 ,0.813 .The root mean square errors of calibration sets and prediction sets are 0.693 ,0.810 respectively .