光谱采集位置和数量对番石榴无损检测的影响

基于漫反射法的番石榴可溶性固形物含量（SSC）无损检测中,确定番石榴最佳光谱取样的位置和数量,对提高检测精度具有重要意义。分别采集番石榴顶端、赤道和底部区域漫反射光谱,每个区域采集4处,以赤道1处、赤道4处平均和全部12处平均光谱作为各样本的光谱,建立PLS模型并对独立预测集样本进行预测。结果显示,12处平均光谱数据建模效果最好,其预测相关系数Rp=0.962,预测均方根误差RMSEP=0.432;赤道4处平均光谱建模效果次之,Rp=0.793,RMSEP为0.588;赤道1处光谱建模效果最差,Rp=0.687,RMSEP=0.599。再经过连续投影算法（SPA）筛选全谱变量,得到23个特征波...

Full description

Saved in:
Bibliographic Details
Published in广东农业科学 Vol. 41; no. 6; pp. 205 - 208
Main Author 李攀 代芬 欧少章 余国雄 陈珊 邓小玲 姚伟璇
Format Journal Article
LanguageChinese
Published 华南农业大学工程学院,广东广州,510642%华南农业大学工程学院,广东广州510642 2014
南方农业机械与装备关键技术教育部重点实验室,广东广州510642
国家柑橘产业技术体系机械研究室,广东广州510642
Subjects
光谱取样位置
可溶性固形物
漫反射
番石榴
近红外光谱
漫反射
番石榴
光谱取样位置
可溶性固形物
近红外光谱
Online AccessGet full text
ISSN1004-874X

Cover

More Information
Summary:基于漫反射法的番石榴可溶性固形物含量（SSC）无损检测中,确定番石榴最佳光谱取样的位置和数量,对提高检测精度具有重要意义。分别采集番石榴顶端、赤道和底部区域漫反射光谱,每个区域采集4处,以赤道1处、赤道4处平均和全部12处平均光谱作为各样本的光谱,建立PLS模型并对独立预测集样本进行预测。结果显示,12处平均光谱数据建模效果最好,其预测相关系数Rp=0.962,预测均方根误差RMSEP=0.432;赤道4处平均光谱建模效果次之,Rp=0.793,RMSEP为0.588;赤道1处光谱建模效果最差,Rp=0.687,RMSEP=0.599。再经过连续投影算法（SPA）筛选全谱变量,得到23个特征波长,此时PLS模型的Rp=0.902,RMSEP=0.438。试验结果表明,番石榴多处平均的漫反射光谱充分携带其内在品质信息,建模效果优于单处或单区域采样光谱。
Bibliography:44-1267/S
guava;soluble solid content;diffuse reflection;spectrum sampling position;near-infrared spectra
LI Pan, DAI Fen, OU Shao-zhang, YU Guo-xiong, CHEN Shan, DENG Xiao-ling, YAO Wei-xuan （1.College of Engineering, South China Agricultural University, Guangzhou 510642, China; 2.Key Lcdooratory of Key Technology for South Agricultural Machirtery and Equipment, Ministry of Education, Guangzhou 510642, China; 3.Division of Citrus Machinery, China Agriculture Research System, Guangzhou 510642, China）
It is important to select optimum spectra sampling position and quantity in noninvasive examination of guava based on near-infrared diffuse reflection. Firstly, 12 kinds of spectra were collected from top, equator and bottom, respectively, then models were built based on 1 spectrum in equator, average spectrum of 4 spectra in equator and average spectrum of 12 spectra all over the samples. The prediction results demonstrated that the model based on average spectrum of 12 spectra all over the samples was the best, wi
ISSN:1004-874X