基于近红外光谱技术的红枣品质分析方法研究

CORC > 沈阳自动化研究所 > 中国科学院沈阳自动化研究所 > 智能检测与装备研究室

题名	基于近红外光谱技术的红枣品质分析方法研究
作者	张翠侠
学位类别	硕士
答辩日期	2016-05-25
授予单位	中国科学院沈阳自动化研究所
导师	马钺
关键词	近红外光谱技术红枣光谱预处理定性分析定量分析
其他题名	Research on the Analysis Methods of Jujube Quality by Near Infrared Spectroscopy
学位专业	检测技术与自动化装置
中文摘要	本文结合红枣的光学特性设计红枣动态检测实验装置，构建了适合红枣内部品质无损检测的光谱采集系统，开展了一系列理化性质采集实验。使用近红外光谱仪采集红枣光谱数据，以采集到的光谱作为研究对象，对光谱信息进行提取，并针对光谱数据做了以下方法的研究。研究预处理方法对提高校正模型性能的作用。分别采用光谱预处理方法：一阶微分、二阶微分、多元散射校正（MSC）以及小波变换对红枣的原始光谱进行预处理。由于动态光谱采集时干扰因素较多，提出了采用两种预处理方法相结合的方法，进一步减少了噪声干扰。研究利用马氏距离判别分析法来建立红枣病虫害的定性分析模型，并提出将簇类独立软模式（SIMCA）引入定性分析之中，SIMCA法在进行判别前加入了两次主成分分解，提取主要影响因素，以光谱残差的数量级作为判别依据，提供了一种新的建模思路。在短波段、长波段以及全波段三个区域内，并结不同合预处理方法，结果显示：经过各种预处理后的校正集和预测集均有提高，范围2.00-12.50%内。二阶微分法结合小波变换能够有效地提高判别法建模的分析精度，判别法利用原始光谱进行分类建模的正确性是85.00%，小波结合微分建模结果最优准确率为97.50%。MSC结合一阶微分能有效提高基于SIMCA的建模分析精度，准确率由原来的84.38%提高96.25%。判别分析的准确率更高，且建模波段为长波区（910~2100nm）数据量小，模型分析效率高。利用近红外技术进行红枣分类建模是可行的且具有较高的识别准确度。进行定量分析时，建模波段为785.35~993.61nm，研究利用主成分回归分析法（PCR）和偏最小二乘回归法（PLS）建立红枣糖分即可溶性固态物质（SSC）的校正分析模型，其中PLS在建模是在PCR基础上做了改进，加入对物质含量的分析。对校正模型采用留一交互验证法，依据交互验证均方根误差值（RMSECV）来判断模型预测性能，最小的RMSECV对应建模最佳主成分因子数。在分析的红枣可溶性固态物质（SSC）作为品质指标时试验结果显示：利用主成分回归分析法（PCR）建模的最佳预处理方法为多元校正散射（MSC）结合小波变换（WT），最佳主成分数为6；利用偏最小二乘回归法（PLS）最佳预处理方法为一阶微分结合多元散射校正（MSC），最佳主成分数为5；相较于PCR模型，PLS的预测相关系数从0.7460提高到0.9716，预测均方误差从1.62降低到1.16，PLS建模效果优于PCR。总体的预测性能有所提高，且所用因子数更小有助于提高分析效率。分析结果表明：通过合适的光谱预处理可以提高模型的分析精度；最优的定性分析模型准确率达到96%以上；定量分析最优模型预测相关系数为0.9716，预测均方根误差为1.16；两种分析类型的预测能力都较好，说明利用近红外漫反射光谱技术对鲜枣进行无损检测具有可行性。本实验开发了近红外红枣内部品质动态检测实验装置，根据近红外光谱检测原理特性以及工作流程，研究了光谱预处理方法、建模波段、建模方法，开展对红枣好坏以及可溶性固形物含量的无损检测试验研究。试验结果表明，该试验平台已经基本满足了检测的要求，但仍有部分功能尚不完整不确定，需要在以后的研究中进行不断完善和改进。
英文摘要	Near infrared spectroscopy become more and more popular with its detection speed, high analysis accuracy, low cost and nondestructive detection. It also has a good performance in online analysis and remote control, so it is widely used in non-destructive detection and agricultural products. According to the laboratory conditions and the optical properties of jujube, we built internal quality spectrum acquisition system for non-destructive detection of jujube. As the same time, we also built the physical and chemical properties acquisition experiments. This work uses diffuse reflectance spectrum of NIR as sample to study the qualitative analysis of jujube, and classify infested and intact jujubes. The effects of different spectral pretreatment methods on the results of spectral analysis are studied. The results show that derivative can correct the baseline effects and increase the analysis accuracy. Noise from the derivative spectra has great detriment to the analysis of jujube. The wavelet transform has a good de-noising function. De-noising of wavelet transform can improve the analysis accuracy. The results show that the combination of wavelet transform and derivative can effectively improve the accuracy of the near-infrared spectrum. The correctness of classification of original spectrum is 87.08%, and it is improved by 2-8% after date pretreatment in both calibration and prediction set. The best result is 96.67%. MSC combined with first-order differential can improve the analysis accuracy of SIMCA model, the accurate rate of classification increased from 84.38% to 96.25%. Discriminant analysis has a higher accuracy, and the model only used the date of 910 ~ 2100nm region. As the decrease of date to operated, discriminant analysis is more efficient than SIMCA. NIR is illustrated could be applied to classify jujubes. This work is aimed to study feasible application for evaluating sugars content in jujube. The date in 785.35~993.61nm region of NIR diffuse reflectance was used as samples and we try to extract the primary wavelengths related to the prediction of soluble solids content (SSC) for jujube. Derivative and smooth was used as pretreatment methods to improve the accuracy of the model. Principal component regression (PCR) and partial least squares (PLS) were used to develop the models. The influence of the principal component count to the PLS model was also studied. Compared to PCR, the model performance of PLS improved. Rp increases from 0.7460 to 0.9716，and RMSEP decreases from 1.62 to 1.16. When the principal component count is 5, model predictive power achieved the best predict results: Rc is 0.9859、 RMSEC is 0.696、 Rp is 0.9716、 RMSEP is 1.16. The study shows that the near infrared diffuse reflectance technique was feasible to predict SSC content in jujube on-line. The results showed that, the choice of appropriate spectral pretreatment methods and optimal number of principal components can help to improve the prediction accuracy of jujube’s internal quality, and provide a new effective method for the online quality detection of jujube. The research uses near infrared spectroscopy fruit to develop the online non-destructive detection of jujube internal quality. We have studied the qualitative analysis of jujube to infested and intact jujubes and evaluating sugars content in jujube. The results show that the test platform has basically met the requirements of online detection. But there are still some existing problems of the system, and it needs further studying and improving.
语种	中文
产权排序	1
页码	59页
内容类型	学位论文
源URL	[http://ir.sia.cn/handle/173321/19662]
专题	沈阳自动化研究所_智能检测与装备研究室
推荐引用方式 GB/T 7714	张翠侠. 基于近红外光谱技术的红枣品质分析方法研究[D]. 中国科学院沈阳自动化研究所. 2016.