在树果实品质快速检测的方法研究

发布时间：2018-06-06 17:15

本文选题：近红外光谱 + 水果　；参考：《江苏科技大学》2017年硕士论文

【摘要】：目前,对在树和采后的核果类水果的大部分品质参数多采用传统方法进行检测,这些方法常常具有破坏性且时间依赖性高,而近红外光谱分析技术具有分析速度快、绿色、无损等优点,可以实现水果品质的快速无损评价。(1)实现了无损检测单颗葡萄中可溶性固形物(SSC)含量,获得个体和群体信息,以期指导田间管理、葡萄储存条件设置及满足消费者对葡萄不同的口味需求。采用手持式NIR光谱仪MicroNIR 1700在950-1650 nm波长范围采集葡萄的近红外光谱,应用偏最小二乘(PLS)回归建立葡萄SSC预测模型。为了减少冗余无信息变量,增加模型的预测精度和稳定性,采用无信息变量消除法(UVE)、随机蛙算法(RF)筛选出与葡萄SSC相关的重要波长变量。研究结果表明RF筛选建立的SSC预测模型优于全光谱PLS和UVE筛选建立的模型。RF-PLS模型的校正集、交叉验证及预测集RC,RCV和RP分别为0.9801、0.9661和0.9646,均方根误差RMSEC,RMSECV和RMSEP分别为0.6382°Brix、0.8299°Brix和0.8688°Brix。本研究表明,通过波长优选后的,手持式近红外光谱仪在预测单颗葡萄SSC的应用上完全可行,有较高的预测精度。(2)SSC、干物质含量(DMC)、总酚含量(TPC)以及总黄酮含量(TFC)是评价果用桑椹营养品质的重要指标。利用漫反射近红外光谱分析技术建立快速、实时无损地检测桑椹中营养成分的方法。首先用便携式近红外光谱仪MicroNIR1700采集桑椹的近红外光谱,光谱经预处理后,应用PLS建立桑椹SSC、DMC、TPC以及TFC的预测模型,并用UVE、自适应重加权采样(CARS)和RF三种方法筛选出最优波长变量,提高PLS模型精度。相较于UVE和RF法,CARS方法优选出更少的波长变量,所建PLS模型的预测效果更好。SCC、DMC、TPC以及TFC的CARS-PLS模型的校正相关系数RC分别为0.9807、0.9697、0.9491和0.9697,RMSEC分别为0.7035°Brix、0.6266%、0.3823 mg/g没食子酸和0.2666 mg/g芦丁,RP分别为0.9514、0.9430、0.8667和0.8771,RMSEP分别为1.2100°Brix、0.9178%、0.6352 mg/g没食子酸和0.6939mg/g芦丁。研究结果表明,手持式近红外光谱仪MicroNIR 1700结合化学计量学方法,能够用于现场对桑椹SCC、DMC、TPC以及TFC的快速无损检测。(3)实现了在两种相同型号仪器上苹果中SSC近红外模型的转移。本文使用K/S方法探讨了转移集样品数量对RMSEP的影响,确定样品数为18时转移效果最佳。采用直接校正法(DS)将从机光谱进行校正,结合CARS波长优选算法,筛选出25个波长点建立PLS分析模型。最终,苹果的SSC近红外模型成功的从主机转移到从机上,Rp为0.8603,RMSEP为0.7914°Brix。DS算法结合CARS波长优选方法可以有效的应用到模型转移的研究中。
[Abstract]:At present, traditional methods are used to detect most of the quality parameters of drupe fruits in trees and postharvest fruits. These methods are often destructive and time-dependent, while the near infrared spectroscopy (NIR) analysis technology is fast and green. In order to guide field management, it can realize the fast nondestructive evaluation of fruit quality and the content of soluble solids (SSCS) in a single grape, so as to obtain individual and group information. Grape storage conditions and to meet the needs of consumers for different tastes of grapes. The near infrared spectra of grape were collected by a hand-held NIR spectrometer MicroNIR 1700 in the wavelength range of 950-1650 nm. The prediction model of grape SSC was established by partial least squares regression. In order to reduce redundant uninformed variables and increase the prediction accuracy and stability of the model, an information free variable elimination method and a random frog algorithm were used to screen the important wavelength variables associated with grape SSC. The results show that the SSC prediction model established by RF screening is superior to the calibration set of full-spectrum PLS and UVE model. The cross validation and prediction set RCU RCV and RP are 0.9801g 0.9661 and 0.9646, respectively. The RMSEC-RMSECV and RMSEP are 0.6382 掳Brix 0.8299 掳Brix and 0.8688 掳Brix, respectively. The results show that the application of hand-held near infrared spectrometer in predicting single grape SSC is feasible after wavelength optimization. The high prediction accuracy of SSCS, the content of dry matter (DMC), the content of total phenol (TPC) and the content of total flavonoids (TFCC) are important indexes for evaluating the nutritive quality of fruit mulberry. A rapid, real-time and nondestructive method for the determination of nutrients in mulberry was established by means of diffuse reflectance near infrared spectroscopy. At first, the near infrared spectra of mulberry were collected by portable near infrared spectrometer MicroNIR1700. After the spectrum was pretreated, the prediction models of mulberry SSCN DMCT TPC and TFC were established by PLS, and the optimal wavelength variables were screened out by three methods: UVE, adaptive re-weighted sampling (CARSs) and RF. Improve the precision of PLS model. Compared with the UVE and RF car methods, fewer wavelength variables are selected. The correlation coefficient RC of the model was 0.9807 / 0.9697 / 0.9491 and 0.9697, respectively. The RMSECs of 0.3823 mg/g Gallic acid and 0.2666 mg/g rutin RMSEP were 0.7035 掳Brix 0.9697 mg/g Gallic acid and 0.2666 mg/g rutin 0.8667 and 0.8771g RMSEP were 1.2100 掳Brix 0.917852 mg/g Gallic acid and 0.6352 0.6939mg/g rutin, respectively. The results show that the hand-held near infrared spectrometer MicroNIR 1700 combined with chemometrics can be used for the fast nondestructive testing of mulberry fruit SCC DMCN TPC and TFC in the field. The transfer of the near infrared model of apple on the same type of instrument is realized. In this paper, the effect of sample number on RMSEP is discussed by using the K / S method, and it is determined that the best transfer effect is at 18:00. The slave spectrum is corrected by direct correction method (DSS), and 25 wavelength points are selected to establish PLS analysis model combined with car wavelength selection algorithm. Finally, Apple's SSC near-infrared model is successfully transferred from host to machine with a RP of 0.8603 and RMSEP of 0.7914 掳Brix.DS algorithm combined with car wavelength optimization can be effectively applied to the research of model transfer.
【学位授予单位】：江苏科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S66

【参考文献】