【作者】 介邓飞；

【导师】 应义斌； 谢丽娟；

【作者基本信息】 浙江大学 ， 农业机械化工程， 2014， 博士

【摘要】 中国是世界水果生产和消费大国,西瓜的产量居世界之冠,但是出口量却非常少,分析其主要原因是由于中国水果商品化水平比较低,水果检测技术和检测水平比较落后。水果商品化水平对水果的内部品质要求较高,对西瓜生产过程中内部品质的检测是西瓜出口供应的一个必需环节。国内外对西瓜内部品质在线无损检测的研究报道还较少,现有的在线无损检测技术和检测方法还不是很成熟,只有国外少数国家拥有西瓜等大型厚皮类水果的无损检测生产设备,而国内水果无损检测装备制造企业设计生产的检测设备分选能力相对较弱,目前,国内还没有西瓜内部品质在线无损检测设备。因此,国内大型水果生产企业每年要花费大量的资金购买和维护国外的水果无损检测设备。本课题研究的最终目标是研制开发西瓜内部品质在线无损检测系统,以增强中国水果品质检测装备制造业的技术实力和技术水平,提高国内水果生产加工企业的效益和水果的品质质量。本研究选择厚皮类瓜果麒麟瓜为研究对象,对麒麟瓜的可溶性固形物(Soluble solids content, SSC)和总酸(Total acidity, TA)旨标进行了理化分析和测定,对麒麟瓜内部品质在线无损检测过程中的关键技术问题进行了研究,初步设计完成了麒麟瓜内部品质在线无损检测试验台。该检测系统是结合光学、机械学、电学、生物物料学和化学计量学于一体的光机电控在线无损检测系统,其主要涉及光谱分析、信息融合、自动控制等相关学科知识。本课题针对麒麟瓜内部品质在线无损检测系统的关键技术主要进行了如下几个方面的研究：(1)针对麒麟瓜体积大以及皮厚的特点,设计了麒麟瓜内部品质在线无损检测系统的总体结构,包括输送装置、光谱采集、控制单元以及光谱分析模型算法等。研制开发了基于近红外光谱(Near infrared spectroscopy, NIRS)技术麒麟瓜内部品质在线无损检测系统,初步解决了麒麟瓜体积大、皮厚难以检测内部品质的难点问题,以满足其内部品质快速在线无损检测要求；(2)该检测系统初步解决了麒麟瓜内部品质快速在线无损检测模型的关键问题。本研究主要对可溶性固形物和总酸两个指标进行了建模研究。并通过不同光谱预处理方法优化预测模型,应用遗传算法(Genetic algorithm, GA)、蒙特卡罗-无信息变量消除(Monte-Carlo uninformative variable elimination, MC-UVE)、蒙特卡罗-无信息变量消除结合遗传算法(MC-UVE-GA)等变量优选方法提取特征变量,剔除无关信息变量,减少了变量数目,初步建立了适用于该检测系统的偏最小二乘回归(Partial least squares regression, PLSR)预测模型；(3)研究了麒麟瓜瓜脐、瓜梗、赤道和接地端四种检测部位和0.1m/s、0.3m/s、0.5m/s和0.8m/s四种输送速度对检测精度的影响。采用美国海洋光学公司QE65000光谱仪,光谱采集采用漫透射方式,入射角度为120°(检测透镜和光源相对水果的角度),12盏150W的卤钨灯,光源功率可调。研究结果显示,预处理结合变量优选方法进行模型简化处理之后,在本研究采用的四种传输速度中,以0.8m/s输送速度在瓜脐处检测对可溶性固形物的结果较其余三种好,采用14个波长,它们分别为800.8、801.5、802.3、806.7、806.0、800.0、807.5、788.8、829.1、775.3、788.1、820.9、781.3和787.3nm,所建PLSR预测模型的校正集相关系数(Correlation coefficient of calibration, rcal)为0.847,校正均方根误差(Root mean square error of calibration, RMSEC)为0.550%,预测集相关系数(Correlation coefficient of prediction, rpre)为0.836,预测均方根误差(Root mean square error of prediction, RMSEP)为0.500%；以0.5m/s输送速度在赤道部位检测对总酸有较优的预测结果,采用8个波长,它们分别为792.5、757.3、791.8、793.3、774.6、758.1、775.3和756.6nm,所建PLSR预测模型的rcal为0.785,RMSEC为0.01076%,rpre为0.763,RMSEP为0.01106%；通过对四种检测条件下的模型研究证明,不同检测速度和检测部位都会对检测模型精度造成影响。同时,该检测系统检测精度还有待进一步提高,距离实际生产应用还有一定差距；(4)研究了麒麟瓜成熟度的定性判别分析：采用美国海洋光学公司QE65000微型光纤光谱仪采集的漫透射光谱建立定性判别模型；创新性地提出了光谱波峰比值法(Ratio of peak1to peak2, RPP),并与其他常见的化学计量学判别方法进行了比较,包括线性比较判别分析(Linear discriminant analysis, LDA)、簇类的独立软模式分类法(Soft independent modeling of class analogy, SIMCA)和最小二乘-支持向量机(Least squares support vector machines, LS-SVM)分类法。光谱波峰比值法利用720-740nm和803-805nm处光谱峰值进行麒麟瓜成熟度鉴别,成功地将麒麟瓜成熟度归纳为四类：不成熟,半成熟,成熟和过成熟。具体划分方法为,当RPP在0.4和0.69之间时为过成熟麒麟瓜；当RPP在0.7和0.96之间为成熟麒麟瓜；当样本的RPP位于0.97到1.23之间为半成熟麒麟瓜；而当样本的RPP低于1.24时被归为不成熟的一类。研究结果表明该方法的预测集准确率为82.1%；(5)设计并初步实现了麒麟瓜内部品质在线无损检测系统,该在线无损检测系统实验中运行的速度范围为0.1-0.8m/s,可以用于厚皮类瓜果麒麟瓜内部品质的在线无损检测。通过四种不同输送速度和四种检测部位条件下的试验研究证实,该系统初步具有在线无损检测麒麟瓜内部品质的功能,但是检测的精度不是很高,检测系统的稳定性和可靠性还需进一步提高。更多还原

【Abstract】 China has large yield and consumption of fruit, with the highest yield of watermelon in the world. However the export is very low due to the relatively low level of fruit’s commercialization. The detection technology of fruit in our country is relatively far behind from many other countries. For watermelon, it is necessary to grade according to the internal quality before exportation. At present, the reports about non-destructive on-line detection of watermelon at home and abroad are limited. The existing non-destructive on-line detective technology needs to be improved. There are only a few of foreign countries own the grading equipment for the large and thick-skinned fruit such as watermelon. It is still lack of comprehensive scientific research. Besides, the domestic manufacturers can’t meet the needs of fruit production companies with relatively week abilities in grading equipment production. Thus, the domestic fruit production companies spend a lot of money every year on buying foreign detection equipment. To date, there is no commercial watermelon internal quality detection equipment in China.The final goal of this research is to develop the on-line watermelon internal quality detection system. We hope to enhance the competition of the Chinese equipment manufacture industry for the fruit quality detection, and improve the benefit of domestic fruit production and processing enterprises.In this study, we chose the Qilin watermelon that is a variety with thick skin as the object. We determined the soluble solids content (SSC) and total acid (TA), which are indicators of maturity; the traditional chemical analysis methods were adopted. We discussed the key technical issues during the on-line detection of internal quality of watermelon. We designed and finished a watermelon internal quality on-line testing system. The detection system combined the optical, mechanical, electrical, biological and chemical materials metrology technology, it is a system integrated the light-controlled electromechanical operator. It mainly involves spectral analysis, information fusion, automatic control and more related discipline. This work focuses on the key technologies of on-line internal quality detection system of watermelon. The study contains following aspects:(1) According to the characteristics of watermelon, we designed the overall frame of on-line detection system for the internal quality of watermelon, including system integration, control unit and spectral model algorithms optimization. To meet the needs of fast, on-line nondestructive detection requirements; We designed the on-line detection system based on near-infrared spectroscopy (NIRS) technology, solved the problem that it is hard to determine the internal quality because of the big size and thick skin of watermelon.(2) It is important to create an accurate rapid predictive detection model for SSC and TA of watermelon. This is the key issue in achieving the non-destructive on-line detection for internal quality of watermelons. In this study, different pretreatment methods were used to optimize the predictive model. We also applied the genetic algorithm (GA), Monte Carlo-uninformation variable elimination (MC-UVE) and Monte Carlo-uninformation variable elimination and genetic algorithm (MC-UVE-GA) to extract feature wavelength. We reduced the variables and established partial least squares regression (PLSR) predictive models to adapt our system;(3) In order to study the detection portion and convey speed on detection accuracy, four different positions (stem, cylax, equator and grounded position) and four speed (0.1m/s,0.3m/s,0.5m/s and0.8m/s) were investigated. In this study, we used QE65000Ocean Optics spectrometer, diffuse transmission mode and the incident angle was set as120°(the angle between watermelon and light source), twelve150W halogen light source power are variable. The results showed that after the model optimization by spectral pretreatment and feature wavelength selection, among the four speeds applied in this study, the better predictive results of SSC obtained at0.8m/s and detected at stem position. Under this condition,14wavelengths were picked out; they were800.8,801.5,802.3,806.7,806.0,800.0,807.5,788.8,829.1,775.3,788.1,820.9,781.3and787.3nm, respectively. The correlation coefficient of calibration (rcal) was0.847, root mean square error of calibration (RMSEC) was0.550°Brix, correlation coefficient of prediction (rpre) was0.836, root mean square error of prediction (RMSEP)0.500°Brix. In TA predictive model, modeling with the spectra acquired at0.5m/s detection speed and at the equator position was the best,8wavelengths (792.5,757.3,791.8,793.3,774.6,758.1,775.3and756.6nm) were picked out, rcal was0.785, RMSEC to0.01076%, rpre was0.763, RMSEP was0.01106%. The results proved that different convey speeds and different positions would make an effect the detection accuracy. The detective precision still needs to be improvd to achieve the final purpose.(4) To carry on the qualitative watermelon maturity discriminant analysis:we used the Ocean Optics QE65000miniature fiber optic spectrometers to acquire transmittance spectra and establish qualitative discriminant model. We firstly proposed using the ratio of spectral peaki to peak2(RPP) to classify the maturity. We compared this effect of this method with other common chemometric discriminant method, including linear discriminant analysis (LDA), soft independent modeling of class analogy (SIMCA) and least squares support vector machines (LS-SVM). This method used the ratio of spectral peak at720-740nm and spectral peak at803-805nm to identify the watermelon maturity. We successfully classified them into four groups:immature, mid-mature, mature and over-mature. The specific division method is that when RPP is in0.4to0.69, the watermelon has over-mature; if the RPP is0.7to0.96, watermelon is mature; when the RPP is located between0.97and1.23, the watermelon is mid-ripen; when the RPP greater than1.24, the samples would be classified to immature. The classification accurate was82.1%by this method.(5) We designed and built an on-line detection system of watermelon, the speed range was0.1-0.8m/s. This system can be used for on-line detection of Qilin watermelon. We verified the function of this system through experiment, but the detection accuracy is not high enough. The stability and reliability of the system are still need to further improvement.更多还原