基于近红外光谱的马铃薯品种鉴别及干物质含量检测方法研究

发布时间：2018-06-16 13:01

本文选题：近红外光谱 + 马铃薯　；参考：《黑龙江八一农垦大学》2016年硕士论文

【摘要】：立足粮食的供求形势,继稻米、小麦、玉米之后,马铃薯成为我国新兴的第四大主粮。近几年,我国深入研讨了马铃薯主粮化的战略意义,积极推进马铃薯产业化发展,因此,需要更科学、更高效的方法来严控生产中的每一环节。就马铃薯品种鉴别和品质检测而言,多数实验室依旧采用传统方法进行,这些方法不适合生产过程中大批量样品的实时分析,并对样品存在破坏性。因此,研究新方法来提高马铃薯各指标检测效率,意义十分重大。近红外光谱分析技术是我国近二十年逐步发展起来的一项快速分析技术,它高效、无损的分析特点被很多领域广泛应用,结合马铃薯当前的产业需求,本文开展了基于近红外光谱分析的马铃薯品种鉴别及干物质含量检测方法研究,主要内容如下:1、基于近红外光谱的马铃薯品种快速鉴别方法的研究。试验以3种不同品种共计352个样本的马铃薯作为主要研究对象,并随机将其分为建模集(307个样本)和预测集(45个样本)。使用可见-近红外光谱仪获取建模集和预测集样品的光谱图,将获取的光谱图通过多元散射校正(multiplicative scatter correction,MSC)和窗口大小为9的Savitzky-Golay(S-G)一阶卷积求导方法预处理,消除颗粒大小、表面散射及光程变化对漫反射光谱的影响,降低原始光谱曲线的随机噪声影响。然后用偏最小二乘法(partial least square,PLS)对数据进行降维、压缩,使用主成分分析方法(principal component analysis,PCA)获得的前4个主成分累计贡献率达到96%以上,并从前4个主成分图谱中提取20个吸收峰作为输入变量,经过试验,得到一个20(输入)-12(隐含)-3(输出)结构的3层BP神经网络。最后利用该模型对预测集样本进行品种鉴别,识别正确率达到100%。此方法能较为快速、准确地鉴别马铃薯的品种,为马铃薯品种的快速鉴别提供了新思路。2、基于近红外光谱技术的马铃薯干物质含量检测研究。以207个具有代表性的马铃薯样本作为研究对象,其中115个用于马铃薯切片样本的研究,92个用于完整马铃薯的研究,通过对比两种样本的模型预测效果,探讨采用可见-短波近红外光谱进行马铃薯干物质含量的完全无损检测。切片样本光谱数据用Savitzky-Golay(S-G)一阶卷积求导方法预处理,根据局部最大值最小值原则和含氢基团(C-H、O-H)伸缩振动的敏感波段选定了5段特征波长参与建模,模型外部检验决定系数R2=0.9416,标准误差RMSE=3.91。完整马铃薯样本光谱数据经过多元散射校正处理的基础上使用S-G一阶卷积求导方法预处理,选取了线性关系较好的5段波长参与建模。模型外部检验决定系数R2=0.8475,标准误差RMSE=4.07。结果表明,完整马铃薯样本模型的检测效果虽然没有切片样本效果理想,但仍可以作为实际生产中进行马铃薯干物质含量检测的有效手段。
[Abstract]:Based on the supply and demand situation of grain, potato has become the fourth main grain after rice, wheat and corn. In recent years, China has deeply discussed the strategic significance of potato grain production and actively promoted the development of potato industrialization. Therefore, more scientific and efficient methods are needed to strictly control every link in production. As far as potato variety identification and quality detection are concerned, traditional methods are still used in most laboratories. These methods are not suitable for real-time analysis of large quantities of samples in the production process and are destructive to the samples. Therefore, it is of great significance to study new methods to improve the detection efficiency of potato indexes. Near-infrared spectroscopy (NIR) analysis technology is a rapid analysis technology developed gradually in China in the past two decades. It has been widely used in many fields because of its high efficiency and nondestructive characteristics, combining with the current demand of potato industry. In this paper, the identification of potato varieties based on near infrared spectroscopy and the determination of dry matter content were studied. The main contents are as follows: 1. The rapid identification method of potato varieties based on near infrared spectroscopy. In the experiment, 352 samples of 3 different varieties of potato were selected as the main research objects, which were randomly divided into two groups: the modeling set (307 samples) and the prediction set (45 samples). The spectral images of the modeling and prediction samples were obtained by using the visible-near infrared spectrometer. The obtained spectra were preprocessed by multiple scattering correction multiple scatter correction MSCs and Savitzky-Golayay S-GG convolution method with window size 9 to eliminate the particle size. The effect of surface scattering and optical path change on diffuse reflectance spectrum is reduced, and the random noise effect of the original spectral curve is reduced. Then the data are reduced and compressed by partial least square (PLS). The cumulative contribution rate of the first four principal components obtained by principal component analysis (PCA) is over 96%. Twenty absorption peaks were extracted from the first four principal component maps as input variables, and a three-layer BP neural network with 20 (input) -12 (implicit) (output) structure was obtained. Finally, the model is used to identify the samples of the prediction set, and the recognition accuracy is 100%. This method can identify potato varieties quickly and accurately, and provide a new way of thinking for rapid identification of potato varieties. 2. The determination of dry matter content in potato based on near infrared spectroscopy is studied. Among 207 representative potato samples, 115 were used for potato slicing and 92 for intact potato. A complete nondestructive test of dry matter content in potato by visible-short-wave near-infrared spectroscopy was studied. The spectral data of slice samples were preprocessed by Savitzky-Golay-S-G) first order convolution method. According to the principle of local maximum and minimum and the sensitive band of H-containing group C-HO-H) stretching vibration, five sections of characteristic wavelengths were selected to participate in modeling. The determination coefficient of external test is 0.9416, and the standard error is RMSE 3.91. On the basis of multivariate scattering correction of the complete potato sample spectral data, S-G first order convolution method was used to preprocess, and 5 wavelengths with good linear relationship were selected to model the model. The determination coefficient of external test of the model is 0.8475, and the standard error RMSE is 4.07. The results showed that the detection effect of intact potato sample model was not satisfactory, but it could be used as an effective method to detect the dry matter content of potato in practical production.
【学位授予单位】：黑龙江八一农垦大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：S532;O657.33

【相似文献】