基于可见近红外光谱分析技术的马铃薯品种鉴别方法

摘要

基于可见-近红外光谱分析技术，提出了一种快速鉴别马铃薯品种的方法。以三种不同品种共计352个样本的马铃薯作为主要研究对象，随机将其分为建模集(307个样本)和预测集(45个样本)。对其中的建模集样品进行可见-近红外光谱分析，将获取的光谱图像通过多元散射校正(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(PLS)对数据进行降维、压缩，使用主成分分析方法(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，为马铃薯品质检测与鉴别提供了新思路。%Potato (Solanum tuberosum L.),as one of the most important carbohydrate food crops in the China ranking the-fourth after rice,wheat and maize,plays a significant role in national economy.Since there are many varieties of potato,the quality such as physical sensory property and chemical components,differ drastically with the variety of potato.Different potato varieties are suitable for different utilization.Thus,the rapid and nondestructive identification of potato cultivars plays an impor-tant role in the better use of varieties.Near infrared (NIR)spectroscopy has raised a lot of interest in the classification and iden-tification of agricultural products because it is a rapid and non-invasive analytical technique.In this study,a rapid visible (VIS) and near infrared (NIR)spectroscopic system was explored as a tool to measure the diffuse spectroscopy of three different spe-cies of potatoes.352 potato samples (Sample A 142,Sample B 84,Sample C 126)from different sites in Heilongjiang province of China,obtained from peddlers market,were randomly divided into two sets at random:calibration set and prediction set,with 307 samples and 45 samples respectively for each set.The potatoes in the calibration set were tested with visible-near infrared spectroscopy method.The spectral data obtained from this test were analyzed with near infrared spectral technology,along with data processing algorithm,i.e.,Savitzky-Golay (S-G)smoothing and multiplicative scatter correction (MSC).The spectra data was firstly transformed by multiplicative scatter correction (MSC)to compensate for additive and/or multiplicative effects.In or-der to reduce the noise components from a raw spectroscopic data set,Savitzky-Golay smoothing and differentiation filter method were introduced.It was proved that,with the soothing segment size of 9,many high frequency noises components can be elimi-nated.Based on the following analysis with principal component analysis (PCA),partial least square (PLS)regression and back propagation artificial neural network (BP-ANN),a near infrared discrimination model was established.The results obtained from the partial least squares (PLS)analysis showed a positive cumulate reliability of more than 9 6% for the first four compo-nents.The clustering effect was also getting better.After that,twenty absorption peaks extracted from the first four principal components were applied as BP neural network inputswhile a three layers BP neural network [20(input)-12(implicit)-3 (out-put)]was constructed,upon which the recognition accuracy of potato varieties for those Prediction Set samples reaches 100%. As a result,the model established in this study can rapidly and accurately identify potato varieties without any destruction, which provides a new way for potato quality detection and variety identification.