摘要： 小麦是制作馒头的主要原料之一，小麦中水、蛋白质、淀粉会因产地以及烘干程度的差异而不同，进而影响到加工成馒头的品质。所以实现对小麦产地和烘干程度的快速鉴别就显得尤为重要。感官评定是鉴别小麦产地和烘干程度常用的方法，对比感官评定，光谱分析可以识别样品中的分子结构等信息。基于此，尝试利用近红外和中红外光谱融合技术实现对不同产地和不同烘干程度的小麦同时鉴别。首先选取了两个不同产地的小麦，再利用微波干燥法对两个不同产地的小麦做烘干预处理，使烘干的小麦水含量为12%±0.5%，原麦水含量为18%±0.5%。分别标记为原麦A，烘干A，原麦B，烘干B，再将小麦研磨成粉末，过100目筛网筛选后，置于自封袋中备用。随后分别采集四种小麦样品的近红外和中红外光谱信息，在Matlab 7.10的环境下使用标准正态变量变换(standard normal variable transformation, SNVT)对采集到的原始光谱数据进行预处理，利用主成分分析对预处理后的数据进行降维处理，再结合线性判别分析（linear discriminant analysis，LDA）和支持向量机（support vector machine, SVM）分别建立小麦近红外、中红外光谱数据识别模型。另外利用联合区间偏最小二乘法（synergy interval partial least square, SiPLS）筛选出利用标准正态变量变换(SNVT)预处理后的小麦近红外和中红外光谱数据特征光谱区间，将筛选出的近红外和中红外光谱数据特征光谱区间融合后再结合线性判别分析（LDA）和支持向量机（SVM）建立小麦融合光谱信息的识别模型。然后比较同种光谱数据下利用线性判别分析（LDA）和支持向量机（SVM）建立的小麦识别模型识别率、比较同种建模方法下近红外和中红外光谱数据建立小麦识别模型识别率、比较同种建模方法下光谱数据融合和单一光谱数据建立小麦识别模型识别率。结果表明，同种光谱分析方法，利用SVM建立的四种小麦识别模型识别率高于利用LDA建立的小麦识别模型识别率。同种建模方法，近红外光谱数据建立的小麦识别模型识别率优于中红外光谱数据建立的小麦识别模型识别率。而在同种建模方法下，利用SiPLS筛选出近红外和中红外光谱数据的特征光谱区间数据融合后建立小麦识别模型识别率最高，光谱数据融合后结合LDA建立的小麦识别模型校正集识别率为98.75%，预测集识别率为97.50%；而将此选择的变量结合SVM建立的小麦识别模型的校正集和预测集识别率都达到100.0%。对比利用单一光谱数据建立的小麦识别模型识别率，光谱数据融合之后建立的小麦识别模型识别率得到显著提高，该研究从纵向和横向上全面地比较了光谱数据建立的小麦模型识别率，结果可为更准确地运用光谱融合技术建立小麦产地以及烘干程度识别模型提供参考。
关键词：小麦；光谱分析技术；联合区间偏最小二乘法；线性判别分析；支持向量机
Abstract：Wheat is one of the main raw materials for making steamed bread, and the water, protein and starch in wheat vary depending on the place of production and the degree of drying, which in turn affects the quality of processed steamed bread. Therefore, it is particularly important to quickly identify the place of origin and degree of drying of wheat. Sensory evaluation is a common method used to identify the origin and degree of drying in wheat, in contrast to sensory evaluation, spectral analysis techniques can identify information such as molecular structure in a sample. Based on this, this paper attempts to use the near-infrared and mid-infrared spectral fusion technology to achieve the simultaneous identification of wheat from different producing areas and different degrees of drying. In this study, wheat from two different origins were selected and microwave drying was used to pretreat the wheat from the two different origin so that the moisture content of the dried wheat decreased to 12%±0.5%, while the moisture content of the undried wheat was 18%±0.5%. They were marked as undried wheat A, dried A, undried wheat B, dried B and then ground into powder, sieved with a 100 mesh screen and placed in a sealed bag for use. Subsequently, the near infrared and mid-infrared spectral information of four wheat samples were collected, and then the raw spectral data collected were pre-processed using the standard normal variate transformation (SNVT) using Matlab 7.10 version. The principal component analysis was used to reduce the dimension of the preprocessed data, and then the linear and short-infrared (NIR) and mid-infrared (MIR) spectral data were identified using linear discriminant analysis (LDA) and support vector machine (SVM), respectively to create a recognition model. In addition, using the synergy interval partial least square (SiPLS) method, the characteristic spectral ranges of the near-infrared and mid-infrared spectral data of the wheat pretreated with the standard normal variable transformation (SNVT) were screened out. After the fusion of the characteristic spectral ranges of the near-infrared and mid-infrared spectral data, a linear discriminant analysis (LDA) and support vector machine (SVM) were used to establish the identification model of the fusion spectral information of wheat. The recognition rate of wheat identification model established by linear discriminant analysis (LDA) and support vector machine (SVM) under the same spectral data were compared and the near-infrared and mid-infrared spectral data of the same modeling method established the wheat identification model. Recognition rate, comparison of spectral data fusion under the same modeling method and single spectral data were used to establish the recognition rate of the wheat identification model. The results showed that using the same kind of spectral analysis method, the recognition rates of the four wheat identification models established using SVM were higher than those of wheat identification models established using LDA. The recognition rate of wheat identification model established by near-infrared spectral data using the same modeling method was better than that of wheat identification model established by mid-infrared spectral data. Under the same modeling method, the identification rate of the wheat identification model established by the fusion of the characteristic spectral interval data of the near-infrared and mid-infrared spectral data filtered by SiPLS was the highest. After the fusion of spectral data, the wheat identification model established with LDA was integrated. The recognition rate of the correction set was 98.75%, and the recognition rate of the prediction set was 97.50%. The recognition rate of the correction set and the prediction set of the wheat identification model established by combining this selected variable with the SVM reached 100.0%. Comparing the recognition rate of wheat identification model established by using single spectral data, the recognition rate of wheat identification model established after fusion of spectral data was significantly improved. This study compared the wheat model’s recognition rate established by spectral data from both vertical and horizontal directions. The results can provide a reference for the more accurate use of spectral fusion technology to establish wheat production areas and drying degree identification model.
Key words：Wheat; Spectral analysis technology; Synergy interval partial least square (SiPLS); Linear discriminant analysis (LDA); Support vector machine (SVM)
[1] ZHANG Chun-qing, LI Qing-qi(张春庆, 李晴祺). Scientia Agricultura Sinica(中国农业科学), 1993, 26(2): 39.
[2] FAN Yu-ding, LI Si-shen, SUN Hai-yan, et al(范玉顶, 李斯深, 孙海艳, 等). Journal of the Chinese Cereals and Oils Association(中国粮油学报), 2005, 20(2): 16.
[3] Machadojr J C, Faria M A, Ferreiraimpl V O, et al. Talanta, 2018, 180: 69.
[4] DUAN Guo-hui, ZHANG Yuan, GAO Hai-tao, et al(段国辉, 张 园, 高海涛, 等). Jiangsu Agricultural Sciences(江苏农业科学), 2011, 39(5): 422.
[5] ZHAO Hai-yan, GUO Bo-li, WEI Yi-min, et al(赵海燕, 郭波莉, 魏益民, 等). Scientia Agricultura Sinica(中国农业科学), 2011, 44(7): 1451.
[6] DENG Hai-yan, SONG Xiang-zhong, XIONG Yan-mei(邓海雁, 宋相中, 熊艳梅). Spectroscopy and Spectral Analysis(光谱学与光谱分析), 2016, 36(10): 183.
[7] SHI Ji-yong, HU Xue-tao, ZHU Yao-di, et al(石吉勇, 胡雪桃, 朱瑶迪, 等). Transactions of the Chinese Society for Agricultural Machinery(农业机械学报), 2016, 47(2): 245.
[8] ZHANG Ju-hua, ZHU Xiang-rong, SHANG Xue-bo, et al(张菊华, 朱向荣, 尚雪波, 等). Food Industry Technology(食品工业科技), 2010, 31(10): 421.
[9] ZOU Xiao-bo, CUI Xue-ping, SHI Ji-yong, et al(邹小波, 崔雪平, 石吉勇, 等). Transactions of the Chinese Society for Agricultural Machinery(农业机械学报), 2018, 49(3): 341.
[10] WU Shuang, WANG Jie, YU Ya-ru, et al(吴 双, 王 杰, 俞雅茹, 等). Journal of the Chinese Cereals and Oils Association(中国粮油学报), 2017, 32(11): 158.
[11] WANG Sheng-peng, ZHENG Peng-cheng, GONG Zi-ming, et al(王胜鹏, 郑鹏程, 龚自明, 等). Journal of Huazhong Agricultural University(华中农业大学学报), 2018, 37(1): 89.
[12] Hell J, Prckle R M, Danne R L, et al. Food Control, 2016, 60(5): 365.
