固态酿造中水分含量是检测大曲不同发酵阶段的关键因素,为快速定量检测发酵过程中水分含量及分布情况,通过高光谱成像技术建立BP神经网络(back propagation neural network, BPNN)数据模型,实现曲块发酵水分含量快速检测。根据所采集大曲发酵高光谱数据,应用主成分分析和实验分析对比提取大曲发酵过程中水分特征光谱所对应的PC1、PC2、PC3、1 200、1 450 nm波段图像;通过图像灰度共生矩阵算法提取水分纹理特征;根据所提取图像纹理特征对曲块发酵水分含量运用偏最小二乘回归、BPNN、支持向量机回归等进行建模预测对比,选择最佳预测模型。结果表明,利用BPNN与1 450 nm特征波段光谱图像纹理特征对水含量建模预测的效果最佳,其训练集决定系数(R2)和均方根误差(root mean square error, RMSE)分别为0.826 9和0.033 5,预测集R2和RMSE分别为0.848 4和0.028 7。研究表明,利用高光谱特征光谱所对应图像纹理信息能够对水分含量进行关联预测,为实现对大曲发酵过程水分含量检测提供理论依据。
The moisture content in solid fermentation is a key factor for different fermentation stages of Daqu. In order to detect the moisture content and distribution in the fermentation process rapidly and quantitatively, the hyperspectral imaging technology was applied and the back propagation neural network(BPNN) data model was established. According to the hyperspectral data of Daqu fermentation, principle component analysis (PCA) and experimental methods were firstly used to extract the PC1, PC2, PC3, 1 200 and 1 450nm band images corresponding to the water characteristic spectrum during Daqu fermentation, the GLCM algorithm was then used to extract the water texture features, PLSR, BPNN and SVR were finally used to model and predict the water content of Daqu fermentation according to the extracted image texture features, and then select the best prediction model. The results showed that the effect of BPNN and 1 450 nm spectral image texture features on water content modeling and prediction was the best. The training set determination coefficient (R2) and root mean square error (RMSE) were 0.826 9 and 0.033 5, respectively. The prediction set R2 and RMSE were 0.848 4 and 0.028 7, respectively. The results show that the texture information of hyperspectral characteristic spectrum can be used to predict the water content, which provides a theoretical basis for the detection of water content in Daqu fermentation process.
[1] 程劲松, 李春扬. 白酒质量控制技术的研究进展[J]. 食品安全质量检测学报, 2014, 5(7): 2 248-2 262.
[2] 蒋玉石, 骆婕茹, 赵丽娟. 新常态下的中国白酒行业发展趋势及应对策略研究[J]. 四川理工学院学报(社会科学版), 2015, 30(6): 46-55.
[3] 敖宗华, 陕小虎, 沈才洪, 等. 国内主要大曲相关标准及研究进展[J]. 酿酒科技, 2010(2): 104-108.
[4] 吴翠芳, 刘国英, 何宏魁. 探索新时代白酒可持续发展的途径[J]. 酿酒, 2018, 45(3): 10-12.
[5] 沈怡方. 白酒生产技术全书[M].北京: 中国轻工业出版社, 1998:45-79.
[6] 张芯豪, 黄丹平, 田建平, 等. 基于机器视觉的大曲质量检测系统研究[J]. 食品与机械, 2018, 34(4): 80-84.
[7] 童庆禧, 张兵, 张立福. 中国高光谱遥感的前沿进展[J]. 遥感学报, 2016, 20(5): 689-707.
[8] 刘燕德, 张光伟. 高光谱成像技术在农产品检测中的应用[J]. 食品与机械, 2012, 28(5): 223-226;242.
[9] 王顺, 黄星奕, 吕日琴, 等. 水果品质无损检测方法研究进展[J]. 食品与发酵工业, 2018, 44(11): 319-324.
[10] 王芹志, 强锋, 何建国, 等. 基于可见-近红外光谱预测灵武长枣脆度及模型优化[J]. 食品与发酵工业, 2017, 43(3): 205-211.
[11] 朱荣光, 段宏伟, 王龙, 等. 不同分集方法对牛肉嫩度高光谱检测模型的比较[J]. 食品与发酵工业, 2016, 42(4): 189-192.
[12] 石吉勇, 吴胜斌, 邹小波, 等. 基于高光谱技术的酸奶中常见致病菌的快速鉴别及计数[J]. 光谱学与光谱分析, 2019, 39(4): 1 186-1 191.
[13] 孙红, 刘宁, 吴莉, 等. 高光谱成像的马铃薯叶片含水率分布可视化[J]. 光谱学与光谱分析, 2019, 39(3): 910-916.
[14] 邹小波, 徐艺伟, 陈武, 等. 高光谱图像技术检测枇杷叶片三萜酸含量及叶面分布[J]. 现代食品科技, 2015, 31(9): 263-267;288.
[15] 谢安国, 康怀彬, 王飞翔, 等. 高光谱成像检测煎制中调理牛肉品质的变化[J]. 食品与机械, 2018, 34(11): 20-23;54.
[16] 张达, 郑玉权. 高光谱遥感的发展与应用[J]. 光学与光电技术, 2013, 11(3): 67-73.
[17] BAIANO A, TERRACONE C, PERI G, et al. Application of hyperspectral imaging for prediction of physico-chemical and sensory characteristics of table grapes[J]. Computers and Electronics in Agriculture, 2012, 87:142-151.
[18] 刘建平, 赵英时, 孙淑玲. 高光谱遥感数据最佳波段选择方法试验研究[J]. 遥感技术与应用, 2001(1): 7-13.
[19] 魏峰. 高光谱遥感数据特征提取与特征选择方法研究[D]. 西安:西北工业大学, 2015.
[20] YAMATERA H, FITZPATRICK B, GORDON G. Near infrared spectra of water and aqueous solutions[J]. Journal of Molecular Spectroscopy, 1964, 14(1-4): 268-278.
[21] INOUE A, KOJIMA K, TANIGUCHI Y, et al. Near-infrared spectra of water and aqueous electrolyte solutions at high pressures[J]. Journal of Solution Chemistry, 1984, 13(11): 811-823.
[22] LIU D, PU H, SUN D-W, et al. Combination of spectra and texture data of hyperspectral imaging for prediction of pH in salted meat[J]. Food Chemistry, 2014, 160:330-337.
