Efficient utilization of organic fertilizer through advanced technology and development of regional agriculture according to local conditions are a crucial method for the development of organic dry farming. Combining sheep manure with millet cultivation can significantly enhance millet quality. Unlike traditional detection methods, hyperspectral imaging offers a rapid, non-destructive means of analyzing samples, therefore, it has great potential in achieving rapid detection of millet crude fat and protein. In order to provide reference and theoretical support for improvement of millet cultivation practices and rapid assessment of nutrient content in millet through hyperspectral imaging, in this study, hyperspectral image data of 358 millet samples were collected, traditional detection methods were used to determine the content of crude fat and crude protein. Simultaneously, chemometrics methods were utilized to predict the content of the two components. The results indicated that the partial least squares regression(PLSR) model established by employing the successive projections algorithm(SPA) and iteratively retaining informative variables(IRIV) had the highest accuracy. The correlation coefficient(R2) predicted by the model for crude protein content was 0.88, with a root mean squared error(RMSE) of 0.59 and a relative percent deviation(RPD) of 1.99. The visualization representations by the regression model vividly illustrated the accumulation patterns of crude fat and protein at various application rates of sheep manure, and it was determined that applying 90 m3 of fertilizer per hectare was the optimal application amount.
高光谱检测具有无损、快速、无接触及化学试剂零使用等技术特点,目前,运用光谱技术针对农作物营养成分检测的研究也渐渐被农业科研人员所关注[3-5]。田翔等[6]对小米淀粉、直链淀粉、蛋白质和脂肪应用近红外光谱法进行快速检测,结果表明,采用化学法和近红外法检测结果无显著差异;王浩等[7]运用近红外结合PLSR模型实现小米硒含量的快速预测。此外,高光谱数据具有变量维度高的特点,为了降低变量维度、提升预测模型预测精度,一般采用特征波段提取算法对目标变量数据组进行特征波段选择。其中,变量组合集群分析(Variable Combination Population Analysis,VCPA)算法能够将所有存在相互影响的原始变量自由组合,并通过预设终端输出变量波段数阈值,选取极具代表性的变量波段,赵环等[8]运用该算法预测小麦蛋白质含量并取得较理想预测效果;SPA算法通过选择初始向量,计算在未选变量上的投影,选出最佳波段组合,李颖[9]运用SPA算法对木材可见-近红外光谱数据进行特征波段提取并建立PLS模型,在木材分类过程中取得较为理想效果;而IRIV算法可以在剔除冗余波段的同时将变量强、弱信息波段尽可能多地保留,于雷等[10]运用IRIV算法优选了大豆叶片高光谱波长并对叶绿素含量进行估测,为高光谱技术估测叶片生理参数提供新思路。目前,特征波段提取算法联用在特征波段选择上应用渐广,余海东[11]提出近红外光谱波长选择采用3步联用策略,提高模型预测精度同时,既突出各算法优势,也克服了算法局限性。因此,通过运用高光谱成像结合化学计量学相关知识实现小米粗脂肪、粗蛋白质快速检测具有重要的现实意义。
特征波段提取是从高光谱高维数据中提取、挖掘隐匿信息,消除无关信息干扰并提高回归模型预测能力,提升运算速度的方法。本研究中,提出VCPA-IRIV、SPA-IRIV特征变量提取联用算法,对初次提取特征变量进行“精选”,最大程度上剔除冗余波段,提高模型精度,其中,VCPA变量选择基于指数衰减函数(Uninformative variable elimination strategy of index decreasing function,EDF)、二进制矩阵采样法(Binary matrix sampling,BMS) 和模型总体分析(Model population analysis,MPA)3个重要步骤,EDF使变量空间不断收缩,剩余变量占比(ri )计算如式(4)。
每次执行EDF时,BMS和MPA也被执行,BMS生成与原数据矩阵等尺寸的二进制矩阵,变量被选与否根据BMS中“1”“0”分布决定,“1”为选择,“0”为不选,MPA利用偏最小二乘法计算子集交叉验证均方根误差,并保留该值取最小的变量子集,重复此运算过程,即得到所需波段组合,该算法能够随执行进度剔除低贡献变量,变量空间随之变小,提高高贡献变量组合出现概率,较为有效避免无效信息和干扰信息出现[15-16];SPA是一种前向特征变量选择算法,利用向量投影分析,将变量投影于其他变量上,以投影向量最大波长为选择波长,通过预设变量选择数,最终获得含最少冗余信息和最小共性的变量组合[17-18];IRIV依据模型集群分析思想,并充分考量变量间联合效应,将变量随机组合生成二进制矩阵,该矩阵行为变量组合、列为变量数,基于各行分别建立PLS模型,通过交叉验证均方根误差评估这些组合模型效果。最终,将贡献分为强信息变量、弱信息变量、无信息变量以及干扰信息变量4种,保留变量由DMEAN(Difference of mean values)及P值决定,前者为波长变量计算包含和不包含时的均方根误差均值,后者为曼-惠特尼U检验P值。执行算法多次迭代后排除无效及干扰变量,并经逆向消除从强、弱信息变量中提取获得最佳特征波段[19]。
LIUY J, CHENY H, YANGX S,et al. Research progress on nutrition and functional components of millet[J]. Cereals & Oils,2020,33(5):1-3.
[5]
WUD, SUND W. Advanced applications of hyperspectral imaging technology for food quality and safety analysis and assessment:a review:Part II:applications[J]. Innovative Food Science & Emerging Technologies,2013,19:15-28.
[6]
XIAY, XUY F, LIJ B,et al. Recent advances in emerging techniques for non-destructive detection of seed viability: A review[J]. Artificial Intelligence in Agriculture,2019,5(1):35-47.
[7]
SAHAD, MANICKAVASAGANA. Machine learning techniques for analysis of hyperspectral images to determine quality of food products:a review[J]. Current Research in Food Science,2021,4:28-44.
TIANX, LIUS C, WANGH G,et al. Application of near infrared diffuse reflectance spectroscopy in rapid detection of crude protein and starch in foxtail millet[J]. Food Science,2017,38(16):140-144.
WANGH, YUG H, HOUY,et al. Establishment and evaluation of near-infrared prediction model for selenium content in millet[J]. Journal of China Agricultural University,2021,26(10):157-163.
ZHAOH, HUANK W, ZHENGF,et al. Research on variable selection of protein in wheat near infrared spectroscopy based on latent projective graph[J]. Journal of Changchun University of Science and Technology(Natural Science Edition),2016,39(5):51-54.
YUL, ZHANGT, ZHUY X,et al. Determination of soybean leaf SPAD value using characteristic wavelength variables preferably selected by IRIV algorithm[J]. Transactions of the Chinese Society of Agricultural Engineering,2018,34(16):148-154.
YUH D. Research on wavelength selection method in near-infrared spectra and its application for the rapid detection of freshness in tilapia fillets[D]. Haikou:Hainan University,2021.
General Administration of Quality Supervision,Inspection and Quarantine of the People's Republic of China,Standardization Administration of the People's Republic of China. Oilseeds-determination of oil content:GB/T 14488.1-2008 [S]. Beijing:Standards Press of China,2009.
National Health and Family Planning Commission,State Food and Drug Administration. National food safety standards-determination of protein for food safety:GB 5009.5-2016 [S]. Beijing:Standards Press of China,2017.
[24]
WUZ S, OUYANGG Q, SHIX Y,et al. Absorption and quantitative characteristics of C-H bond and O-H bond of NIR[J]. Optics and Spectroscopy,2014,117(5):703-709.
[25]
FANN Y, MAX, LIUG S,et al. Rapid determination of TBARS content by hyperspectral imaging for evaluating lipid oxidation in mutton[J]. Journal of Food Composition and Analysis,2021(103):1-10.
[26]
WANG L, LIUG S, HEJ G,et al. Feature wavelength selection and model development for rapid determination of myoglobin content in nitrite-cured mutton using hyperspectral imaging[J]. Journal of Food Engineering,2020(287):1-9.
[27]
MARTÍNJ F G. Optical path length and wavelength selection using Vis/NIR spectroscopy for olive oil’s free acidity determination[J]. International Journal of Food Science and Technology,2015,50(6):1461-1467.
[28]
LIY. Rapid quality discrimination of grape seed oil using an extreme machine learning approach with Near-Infrared(NIR) spectroscopy[J]. Spectroscopy,2022,36:14-20.
[29]
YUNY H,BIN J, LIUD L,et al. A hybrid variable selection strategy based on continuous shrinkage of variable space in multivariate calibration[J]. Analytica Chimica Acta,2019,1058:58-69.
[30]
褚小立. 化学计量学方法与分子光谱分析技术[M]. 北京:化学工业出版社,2011.
[31]
CHUX L. Molecular spectroscopy analytical technology combined with chemometrics and its applications[M]. Beijing:Chemical Industry Press,2011.
[32]
CHANGC W, LAIRDD A, MAUSBACHM J,et al. Near-infrared reflectance spectroscopy–principal components regression analyses of soil properties[J]. Soil Science Society of America Journal,2001,65(2):480-490.
[33]
ZHANY M, GUOW C. Moisture content detection of maize seed based on visible/near-infrared and near-infrared hyperspectral imaging technology[J]. International Journal of Food Science and Technology,2020,55(2):631-640.
[34]
SUNH X, ZHANGS J, RENR,et al. Detection of soluble solids content in different cultivated fresh jujubes based on variable optimization and model update[J]. Foods,2022,11(16):2522.
LIY F, ZHANGX, CHANGD,et al. Effects of fertilization amount and patterns of sheep dung on soil properties and tobacco economic performance in continuous tobacco planting fields[J]. Journal of Shanxi Agricultural Sciences,2020,48(12):1964-1968,1983.
ZHANGH, LIANGT B, FENGW Q,et al. Hyperspectral estimation of total nitrogen and alkali hydrolysable nitrogen contents in tobacco growing soil based on successive projection algorithm[J]. Chinese Tobacco Science,2023,44(5):103-113.
YANGH, LUOB, ZHANGH,et al. Recognition of maize seed variety purity based on hyperspectral imaging technology and IRIV algorithm[J]. Journal of Jiangsu University(Natural Science Edition),2023,44(2):159-165.
[45]
ZHAOJ G, CHENN, ZHUT Y,et al. Simultaneous quantification and visualization of photosynthetic pigments in Lycopersicon esculentum mill.under different levels of nitrogen application with visible-near infrared hyperspectral imaging technology[J]. Plants,2023,12(16):2956.
LIANGK, LIUQ X, PANL Q,et al. Detection of soluble solids content in 'Korla fragrant pear' based on hyperspectral imaging and CARS-IRIV algorithm[J]. Journal of Nanjing Agricultural University,2018,41(4):760-766.