基于流形学习的红松仁脂肪近红外定量检测

doi:10.3969/j.issn.1004-1524.20221056

浙江农业学报 ›› 2023, Vol. 35 ›› Issue (8): 1915-1926.DOI: 10.3969/j.issn.1004-1524.20221056

基于流形学习的红松仁脂肪近红外定量检测

仇逊超¹^,²(), 曹军²^,^*(), 张怡卓²

1.哈尔滨金融学院计算机系,黑龙江哈尔滨 150030
2.东北林业大学机电工程学院,黑龙江哈尔滨 150040

收稿日期:2022-07-17 出版日期:2023-08-25 发布日期:2023-08-29
作者简介:仇逊超(1986—),女,黑龙江哈尔滨人,博士研究生,副教授,研究方向为农林产品无损检测、农林业机械化工程。E-mail:ldqiuxunchao@126.com
通讯作者: *曹军,E-mail:ldcaojun1956@163.com
基金资助:
国家自然科学基金(31270757);黑龙江省省属本科高校基本科研业务费项目(青年学术骨干研究项目)(2021-KYYWF-019);中央高校创新团队与重大项目培育资金项目(E2572016EBC3)

Near-infrared quantitative detection of fat in peeled Korean pine seeds based on manifold learning

QIU Xunchao¹^,²(), CAO Jun²^,^*(), ZHANG Yizhuo²

1. Department of Computer Engineering, Harbin Finance University, Harbin 150030, China
2. College of Mechanical and Electrical Engineering, Northeast Forestry University, Harbin 150040, China

Received:2022-07-17 Online:2023-08-25 Published:2023-08-29

摘要/Abstract

摘要：

红松仁脂肪的定量检测可以作为评价其食用价值和育种价值的重要指标,利用近红外光谱分析技术开展无损检测研究。在变量标准化校正+一阶导数+小波变换对原始光谱进行预处理的基础上,考虑到传统主成分分析降维方法存在对非线性复杂结构不敏感、完全去除特征间线性相关性信息的问题,分别采用流形学习中的等距映射、局部线性嵌入、改进型局部线性嵌入、局部切空间对齐、黑塞特征映射进行非线性降维,以构建的偏最小二乘为定标模型,进一步分别建立岭回归、支持向量回归、极度梯度提升数学模型。结果表明,改进型局部线性嵌入+支持向量回归建立的参数优化模型质量最佳,其降维方法优化参数为:邻域数取30,维度取16,验证集均方差均值为0.646 4,测试集实测值与预测值间的平均相对误差为0.999 2%,可见,该模型可以良好地应用到红松仁脂肪定量检测中。

关键词: 红松仁, 脂肪, 流形学习, 近红外光谱

Abstract:

Quantitative detection of fat in peeled Korean pine seeds is an important indicator of its edible and breeding value. In addition, near-infrared spectroscopy was used to detect nondestructively. Based on the result of standard normalized variate+first derivative+symlet4 (SNV+1^st-Der+Sym4) pretreatment method, considering the traditional dimensionality reduction, principal components analysis (PCA) has some problems, such as insensitive to nonlinear complex structures and removing the linear correlation information between features completely. Isometric mapping (Isomap), locally linear embedding (LLE), modified locally linear embedding (MLLE), local tangent space alignment (LTSA) and Hessian based locally linear embedding (HLLE) were used to reduce dimensions separately. Taking the model which was established by partial least square (PLS) as the calibration model. Furthermore, ridge regression (Ridge), support vector regression (SVR) and extreme gradient boosting (XGBoost) were adopted to establish mathematical models, respectively. As shown by the results, the quality of the parameter optimization model established by MLLE+SVR was the best, and the optimized parameters were as follows: neighborhood number (neighbors) was 30 and dimension (components) was 16, and the mean value of mean squared error of validation (mean-MSEV) was 0.646 4, and the mean relative error (MRE) of test set was 0.999 2%. Therefore, the MLLE+SVR model can be well applied to the quantitative detection of fat in peeled Korean pine seeds.

Key words: peeled Korean pine seeds, fat, manifold learning, near-infrared spectroscopy

中图分类号:

S126
S58

仇逊超, 曹军, 张怡卓. 基于流形学习的红松仁脂肪近红外定量检测[J]. 浙江农业学报, 2023, 35(8): 1915-1926.

QIU Xunchao, CAO Jun, ZHANG Yizhuo. Near-infrared quantitative detection of fat in peeled Korean pine seeds based on manifold learning[J]. Acta Agriculturae Zhejiangensis, 2023, 35(8): 1915-1926.

图/表 13

图1 近红外光谱采集系统示意图

Fig.1 Schematic diagram of near-infrared spectrum acquisition system

图2 红松仁原始近红外光谱曲线图

Fig.2 Original near-infrared spectrum curve graph of peeled Korean pine seeds

图3 红松仁样本脂肪含量分布情况 61.14%为均值-标准差,65.58%为均值+标准差。

Fig.3 Distribution of fat content in peeled Korean pine seeds 61.14% was the result of mean minus standard deviation, and 65.58% was the result of mean plus standard deviation.

表1 十次红松仁脂肪训练集和验证集切分结果

Table 1 Ten times segmentation results of fat in peeled Korean pine seeds’ training and validation sets

切分次数 Number of segmentation	样本集 Sample set	脂肪Fat/%
切分次数 Number of segmentation	样本集 Sample set	最大值 Maximum	最小值 Minimum	均值 Mean	标准差 Standard deviation
1	训练集Training set	69.93	60.04	63.48	2.27
	验证集Validation set	67.64	60.24	62.86	1.90
2	训练集Training set	69.93	60.04	63.25	2.19
	验证集Validation set	68.87	60.46	63.78	2.27
3	训练集Training set	69.93	60.04	63.33	2.23
	验证集Validation set	69.59	60.55	63.48	2.17
4	训练集Training set	69.93	60.04	63.26	2.20
	验证集Validation set	69.21	60.40	63.75	2.24
5	训练集Training set	69.93	60.04	63.36	2.17
	验证集Validation set	69.59	60.23	63.35	2.46
6	训练集Training set	69.93	60.04	63.25	2.57
	验证集Validation set	69.59	60.37	63.79	2.57
7	训练集Training set	69.93	60.04	63.34	2.24
	验证集Validation set	68.18	60.23	63.45	2.11
8	训练集Training set	69.93	60.04	63.45	2.22
	验证集Validation set	68.87	60.37	62.99	2.15
9	训练集Training set	69.93	60.04	63.38	2.22
	验证集Validation set	68.26	60.24	63.29	2.21
10	训练集Training set	69.93	60.04	63.34	2.15
	验证集Validation set	69.59	60.23	63.43	2.53

图4 变量标准化校正+一阶导数+紧支集正交小波变换预处理后红松仁光谱曲线图

Fig.4 Spectrum curve graph of peeled Korean pine seeds after standard normalized variate+first derivative+orthogonal and compactly supported wavelet transformation pretreatment

图5 滤波前后红松仁光谱对比

Fig.5 Spectral comparison of peeled Korean pine seeds before and after filtering

图6 预处理后光谱特征热度图

Fig.6 The spectral signatures’ heat map after preprocessing

图7 Partial least square参数优化验证集均方差均值对比情况

Fig.7 Comparison for partial least square’s parameter optimization of validation sets’ mean value of mean squared error

图8 全波段模型验证集均方差均值比较

Fig.8 Comparison for full wavelengths of validation sets’ mean value of mean squared error

图9 不同降维、建模方法及参数验证集均方差均值比较 mean-MSEV,验证集均方差均值;Ridge,岭回归;SVR,支持向量回归;XGBoost,极度梯度提升;PCA,主成分分析;Isomap,等距映射;LLE,局部线性嵌入;MLLE,改进型局部线性嵌入;LTSA,局部切空间对齐;HLLE,黑塞特征映射。

Fig.9 Comparison for different dimension reduction, modeling methods and parameters of validation sets’ mean value of mean squared error mean-MSEV, Mean value of mean squared error of validation; Ridge, Ridge regression; SVR, Support vector regression; XGBoost, Extreme gradient boosting; PCA, Principal components analysis; Isomap, Isometric mapping; LLE, Locally linear embedding; MLLE, Modified locally linear embedding; LTSA, Local tangent space alignment; HLLE, Hessian based locally linear embedding.

图10 不同降维方法后光谱特征热度图对比 PCA,主成分分析;Isomap,等距映射;LLE,局部线性嵌入;MLLE,改进型局部线性嵌入;LTSA,局部切空间对齐;HLLE,黑塞特征映射。

Fig.10 The comparison of spectral signatures’ heat maps after different dimension reduction PCA, Principal components analysis; Isomap, Isometric mapping; LLE, Locally linear embedding; MLLE, Modified locally linear embedding; LTSA, Local tangent space alignment; HLLE, Hessian based locally linear embedding.

表2 各最优参数模型评价指标比较情况

Table 2 The evaluating indicator comparison of optimal parameter models

模型 Model	最优参数 Optimal parameter	维度 Dimension	mean-MSEV	mean-PCCV
PLS	components=9	9	1.519 2	0.813 3
PCA+Ridge	contribution=0.99	15/16	1.771 2	0.788 2
PCA+SVR	contribution=0.85	3	1.481 9	0.824 0
PCA+XGBoost	contribution=0.96	7	1.887 9	0.745 8
Isomap+Ridge	components=18, neighbors=50	18	1.413 3	0.833 5
Isomap+SVR	components=3, neighbors=90	3	1.436 9	0.825 4
Isomap+XGBoost	components=10, neighbors=100	10	1.403 7	0.838 2
LLE+Ridge	components=18, neighbors=70	18	0.797 4	0.874 7
LLE+SVR	components=16, neighbors=60	16	0.682 2	0.883 7
LLE+XGBoost	components=14, neighbors=60	14	0.989 5	0.851 3
MLLE+Ridge	components=18, neighbors=30	18	0.709 3	0.881 7
MLLE+SVR	components=16, neighbors=30	16	0.646 4	0.914 5
MLLE+XGBoost	components=16, neighbors=80	16	1.072 7	0.849 4
LTSA+Ridge	components=18, neighbors=60	18	0.759 9	0.879 6
LTSA+SVR	components=16, neighbors=20	16	0.743 5	0.880 2
LTSA+XGBoost	components=8, neighbors=70	8	1.193 7	0.841 8
HLLE+Ridge	components=10,neighbors=80	10	0.979 4	0.867 9
HLLE+SVR	components=8, neighbors=50	8	0.830 8	0.874 1
HLLE+XGBoost	components=8, neighbors=70	8	1.193 7	0.841 8

图11 红松仁脂肪实测值与预测值散点分布情况

Fig.11 Scatter distribution of fat in peeled Korean pine seeds’ measured and predicted values

参考文献 25

[1]	孟祥茹, 胡乐乾, 琚荧, 等. 食用油中多环芳烃检测的前处理方法研究进展[J]. 食品科学, 2022, 43(11): 373-382.
	MENG X R, HU L Q, JU Y, et al. Recent advances in pretreatment methods for the detection of polycyclic aromatic hydrocarbons in edible oil[J]. Food Science, 2022, 43(11): 373-382. (in Chinese with English abstract)
[2]	FERREIRA D S, GALÃO O F, PALLONE J A L, et al. Comparison and application of near-infrared (NIR) and mid-infrared (MIR) spectroscopy for determination of quality parameters in soybean samples[J]. Food Control, 2014, 35(1): 227-232.
[3]	FERREIRA D, PALLONE J, POPPI R. Comparison and application of near-infrared (NIR) and mid-infrared (MIR) spectroscopy for determination of quality parameters in soybean samples[J]. Food Control, 2014, 35(1): 227.
[4]	YıLMAZ N. Middle infrared stabilization of individual rice bran milling fractions[J]. Food Chemistry, 2016, 190: 179-185.
[5]	XIE A G, SUN D W, XU Z Y, et al. Rapid detection of frozen pork quality without thawing by Vis-NIR hyperspectral imaging technique[J]. Talanta, 2015, 139: 208-215.
[6]	LÉNÁRT J, SZIGEDI T, DERNOVICS M, et al. Application of FT-NIR spectroscopy on the determination of the fat and protein contents of lyophilized cheeses[J]. Acta Alimentaria, 2012, 41(3): 351-362.
[7]	刘丽娜, 马世伟, 芮玲. 基于可信赖性和连续性的流形降维效果评价方法[J]. 计算机应用研究, 2018, 35(6): 1707-1711.
	LIU L N, MA S W, RUI L. Evaluation method for manifold dimensionality reduction effect based on trustworthiness and continuity[J]. Application Research of Computers, 2018, 35(6): 1707-1711. (in Chinese with English abstract)
[8]	吕飞, 韩敏. 基于深度极限学习机的高光谱遥感影像分类研究[J]. 大连理工大学学报, 2018, 58(2): 166-173.
	LYU F, HAN M. Hyperspectral remote sensing image classification based on deep extreme learning machine[J]. Journal of Dalian University of Technology, 2018, 58(2): 166-173. (in Chinese with English abstract)
[9]	李路, 黄汉英, 赵思明, 等. 大米蛋白质、脂肪、总糖、水分近红外检测模型研究[J]. 中国粮油学报, 2017, 32(7): 121-126.
	LI L, HUANG H Y, ZHAO S M, et al. NIR spectra detection model of protein, fat, total sugar and moisture in rice[J]. Journal of the Chinese Cereals and Oils Association, 2017, 32(7): 121-126. (in Chinese with English abstract)
[10]	周菲, 王文军, 刘岩, 等. 向日葵籽仁脂肪和脂肪酸含量近红外光谱模型的建立[J]. 作物杂志, 2021(2): 200-206.
	ZHOU F, WANG W J, LIU Y, et al. Establishment of near-infrared spectroscopy model for the contents of fat and fatty acids in sunflower husked seeds[J]. Crops, 2021(2): 200-206. (in Chinese with English abstract)
[11]	LOPEZ E, GONZALEZ D, AGUADO J V, et al. A manifold learning approach for integrated computational materials engineering[J]. Archives of Computational Methods in Engineering, 2018, 25(1): 59-68.
[12]	LIANG Y M, SHIH S W, SHIH A C C. Human action segmentation and classification based on the Isomap algorithm[J]. Multimedia Tools and Applications, 2013, 62(3): 561-580.
[13]	杨福臣, 孙兆远, 孙芝杨. 乳粉中蛋白质和脂肪近红外测定模型研究[J]. 食品研究与开发, 2017, 38(2): 169-173.
	YANG F C, SUN Z Y, SUN Z Y. Research on near infrared determination model of milk powder protein and fat[J]. Food Research and Development, 2017, 38(2): 169-173. (in Chinese with English abstract)
[14]	王海莲, 万向元, 胡培松, 等. 稻米脂肪含量近红外光谱分析技术研究[J]. 中国农业科学, 2005, 38(8): 1540-1546.
	WANG H L, WAN X Y, HU P S, et al. Quantitative analysis of fat content in brown rice by near infrared spectroscopy (NIRS) technique[J]. Scientia Agricultura Sinica, 2005, 38(8): 1540-1546. (in Chinese with English abstract)
[15]	李跑, 吴红艳, 李尚科, 等. 近红外光谱技术结合化学计量方法用于大米的快速分析[J]. 食品研究与开发, 2018, 39(19): 117-124.
	LI P, WU H Y, LI S K, et al. Near infrared spectroscopy combined with chemometrics method for rapid analysis of rice[J]. Food Research and Development, 2018, 39(19): 117-124. (in Chinese with English abstract)
[16]	张中卫, 温志渝, 曾甜玲, 等. 微型近红外光纤光谱仪用于奶粉中蛋白质脂肪的定量检测研究[J]. 光谱学与光谱分析, 2013, 33(7): 1796-1800.
	ZHANG Z W, WEN Z Y, ZENG T L, et al. Miniature near-infrared fibre optic spectrometer for the quantitative detection of protein and fat in milk powder[J]. Spectroscopy and Spectral Analysis, 2013, 33(7): 1796-1800. (in Chinese with English abstract)
[17]	洛曲, 于修烛, 张建新, 等. 基于近红外光谱的藏区酥油脂肪和蛋白质含量快速检测分析[J]. 中国油脂, 2018, 43(3): 136-140.
	LUO Q, YU X Z, ZHANG J X, et al. Rapid determination of fat and protein contents in ghee using near-infrared spectroscopy[J]. China Oils and Fats, 2018, 43(3): 136-140. (in Chinese with English abstract)
[18]	雷雨梦, 闫国超, 杨静, 等. 青花菜中矿质元素的近红外光谱快速测定[J]. 浙江农业学报, 2022, 34(3): 548-556.
	LEI Y M, YAN G C, YANG J, et al. Rapid determination of mineral elements in broccoli by near-infrared spectroscopy[J]. Acta Agriculturae Zhejiangensis, 2022, 34(3): 548-556. (in Chinese with English abstract)
[19]	谢德红, 李俊锋, 刘菂, 等. 基于改进Hodrick-Prescott分解模型的近红外自适应降噪方法[J]. 光谱学与光谱分析, 2020, 40(5): 1650-1655.
	XIE D H, LI J F, LIU D, et al. An improved hodrick-prescott decomposition based near-infrared adaptive denoising method[J]. Spectroscopy and Spectral Analysis, 2020, 40(5): 1650-1655. (in Chinese with English abstract)
[20]	李耀翔, 李颖, 姜立春. 基于小波压缩的木材密度近红外光谱的预处理研究[J]. 北京林业大学学报, 2016, 38(3): 89-94.
	LI Y X, LI Y, JIANG L C. Pretreatment of near-infrared spectroscopy of wood based on wavelet compression[J]. Journal of Beijing Forestry University, 2016, 38(3): 89-94. (in Chinese with English abstract)
[21]	CORTÉS V, RODRÍGUEZ A, BLASCO J, et al. Prediction of the level of astringency in persimmon using visible and near-infrared spectroscopy[J]. Journal of Food Engineering, 2017, 204: 27-37.
[22]	ZHANG Y C, DUCHI J C, WAINWRIGHT M J. Divide and conquer kernel ridge regression: a distributed algorithm with minimax optimal rates[EB/OL]. 2013: arXiv: 1305.5029. https://arxiv.org/abs/1305.5029.
[23]	CALDERÓN R, NAVAS-CORTÉS J, ZARCO-TEJADA P. Early detection and quantification of Verticillium wilt in olive using hyperspectral and thermal imagery over large areas[J]. Remote Sensing, 2015, 7(5): 5584-5610.
[24]	LI K, RÜDIGER H, ZIEMSSEN T. Spectral analysis of heart rate variability: time window matters[J]. Frontiers in Neurology, 2019, 10: 545.
[25]	付波霖, 邓良超, 张丽, 等. 联合星载高光谱影像和堆栈集成学习回归算法的红树林冠层叶绿素含量遥感反演[J]. 遥感学报, 2022, 26(6): 1182-1205.
	FU B L, DENG L C, ZHANG L, et al. Estimation of mangrove canopy chlorophyll content using hyperspectral image and stacking ensemble regression algorithm[J]. National Remote Sensing Bulletin, 2022, 26(6): 1182-1205. (in Chinese with English abstract)

基于流形学习的红松仁脂肪近红外定量检测

Near-infrared quantitative detection of fat in peeled Korean pine seeds based on manifold learning

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 13

参考文献 25

相关文章 15

编辑推荐

Metrics

本文评价

[1]	宋雅萍, 雷召雄, 赵毅昂, 姜超, 王兴平, 罗仍卓么, 马云, 魏大为. 牛FoxO1基因CDS区克隆及其在脂肪细胞分化过程中的表达分析[J]. 浙江农业学报, 2023, 35(5): 1016-1027.
[2]	宋碧清, 杨晓东, 郑昀晔, 王国平, 徐盛春, 赵燕, 赵珊珊, 马宇轩, 李素娟. 不同烟草品种烟籽油理化性质及脂肪酸和挥发性成分评价[J]. 浙江农业学报, 2022, 34(6): 1152-1161.
[3]	李燕乐, 钟怀荣, 宣宁, 张燕, 陈高, 季祥. 磷酸泛酰巯基乙胺基转移酶基因过量表达对集胞藻PCC6803脂肪酸合成的影响[J]. 浙江农业学报, 2022, 34(6): 1316-1325.
[4]	李文略, 骆霞虹, 柳婷婷, 金关荣, 葛亚英, 陈常理, 安霞. 不同类型向日葵籽粒的理化性质[J]. 浙江农业学报, 2022, 34(4): 671-677.
[5]	雷雨梦, 闫国超, 杨静, 朱祝军. 青花菜中矿质元素的近红外光谱快速测定[J]. 浙江农业学报, 2022, 34(3): 548-556.
[6]	林雨晴, 陆胜民, 周万怡, 邢建荣, 杨颖. 铁皮石斛叶多糖结构及其益生性质初探[J]. 浙江农业学报, 2022, 34(11): 2504-2511.
[7]	汤佳宁, 王永侠, 刘金松, 曾新福, 杨彩梅. 中链脂肪酸及其酯对致病菌的抑菌作用[J]. 浙江农业学报, 2021, 33(9): 1611-1616.
[8]	胡晓丹, 王来娣, 杨婷, 白皓, 江勇, 王志秀, 王莎莎, 陈国宏, 常国斌. 鸭脂肪细胞中circ-FBLN2鉴定与功能预测[J]. 浙江农业学报, 2021, 33(4): 602-609.
[9]	杜金梁, 曹丽萍, 贾睿, 顾郑琰, 何勤, 徐跑, JENEYGalina, 马玉忠, 殷国俊. 甘草总黄酮对高脂条件下罗非鱼肝损伤的保护作用[J]. 浙江农业学报, 2021, 33(10): 1826-1835.
[10]	路晓媛, 钟怀荣, 夏志洁, 曹月蕾, 陈高, 戴美学. 集胞藻酰基载体蛋白基因过量表达对脂肪酸合成的影响[J]. 浙江农业学报, 2020, 32(7): 1253-1262.
[11]	李美霖, 陈宇眺, 洪晓富, 乔宇颖, 王青霞, 陈喜靖, 沈阿林, 喻曼. 不同氮肥管理方式对稻田土壤微生物群落结构的影响[J]. 浙江农业学报, 2020, 32(2): 308-316.
[12]	乔木, 黄藏宇, 吴俊静, 武华玉, 万绪玲, 周佳伟, 刘贵生, 梅书棋, 彭先文. 猪OLR1基因对肌内前脂肪细胞分化的影响[J]. 浙江农业学报, 2020, 32(12): 2147-2153.
[13]	邵嘉皓, 王杰, 贾先波, 陈仕毅, 赖松家. 家兔中microRNA的研究进展[J]. 浙江农业学报, 2020, 32(1): 176-182.
[14]	朱颍琨, 肖劲邦, 钱柏霖, 姜思汛, 尤留超, 张钺, 刘红, 马莉, 曹随忠, 余树民, 沈留红. 泌乳初期奶牛相关脂肪因子及生理生化指标与脂肪肝的相关性[J]. 浙江农业学报, 2019, 31(5): 722-729.
[15]	王帅帅, 张才, 孟素丹, 黄永志, 朱雪敏, 王玉琴, 杨自军. 犊牛脂肪干细胞的分离、培养与鉴定[J]. 浙江农业学报, 2019, 31(2): 235-241.