种子质量与安全检验的高光谱成像研究进展

浏览次数：4180　发布日期：2019-8-20　来源：本站　仅供参考，谢绝转载，否则责任自负

种子质量是植物育种和生产中的一个基础性和关键性因素，可以通过种子的发芽率或理化特性来衡量，在农业领域已变得越来越重要。一方面，优质种子是植物生长的良好开端，预示着丰收；另一方面，种子质量通常与食品质量密切相关，如质地、风味和营养成分。为了满足消费者的需求，种子在收获后应谨慎加工和储存。在采收、加工和储存过程中，需要一种快速、准确、无损的检测种子质量的方法。高光谱成像作为一种非破坏性、快速的种子质量和安全性评价方法，近年来备受关注。

高光谱成像技术结合了光谱技术和成像技术的优点，可以同时获取光谱和空间信息。也就是说，它可以同时获得不均匀样品的化学信息和化学成分的空间分布。高光谱技术在农业、食品、医药等行业得到了广泛的应用。高光谱成像技术在种子行业的潜在或实际应用包括种子活性、活力、缺陷、疾病、净度检测，种子成分测定。

本文总结和分析了高光谱技术在种子质量和安全检验方面的发展，介绍了该技术在种子分类分级、活性和活力检测、损伤（缺陷和真菌）检测、净度检测和种子成分测定等方面的能力，综述了该技术在种子质量检测和安全检测中的应用，包括分析的光谱范围、样品种类、样品状态、样品数量、特征（光谱特征、图像特征、特征提取方法）、信号模式等。

表1高光谱成像应用于种子分类和分级的参考文献摘要

Seed	Varieties	Features		Data analysis strategies		Main application type	Classification result (highest accuracy)
Seed	Varieties	Spectra/image	Extraction/selection methods	Analysis level	Classification/regression methods	Main application type	Classification result (highest accuracy)
Barley, wheat and sorghum	1 variety of each kind of grain	Spectra	PCA	PWbprediction map and OWc(single kernels)	–	Grain topography classification	–
Black bean	3	Spectra and image	SPA, PCA, GLCM	OW (single kernels)	PLS-DA, SVM	Variety classification	98.33% (PLS-DA)
Grape seed	3 varieties, two growth soil	Spectra	PCA	OW (single kernels), PW PCA and prediction map	GDA	Assess Stage of maturation of grape seeds	> 95%
Grape seed	3	Spectra and image	PCA	OW (single kernels)	SVM	Variety classification	94.30%
Maize	2 (transgenic and non-transgenic)	Spectra	PCA, CARS	PW PCA and prediction map, OW (single kernels)	PLS-DA, SVM	Transgenic and non-transgenic classification	99.5% (PLS-DA)
Maize	4 varieties, 3 crop years	Spectra	no	OW (single kernels)	LS-SVM	Variety classification	91.50%
Maize	4 varieties, 3 crop years	Spectra	no	OW (single kernels)	LS-SVM	Variety classification	94.80%
Maize	4 varieties, 3 crop years	Spectra	no	OW (single kernels)	LS-SVM	Variety classification	94.40%
Maize	17	Spectra and image	PCA, SPA, GLCM, MDS	OW (single kernels)	LS-SVM	Variety classification	94.40%
Maize	18	Spectra and image	PCA	OW (single kernels), PW PCA and prediction map	PLS-DA	Textural, vitreous, floury and the third type endosperm	85% (PLS-DA)
Maize	3 hardness	Spectra and image	PCA	PW PCA and prediction map, OW (single kernels)	PLS-DA	Hardness classification	97% (PLS-DA)
Maize	14	Spectra	joint skewness-based wavelength selection	OW (single kernels)	LS-SVM	Variety classification	98.18%
Maize	3	Spectra and image	PCA	OW (single kernels)	SVM, RBFNN	Variety classification	93.85% (RBFNN)
Maize	6	Spectra and image	PCA, KPCA, GLCM	OW (bulk samples)	LS-SVM, BPNN, PCA, KPCs	Classes classification	98.89% (PCA-GLCM-LS-SVM)
Rice	4 origins	Spectra and image	PCA, GLCM	OW (single kernels)	SVM	Variety classification	91.67%
Rice	4	Spectra	PLS-DA, PCA	PW PCA and OW (bulk samples)	KNN, PLS-DA, SIMCA, SVM, RF	Seed cultivars classification	100% (SIMCA, SVM, and RF)
Soybean, maize and rice	3 of each kind of seed	Spectra	neighborhood mutual information	OW (single kernels)	ELM, RF	Variety classification	100% (ELM)
Waxy corn	4	Spectra and image	SPA, GLCM	OW (single kernels)	PLS-DA, SVM	Variety classification	98.2% (SVM)
Wheat	8	Image	WT, STEPDISC, PCA	PW and OW (bulk samples)	BPNN, LDA, QDA	Classes classification	99.1% (LDA)
Wheat	8	Spectra	STEPDISC	OW (bulk samples)	LDA, QDA, Standard BPNN, Wardnet BPNN	Variety classification	94–100% (LDA)
Wheat	5	Spectra	STEPDISC	PW PCA and OW (bulk samples)	LDA, QDA	Classes classification	90–100% (LDA)

表2 高光谱成像应用于种子活力和活力检测的参考文献摘要

Seed	Varieties	Features		Data analysis strategies		Main application type	Classification result (highest accuracy)
Seed	Varieties	Spectra/image	Extraction/selection methods	Analysis level	Classification/regression methods	Main application type	Classification result (highest accuracy)
Barley	1 variety, 8 treatments	Spectra	PCA, MNF	PWbprediction map and OWc(single kernels)	Maximum likelihood multinomial, regression classifier	Germination level detection	97% when single kernels grouped into the three categories
Corn	3 varieties, 2 treatments	Spectra	No	OW (single kernels)	PLS-DA	Viability prediction	> 95.6%
Cryptomeria japonica and Chamaecyparis obtuse	2 treatments of each kind of seed	Spectra	No	OW (single kernels)	Spectral index	Viability prediction	98.30%
Cucumber	1 variety, 2 treatments	Spectra	No	OW (single kernels), PW prediction map	PLS-DA	Viability prediction	100%
Muskmelon	1 variety, 4 treatments	Spectra	VIP, SR, and SMC	OW (single kernels)	PLS-DA	Viability prediction	94.60%
Norway spruce	1 variety, 3 treatments	Spectra and image	L1-regularized logistic regression based feature selection	OW (single kernels)	SVM	Viability prediction	> 93%
Pepper	1 variety, 2 treatments	Spectra	No	OW (single kernels), PW prediction map	PLS-DA	Germination level detection	> 85%
Tree seeds	3 varieties, 8 treatments	Spectra	LDA	OW (single kernels)	LDA	Germination level detection	> 79%
Wheat, barley and sorghum	B: 3 varieties W: 3 varieties S: 2, varieties 6 treatments	Spectra	PCA	OW (single kernels), PW prediction map	PLS-DA, PLSR	Viability prediction	R = 0.92 (PLS-DA)

表3 高光谱成像应用于种子质量缺陷检测的参考文献摘要

Seed	Varieties	Features		Data analysis strategies		Main application type	Classification result (highest accuracy)
Seed	Varieties	Spectra/image	Extraction/selection methods	Analysis level	Classification/regression methods	Main application type	Classification result (highest accuracy)
Mung bean	1 variety, 8 treatments	Spectra and image	PCA	OWb(single kernels)	LDA, QDA	Insect damage detection	> 82%
Soybean	1 variety, 5 treatments	Spectra and image	GLCM	OW (single kernels)	LDA, QDA	Insect damage detection	99% (QDA)
Wheat	1 variety, 4 insect varieties	Spectra and image	STEPDISC, GLCM, GLRM, PCA	OW (single kernels)	LDA, QDA	Insect damage detection	95.3–99.3%
Wheat	1 variety, 3 treatments	Spectra and image	PCA	PWcprediction map and OW (single kernels)	Spectral index	Seed sprouted detection	> 90%

表4 高光谱成像应用于种子真菌损伤检测的参考文献摘要

Seed	Varieties	Features		Data analysis strategies		Main application type	Classification result (highest accuracy)
Seed	Varieties	Spectra/image	Extraction/selection methods	Analysis level	Classification/regression methods	Main application type	Classification result (highest accuracy)
Barley	1 variety, 2 fungi	Spectra and image	PCA	PWbprediction map and OWc(single kernels)	LDA, QDA, MDA	Fungus (Ochratoxin A and Penicillium) damage detection	> 82%
Canola	1 variety, 2 fungi,	Spectra and image	PCA	OW (single kernels)	LDA, QDA, MDA	Fungus (Aspergillus glaucus and Penicilliumspp.) damage detection	> 90%
Corn	3 varieties, 5 treatments	Spectra	No	OW (single kernels), PW prediction map	PLS-DA	Fungus (Aflatoxin B1) damage detection	96.90%
Corn	1 variety, 3 treatments	Spectra	No	PW spectra	spectral index	Fungus (Aflatoxin A. flavus) damage detection	93%
Corn	1 variety, 3 treatments	Spectra	PCA	OW (single kernels), PW PCA	LS-SVM, KNN	Fungus (Aflatoxin A. flavus) damage detection	> 91% (KNN)
Hick peas, green peas, lentils, pinto beans and kidney beans	5 different pulses, 2 fungi	Spectra and image	PCA	OW (single kernels), PW PCA	LDA, QDA	Fungus (Penicillium commune Thom, C. and A. flavus Link, J.) damage detection	96%-100%
Maize	4 varieties	Spectra	PCA	OW (single kernels), PW prediction map	SVM, SVR	Fungus (Aflatoxin B1) damage detection	R2 = 0.77
Maize	1 variety, 5 treatments	Spectra	PCA, FDA	OW (single kernels), PW PCA	FDA	Fungus (Aflatoxin B1) damage detection	88%
Maize	1 variety, 5 treatments	Spectra	PCA	OW (single kernels)	FDA	Fungus (Aflatoxin B1) damage detection	98%
Maize	1 variety, 3 treatments	Spectra	No	OW (single kernels), PW prediction map	PLS-DA	Fungus (Fusarium) damage detection	77% (PLS-DA)
Maize	1 variety, nine treatments	Spectra	PCA, variable importance plots	OW (single kernels), PW PCA and prediction map	PLSR	Fungus damage detection	R2 = 0.87
maize	1 variety, 2 fungi, 3 treatments	Spectra	No	OW (single kernels)	discriminant analysis	Fungus (Toxigenic and atoxigenic A. flavus) damage detection	94.40%
Maize	12 varieties, 4 fungi	Spectra	PCA	OW (bulk samples), PW PCA	ANOVA, Fisher’s LSD test	Fungus (Aspergillus strains) damage detection	Fisher’s LSD test
Oat50	1 variety, 4 treatments	Spectra	PLSR	OW (single kernels), PW prediction map	PLSR, PLS-LDA	Fungus (Fusarium) damage detection	R2 = 0.8
Peanut	1 variety, 2 treatments	Spectra	PCA	OW (single kernels), PW prediction map	PCA	Moldy kernel detection	98.73%
Peanut	1 variety, 2 treatments	Spectra	ANOVA, NWFE	OW (single kernels), PW prediction map	SVM	Fungus (Aflatoxin) damage detection	> 94%
Rice	1 variety, 6 treatments	Spectra	No	OW (bulk samples)	SOM, PLSR	Fungus (Aspergillus) damage detection	R2 = 0.97
Watermelon	1 variety, 2 treatments	Spectra	Intermediate PLS (iPLS)	OW (single kernels) PW prediction map	PLS-DA, LS-SVM	Fungus (Cucumber green mottle mosaic virus) damage detection	83.3% (LS-SVM)
Watermelon	1 variety, 2 treatments	Spectra	Intermediate PLS (iPLS)	OW (single kernels), PW prediction map	PLS-DA, LS-SVM	Fungus (Acidovorax citrulli) damage detection	> 90%
Wheat	4 varieties, 2 fungi	Spectra	PCA	OW (single kernels), PW spectra	LDA	Fungus (Fusarium) damage detection	> 91%
Wheat	33 varieties, 3 treatments	Spectra	No	OW (single kernels), PW spectra	spectral index	Fungus (Fusarium head blight) damage detection	81%
Wheat	1 variety, 3 treatments	Spectra and image	PCA, STEPDISC	OW (single kernels)	LDA	Fungus (Fusarium) damage detection	92%
Wheat	1 variety, 3 fungi	Spectra and image	STEPDISC, GLCM, GLRM, PCA	OW (single kernels)	LDA, QDA, MDA	Fungus (Penicilliumspp., Aspergillus glaucus group, and Aspergillus niger) damage detection	> 95%
Wheat	3 varieties	Spectra	PCA	OW (bulk, single kernels), PW PCA	PLS-DA, iPLS-DA	Fungus (Fusarium) damage detection	99%

高光谱成像是一个复杂的、多学科的领域，其目的是在不进行单调的样品制备情况下，同时对多种化学成分和物理属性的含量和空间分布进行有效和可靠的测量，因此为种子自动分级和缺陷检测系统的设计提供了可能。本文概述的各种应用表明，在种子分级、活力和活力检测、缺陷和疾病检测、清洁度检测和种子成分测定方面，高光谱成像具有很大的应用潜力。可以预见，采用该技术的实时种子监测系统将在不久的将来满足现代种子工业控制和分选系统的需求。

全文阅读

Feng L, Zhu S, Liu F, et al, et al. Hyperspectral imaging for seed quality and safety inspection: a review. Plant Methods, 2019, 15(1): 1-25.

索取资料

来源：上海泽泉科技股份有限公司
联系电话：021-32555118
E-mail：michael.shen@zealquest.com

【点击可查看上海泽泉科技股份有限公司相关产品】

标签：种子质量安全检验高光谱成像

分享到：QQ空间新浪微博腾讯微博微信

【所有文章】【本类新闻】【相关产品】【关闭窗口】

本类文章

本类新闻