|
Registro Completo |
|
Biblioteca(s): |
Embrapa Algodão. |
|
Data corrente: |
04/02/2022 |
|
Data da última atualização: |
04/02/2022 |
|
Tipo da produção científica: |
Artigo em Periódico Indexado |
|
Autoria: |
ROCHA, P. D.; MEDEIROS, E. P. de; SILVA, C. S.; SIMÕES, S. da S. |
|
Afiliação: |
PRISCILLA DANTAS ROCHA, UEPB; EVERALDO PAULO MEDEIROS, CNPA; CAROLINA SANTOS SILVA, UNIVERSIDADE DE MALTA; SIMONE DA SILVA SIMÕES, UEPB. |
|
Título: |
Chemometric strategies for near infrared hyperspectral imaging analysis: classification of cotton seed genotypes. |
|
Ano de publicação: |
2021 |
|
Fonte/Imprenta: |
Royal Society of Chemistry, ñ p., 2021. |
|
Páginas: |
10 p. |
|
ISSN: |
0260-6291 |
|
DOI: |
10.1039/d1ay01076j |
|
Idioma: |
Inglês |
|
Conteúdo: |
Hyperspectral images have been increasingly employed in the agricultural sector for seed classification for different purposes. In the present paper we propose a new methodology based on HSI in the near infrared range (HSI-NIR) to distinguish conventional from transgenic cotton seeds. Three different chemometric approaches, one pixel-based and two object-based, using partial least squares discriminant analysis (PLS-DA) were built and their performances were compared considering the pros and cons of each approach. Specificity and sensitivity values ranged from 0.78-0.92 and 0.62-0.93, respectively, for the different approaches. Hyperspectral images (HSIs) associated with multivariate analysis (chemometrics) have gained an important role in agricultural studies recently. From satellite to lab-scale analysis, the applications of this type of image are many. The main advantage associated with the use of HSIs when compared to benchtop equipment is the localization of specific compounds provided by the spatial distribution information that is inherent to image analysis methods; this is important information for the agricultural sector. In the literature, it is possible to find a plethora of interesting applications of HSIs. Regarding applications for the quality control of food products, HSIs are commonly used to evaluate food ripeness, for food fraud/ authentication, to validate the shelf-life, to identify defects and diseases and for the quantification of pesticides. With respect to applications related to seed analysis, regression methods have been reported for quantication of specific compounds and properties, although classification methods are more oen employed to tackle other issues. For this, it is usually necessary to analyze a large number of samples quickly. This presents a drawback when using benchtop equipment, especially if the information needed is related to individual seeds. The identification of genetically modified seeds, for instance, usually requires identification by an individual seed-by-seed approach. This is due to the fact that mixing transgenic and conventional cultivars may lead to contamination or extinction of the conventional species, as in the case of transgenic cotton seeds. Environmental conditions for cotton cultivars in Brazil are extremely favorable for the occurrence of pests and diseases, encouraging the development of transgenic genotypes that are able to resist the damages caused by pests, fungi, and insects and, consequently, decreasing the need for the use of pesticides. Since DNA-based analysis to identify thosetransgenic varieties can be laborious and time-consuming, hyperspectral imaging in the near infrared range (HSI-NIR) is currently proposed as an alternative to identify these genotypes. Applications of HSI-NIR regarding the identification of infested Norway spruce seeds,6 different maize varieties and identifi- cation of ergot bodies in cereals have been reported, using different chemometric approaches. Apart from the steps related to conventional spectral analysis, the analysis of HSI datasets requires a series of essential procedures to achieve a reliable model. Among these, spatial preprocessing methods, the choice of the region of interest (ROI), masking, and binning can be mentioned. Regarding the model development per se, it is possible to observe different ways to analyze an image dataset, bearing in mind that differentapproaches will depend on the type of data available, the equipment used and the application purposes. The last factorincludes different considerations such as the analytical goal of classification and model implementation (lab, field or industry scales). For lab-scale and proof of concept research, pixel-based approaches are often employed, where each individual pixel taken from the object is used for model training and prediction. However, in certain scenarios, pixel-based approaches can be time-consuming or even prove ambiguous for a final decision. In those cases, some object-based approaches can be implemented for practical reasons, where all pixels that belong to an object are represented by a single spectrum or a set of descriptors. It is important to mention that pixel-based approaches can seldom be completely ignored during the model building step. This is due to the fact that this approach provides a better understanding of data variability and the impact of subtle variation in the performance of the final model. As an example, the need for a single-seed-identification approach led Cruz-Tirado and co-workers to evaluate and compare different classification models using a pixel-based approach training with four different object-based decision rules to classify hybrid varieties of cocoa beans. An approach developed by Soares and collaborators enabled the identification of individual seeds of four varieties of cotton from the reduction in the special variation of HSI-NIR images by averaging the pixel spectra. Given this scenario, the present work reports the development of methodologies based on HSI-NIR spectroscopy combined with supervised pattern recognition to distinguish between cotton seed genotypes of transgenic and conventional cultivars in a reliable and fast manner. Moreover, detailed discussion on different chemometric strategies for hyperspectral image analysis is also proposed to assess the suitability of this method for practical applications in industries, providing reliable classification models with objective results, and easy interpretation. MenosHyperspectral images have been increasingly employed in the agricultural sector for seed classification for different purposes. In the present paper we propose a new methodology based on HSI in the near infrared range (HSI-NIR) to distinguish conventional from transgenic cotton seeds. Three different chemometric approaches, one pixel-based and two object-based, using partial least squares discriminant analysis (PLS-DA) were built and their performances were compared considering the pros and cons of each approach. Specificity and sensitivity values ranged from 0.78-0.92 and 0.62-0.93, respectively, for the different approaches. Hyperspectral images (HSIs) associated with multivariate analysis (chemometrics) have gained an important role in agricultural studies recently. From satellite to lab-scale analysis, the applications of this type of image are many. The main advantage associated with the use of HSIs when compared to benchtop equipment is the localization of specific compounds provided by the spatial distribution information that is inherent to image analysis methods; this is important information for the agricultural sector. In the literature, it is possible to find a plethora of interesting applications of HSIs. Regarding applications for the quality control of food products, HSIs are commonly used to evaluate food ripeness, for food fraud/ authentication, to validate the shelf-life, to identify defects and diseases and for the quantification of pesticides. With respec... Mostrar Tudo |
|
Palavras-Chave: |
HSI; Imagem Hiper Espectral; Quimiometria. |
|
Thesagro: |
Algodão; Organismo Transgênico. |
|
Thesaurus Nal: |
Chemometrics; Cotton; Transgenic plants. |
|
Categoria do assunto: |
-- |
|
Marc: |
LEADER 06360naa a2200289 a 4500
001 2139690
005 2022-02-04
008 2021 bl uuuu u00u1 u #d
022 $a0260-6291
024 7 $a10.1039/d1ay01076j$2DOI
100 1 $aROCHA, P. D.
245 $aChemometric strategies for near infrared hyperspectral imaging analysis$bclassification of cotton seed genotypes.$h[electronic resource]
260 $c2021
300 $a10 p.
520 $aHyperspectral images have been increasingly employed in the agricultural sector for seed classification for different purposes. In the present paper we propose a new methodology based on HSI in the near infrared range (HSI-NIR) to distinguish conventional from transgenic cotton seeds. Three different chemometric approaches, one pixel-based and two object-based, using partial least squares discriminant analysis (PLS-DA) were built and their performances were compared considering the pros and cons of each approach. Specificity and sensitivity values ranged from 0.78-0.92 and 0.62-0.93, respectively, for the different approaches. Hyperspectral images (HSIs) associated with multivariate analysis (chemometrics) have gained an important role in agricultural studies recently. From satellite to lab-scale analysis, the applications of this type of image are many. The main advantage associated with the use of HSIs when compared to benchtop equipment is the localization of specific compounds provided by the spatial distribution information that is inherent to image analysis methods; this is important information for the agricultural sector. In the literature, it is possible to find a plethora of interesting applications of HSIs. Regarding applications for the quality control of food products, HSIs are commonly used to evaluate food ripeness, for food fraud/ authentication, to validate the shelf-life, to identify defects and diseases and for the quantification of pesticides. With respect to applications related to seed analysis, regression methods have been reported for quantication of specific compounds and properties, although classification methods are more oen employed to tackle other issues. For this, it is usually necessary to analyze a large number of samples quickly. This presents a drawback when using benchtop equipment, especially if the information needed is related to individual seeds. The identification of genetically modified seeds, for instance, usually requires identification by an individual seed-by-seed approach. This is due to the fact that mixing transgenic and conventional cultivars may lead to contamination or extinction of the conventional species, as in the case of transgenic cotton seeds. Environmental conditions for cotton cultivars in Brazil are extremely favorable for the occurrence of pests and diseases, encouraging the development of transgenic genotypes that are able to resist the damages caused by pests, fungi, and insects and, consequently, decreasing the need for the use of pesticides. Since DNA-based analysis to identify thosetransgenic varieties can be laborious and time-consuming, hyperspectral imaging in the near infrared range (HSI-NIR) is currently proposed as an alternative to identify these genotypes. Applications of HSI-NIR regarding the identification of infested Norway spruce seeds,6 different maize varieties and identifi- cation of ergot bodies in cereals have been reported, using different chemometric approaches. Apart from the steps related to conventional spectral analysis, the analysis of HSI datasets requires a series of essential procedures to achieve a reliable model. Among these, spatial preprocessing methods, the choice of the region of interest (ROI), masking, and binning can be mentioned. Regarding the model development per se, it is possible to observe different ways to analyze an image dataset, bearing in mind that differentapproaches will depend on the type of data available, the equipment used and the application purposes. The last factorincludes different considerations such as the analytical goal of classification and model implementation (lab, field or industry scales). For lab-scale and proof of concept research, pixel-based approaches are often employed, where each individual pixel taken from the object is used for model training and prediction. However, in certain scenarios, pixel-based approaches can be time-consuming or even prove ambiguous for a final decision. In those cases, some object-based approaches can be implemented for practical reasons, where all pixels that belong to an object are represented by a single spectrum or a set of descriptors. It is important to mention that pixel-based approaches can seldom be completely ignored during the model building step. This is due to the fact that this approach provides a better understanding of data variability and the impact of subtle variation in the performance of the final model. As an example, the need for a single-seed-identification approach led Cruz-Tirado and co-workers to evaluate and compare different classification models using a pixel-based approach training with four different object-based decision rules to classify hybrid varieties of cocoa beans. An approach developed by Soares and collaborators enabled the identification of individual seeds of four varieties of cotton from the reduction in the special variation of HSI-NIR images by averaging the pixel spectra. Given this scenario, the present work reports the development of methodologies based on HSI-NIR spectroscopy combined with supervised pattern recognition to distinguish between cotton seed genotypes of transgenic and conventional cultivars in a reliable and fast manner. Moreover, detailed discussion on different chemometric strategies for hyperspectral image analysis is also proposed to assess the suitability of this method for practical applications in industries, providing reliable classification models with objective results, and easy interpretation.
650 $aChemometrics
650 $aCotton
650 $aTransgenic plants
650 $aAlgodão
650 $aOrganismo Transgênico
653 $aHSI
653 $aImagem Hiper Espectral
653 $aQuimiometria
700 1 $aMEDEIROS, E. P. de
700 1 $aSILVA, C. S.
700 1 $aSIMÕES, S. da S.
773 $tRoyal Society of Chemistry, ñ p., 2021.
Download
Esconder MarcMostrar Marc Completo |
|
Registro original: |
Embrapa Algodão (CNPA) |
|
