Non-destructive evaluation of external and internal table grape quality
| dc.contributor.advisor | Opara, Umezuruike Linus | en_ZA |
| dc.contributor.advisor | Nieuwoudt, Hélène H | en_ZA |
| dc.contributor.author | Daniels, Andries Jerrick | en_ZA |
| dc.contributor.other | Stellenbosch University. Faculty of AgriSciences. Dept. of Viticulture and Oenology. | en_ZA |
| dc.date.accessioned | 2021-05-24T11:31:07Z | |
| dc.date.available | 2021-05-24T11:31:07Z | |
| dc.date.issued | 2021-03 | |
| dc.description | Thesis (PhDAgric)--Stellenbosch University, 2021. | en_ZA |
| dc.description.abstract | ENGLISH ABSTRACT: Determining the correct harvest maturity parameters of table grapes is an essential step before harvesting. The chemical analysis of table grapes to determine harvest and quality parameters such as total soluble solids (TSS), titratable acidity (TA) and pH, is very time-consuming, expensive, and destructive. Developing faster and more cost-effective methods to obtain the information can benefit the table grape industry by reducing losses suffered at the postharvest stage. There are multitudes of factors that can influence table grape postharvest quality leading to huge losses. These losses are exacerbated even further by the long list of postharvest external and internal defects that can occur, including browning in all its various manifestations. The application of cutting-edge technologies such as Fourier Transform Near-Infrared (FT-NIR) spectroscopy that can accurately assess the external and internal quality of fruit is, therefore, essential. This particularly concerns the identification of defects or assessment of the risks of defects that are likely to develop during post storage. The aim of this application would thus be to evaluate these new technologies to monitor table grape quality non-destructively, before, during, and/or after harvest. This study, therefore, focussed on the development and optimisation of faster, cost- effective, and fit-for-purpose methods to monitor harvest maturity and quality of table grapes in the vineyard before harvesting and during packaging and cold storage. Harvest of three different cultivars, namely, Thompson Seedless, Regal Seedless and Prime, happened over two seasons (2016 and 2017) from six different commercial vineyards. Five of these vineyards were in the Western Cape (two in the Hex River Valley, three in Wellington) and one in the Northern Cape (Kakamas), South Africa. Harvest occurred twice at each vineyard, at optimum ripeness and two weeks later (after the optimum harvest date). The incidence and intensity of browning on each berry on a bunch were evaluated for different defects and browning phenotypes. Quantitative harvest maturity and indicative quality parameters such as TSS, TA and pH, as well as the sensory-related parameters – sugar:acid ratio (TSS:TA ratio) and BrimA, were investigated by scanning whole table grape bunches contactless with Bruker’s MATRIX-F spectrometer in the laboratory. Partial Least Squares (PLS) regression was used to build prediction models for each parameter. Two different infrared spectrometers, namely the Bruker Multipurpose Analyser Fourier Transform Near-Infrared (MPA FT-NIR) and MicroNIR Pro 1700 were also used to determine TSS on whole table grape berries. The MicroNIR Pro 1700 was utilised in the vineyard and the laboratory and the MPA only in the laboratory. The same spectral dataset used to build the quantitative models was used to build classification models for two browning phenotypes, namely chocolate browning and friction browning. Partial Least Squares Discriminant Analysis (PLS-DA) and Artificial Neural Networks (ANN) were used for the classification tasks. Key results showed that the incidence and intensity of different defects and browning phenotypes such as sulphur dioxide (SO2) damage were prevalent on all three white seedless table grape cultivars. The incidences of fungal infection, sunburn and abrasion damage were high on Regal Seedless and Thompson Seedless in 2016. Contact browning, mottled browning and friction browning and bruising damage had higher incidences in 2017 than in 2016. Overall, the intensity of defects was very high in 2016 except on Regal Seedless from Hex River Valley. Prime from Kakamas and Wellington had the highest intensity of defects in 2017, which appeared on the grapes after 7 weeks of cold storage. Prediction models were successfully developed for TSS, TA, TSS:TA, pH, and BrimA minus acids on intact table grape bunches using FT-NIR spectroscopy in a contactless measurement mode, and applying spectral pre-processing techniques for regression analysis with PLS. The combination of Savitzky-Golay first derivative coupled with multiplicative scatter correction on the original spectra delivered the best models. Statistical indicators used to evaluate the models were the number of latent variables (LV) used to build the model, the prediction correlation coefficient (R2p) and root mean square error of prediction (RMSE). For the respective parameters TSS, TA, TSS:TA ratio, pH, and BrimA, the number of LV used when the models were build according to a random split of the calibration and validation set were 6, 4, 5, 5 and 10, the R2p = 0.81, 0.43, 0.66, 0.27, and 0.71, and the RMSEP = 1.30 °Brix, 1.09 g/L, 7.08, 0.14, and 1.80. When 2016 was used as the calibration set and 2017 as the validation set in model building the number of LV used were 9, 5, 5, 4 and the R2p = 0.44, 0.06, 0.17, 0.05, and 0.05 and the RMSEP = 3.22 °Brix, 2.41 g/L, 14.53, 0.21, and 8.03 for for the respective parameters. Determining TSS of whole table grape berries in the vineyard before and after harvesting using handheld and benchtop spectrometers on intact table grape berries showed that spectra taken in the laboratory with the MicroNIR were more homogenous than those taken in the vineyard with the same spectrometer, over the two years investigated. The results obtained with the MPA were not as good as those obtained with the MicroNIR in the laboratory were. The model constructed with the combined data of 2016 and 2017 taken in the laboratory with the MicroNIR had the best statistics in terms of R2p (0.74) and RPDp (1.97). The model constructed with the 2017 data obtained in the laboratory with the MicroNIR had the lowest prediction error (RMSEP = 1.13°Brix). Good models were obtained using PLS-DA and ANN to classify bunches as either clear or as having chocolate browning and friction browning based on the spectra obtained from intact table grape bunches with the MATRIX-F spectrometer. The classification error rate (CER), specificity and sensitivity were used to evaluate the models constructed using PLS-DA and the kappa score was used for ANN. The CER for chocolate browning (25%) was better than that of friction browning (46%) after Weeks 3 and 4 in cold storage for both class 0 (absence of browning) and class 1 (presence of browning). Both the specificity and sensitivity of class 0 and class 1 of friction browning were not as good as for chocolate browning. With ANN, the testing kappa score to classify table grape bunches as clear or having chocolate browning or friction browning showed that chocolate browning could be classified with the strong agreement after Weeks 3 and 4 and Weeks 5 and 6 and that friction browning could be classified with moderate agreement after three and four weeks in cold storage. Classification of chocolate browning and friction browning phenotypes was done using PLS-DA and ANN and the result showed that both types of browning can be classified with moderate agreement. The implications of the results of this study for the table grape industry are that the industry can move beyond just assessing methods and techniques in the laboratory towards implementation in the vineyard and the packhouse. Much quicker decisions regarding grape quality and destination of export can now be made using a combination of the MicroNIR handheld and MATRIX-F instruments for onsite quality measurement and the models to predict internal (e.g. TSS) and external (browning) quality attributes. | en_ZA |
| dc.description.abstract | AFRIKAANSE OPSOMMING: Die bepaling van die korrekte oesrypheidsparameters van tafeldruiwe is 'n noodsaaklike stap voor oes. Chemiese ontleding van tafeldruiwe om oes- en kwaliteitsparameters te bepaal, soos totale oplosbare vaste stowwe (TOVS), titreerbare suur (TS) en pH, is baie tydrowend, duur en vernietigend. Die ontwikkeling van vinniger en kostedoeltreffender maniere om die inligting te bekom, kan die tafeldruifbedryf bevoordeel deur verliese wat in die na-oesstadium gely word, te verminder. Dit sluit die menigte faktore in wat die gehalte van tafeldruiwe ná oes kan beïnvloed en tot verliese lui. Hierdie verliese word nog verder vererger deur die lang lys van verskillende na-oes-verwante gebreke wat kan voorkom, insluitend verbruining in al sy verskillende manifestasies. Die toepassing van toonaangewende tegnologieë soos Fourier-transform-naby- infrarooi (FT-NIR) spektroskopie wat die eksterne en interne kwaliteit van vrugte akkuraat kan beoordeel, is dus noodsaaklik. Dit is veral die identifisering van gebreke, of die beoordeling van die risiko's van gebreke, wat waarskynlik tydens die opberging kan ontstaan. Die doel van hierdie toepassing was dus om hierdie nuwe tegnologieë te evalueer om die kwaliteit van tafeldruiwe nie-vernietigend te monitor, voor, tydens en/of ná oes. Hierdie studie het dus gefokus op die ontwikkeling en optimalisering van vinniger, koste- effektiewe en geskikte doeleindes om oesrypheid en kwaliteit van tafeldruiwe in die wingerd te monitor voor oes en tydens verpakking en koelopberging. Druiwe-oes van drie verskillende kultivars (Thompson Seedless, Regal Seedless en Prime) het gedurende twee jare (2016 en 2017) uit ses verskillende kommersiële wingerde plaasgevind. Vyf van hierdie wingerde was in die Wes-Kaap (twee in die Hexriviervallei, drie in Wellington) en een in die Noord-Kaap (Kakamas), Suid-Afrika. Die oes het twee keer by elke wingerd plaasgevind, dit wil sê op die beste rypheid en twee weke later ná die optimale oesdatum. Die voorkoms en intensiteit van verbruining op elke korrel op 'n tros is op verskillende defekte en verbruiningsfenotipes geëvalueer. Kwantitatiewe oesrypheid en kwaliteitsindikatiewe parameters, naamlik TOVS, TS en pH, sowel as sensoriese verwante parameters suiker:suur-verhouding (TOVS:TS- verhouding) en BrimA is ondersoek deur heel tafeldruiftrosse sonder kontak met die Bruker se MATRIX-F-spektrometer in die laboratorium te skandeer. Gedeeltelike minste kwadrate (GMK) regressie is gebruik om modelle vir die parameters te bou. Twee verskillende infrarooi- spektrometers naamlik (a) die Bruker Multipurpose Analyzer Fourier Transform Near-Infrared (MPA FT-NIR) en (b) MicroNIR Pro 1700 is ook gebruik om TOVS op heel tafeldruifkorrels te bepaal. Die MicroNIR Pro 1700 is in die wingerd en in die laboratorium gebruik en die MPA slegs in die laboratorium. Met behulp van dieselfde spektrale datastel as die een wat gebruik word om die kwantitatiewe modelle op te stel, is klassifikasiemodelle vir twee verskillende verbruiningsfenotipes (sjokoladeverbruining en wrywingverbruining) gebou. Hierdie keer is gedeeltelike minste-kwadrate-diskriminant-analise (GMK-DA) en kunsmatige neurale netwerke (KNN) gebruik. Die belangrike resultate het getoon dat die voorkoms en intensiteit van verskillende defekte en verbruiningsfenotipes soos swaeldioksied (SO2)-skade op al drie wit pitlose tafeldruifkultivars voorgekom het. Die voorkoms van swaminfeksie, sonbrand en skaafskuur was hoog op Regal Seedless en Thompson Seedless in 2016. Kontak-, gevlekte- en wrywing verbruining sowel as kneusplekke het in 2017 'n hoër voorkoms as in 2016 gehad. Oor die algemeen was die intensiteit van defekte baie hoog in 2016 behalwe op Regal Seedless vanaf die Hexriviervallei. Prime van Kakamas en Wellington het in 2017 die hoogste intensiteit van gebreke gehad wat ná 7 weke se koelopberging op die druiwe verskyn het. Die suksesvolle ontwikkeling van modelle vir TOVS, TS, TOVS:TS verhouding, pH en BrimA op heel tafeldruiftrosse met behulp van FT-NIR-spektroskopie is bewys as inderdaad moontlik – veral as GMK met verskillende spektrale voorverwerkingstegnieke gepaard gaan. Statistiese aanwysers wat gebruik is om die modelle te evalueer, was die aantal latente veranderlikes (LV) wat gebruik is om die model te bou, die voorspellingskorrelasiekoëffisiënt (R2p) en wortelgemiddelde vierkante voorspellingsfout (WGVVF). Die kombinasie van die eerste afgeleide Savitzky-Golay tesame met die vermenigvuldigende verstrooiingskorreksie op die oorspronklike spektra het die beste modelle gelewer. Statistiese aanwysers wat gebruik is om die modelle te evalueer, was die aantal latente veranderlikes (LV) wat gebruik is om die model te bou, die voorspellingskorrelasiekoëffisiënt (R2p) en wortelgemiddelde vierkante voorspellingsfout (RMSE). Vir die onderskeie parameters TSS, TA, TSS: TA-verhouding, pH en BrimA, was die aantal LV wat gebruik is toe die modelle volgens 'n ewekansige verdeling van die kalibrasie- en valideringstel gebou is, 6, 4, 5, 5 en 10, die R2p = 0,81, 0,43, 0,66, 0,27 en 0,71, en die RMSEP = 1,30 ° Brix, 1,09 g / l, 7,08, 0,14 en 1,80. Toe 2016 as die kalibrasiestel gebruik is en 2017 as die validasieset in modelbou, was die aantal gebruikte LV 9, 5, 5, 4 en die R2p = 0,44, 0,06, 0,17, 0,05 en 0,05 en die RMSEP = 3,22 ° Brix, 2,41 g / l, 14,53, 0,21 en 8,03 vir die onderskeie parameters. Die bepaling van TOVS van heel tafeldruifkorrels in die wingerd voor en ná oes oor twee jaar met behulp van hand- en tafelbladspektrometers het getoon dat spektra wat in die laboratorium met die MicroNIR geneem is meer homogeen was as dié wat in die wingerd met dieselfde spektrometer geneem is. Die resultate wat met die MPA behaal is, was nie so goed soos met die MicroNIR in die laboratorium nie. Die model wat saamgestel is met die gekombineerde data van 2016 en 2017 wat in die laboratorium met die MicroNIR geneem is, het die beste statistieke gehad in terme van die R2p (0.74) en die RPDp (1.97). Die model wat opgestel is met die 2017 data wat in die laboratorium met die MicroNIR verkry is, het die laagste voorspellingsfout (RMSEP = 1.13°Brix) gehad. Goeie modelle is verkry met behulp van GMK-DA en KNN om trosse as skoon te klassifiseer, of as sjokoladeverbruining en wrywingsverbruining gebaseer op die spektra van die heel tafeldruiftrosse wat met die MATRIX-F-spektrometer geneem is. Die klassifikasiesyfer (KS), spesifisiteit en sensitiwiteit is gebruik om die modelle wat met behulp van GMK-DA saamgestel is, te evalueer en die kappa-telling is vir KNN gebruik. Die KS vir sjokoladeverbruining (25%) was beter as dié van wrywingsverbruining (46%) vir week 3 en week 4 vir beide klas 0 (afwesigheid van verbruining) en klas 1 (teenwoordigheid van verbruining). Beide die spesifisiteit en sensitiwiteit van klas 0 en klas 1 vir wrywingverbruining was nie so goed soos vir sjokoladeverbruining nie. Met KNN het die toetskappa-telling om tafeldruiftrosse as skoon of sjokoladeverbruining of wrywingsverbruining te klassifiseer, getoon dat sjokoladeverbruining tydens Week 3 en Week 4 en Week 5 en Week 6 met 'n matige ooreenstemming geklassifiseer kan word en dat wrywingsverbruining met matige ooreenstemming tydens Week 3 en Week 4 geklassifiseer kan word. Die implikasies van hierdie resultate vir die tafeldruifbedryf is van so 'n aard dat die bedryf nou verder kan gaan as om net metodes en tegnieke in die laboratorium te beoordeel, maar kan beweeg na implementering in die wingerd en die pakhuis. Die neem van baie vinniger besluite rakende die kwaliteit van die druiwe, dit wil sê in watter klas druiwe geplaas kan word en na watter uitvoermark druiwe gestuur kan word, is nou moontlik. Veel vinniger besluite rakende druiwekwaliteit en bestemming van uitvoer kan nou geneem word met behulp van 'n kombinasie van die MicroNIR-hand- en MATRIX-F-instrumente vir kwaliteitsmeting in situ en die modelle om interne (bv. TOVS) en eksterne (verbruining) kwaliteitseienskappe te voorspel. | af_ZA |
| dc.description.version | Doctoral | en_ZA |
| dc.format.extent | xviii, 142 pages : illustrations (some color) | en_ZA |
| dc.identifier.uri | http://hdl.handle.net/10019.1/110489 | |
| dc.language.iso | en_ZA | en_ZA |
| dc.publisher | Stellenbosch : Stellenbosch University | en_ZA |
| dc.rights.holder | Stellenbosch University | en_ZA |
| dc.subject | Table grapes -- Breeding | en_ZA |
| dc.subject | Table grapes -- Quality | en_ZA |
| dc.subject | Fourier Transform Near-Infrared (FT-NIR) spectroscopy | en_ZA |
| dc.subject | Table grapes -- Economic aspects -- South Africa | en_ZA |
| dc.subject | Evaluation research (Social action programs) -- South Africa | en_ZA |
| dc.subject | Harvest maturity | en_ZA |
| dc.subject | UCTD | en_ZA |
| dc.title | Non-destructive evaluation of external and internal table grape quality | en_ZA |
| dc.type | Thesis | en_ZA |