Full Text Available
Note: Clicking the button above will open the full text document at the original institutional repository in a new window.
Approaching authenticity issues in processed meat products by using hyperspectral imaging and chemometrics
Thesis (PhDFoodSc)--Stellenbosch University, 2023.
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Language: | English |
| Published: |
Stellenbosch : Stellenbosch University
2023
|
| Subjects: | |
| Tags: |
No Tags, Be the first to tag this record!
|
| _version_ | 1867613887440879616 |
|---|---|
| access_status_str | Open Access |
| author | Edwards, Kiah |
| author2 | Williams, Paul James |
| author_browse | Edwards, Kiah Williams, Paul James |
| author_facet | Williams, Paul James Edwards, Kiah |
| author_sort | Edwards, Kiah |
| collection | Thesis |
| dc_rights_str_mv | Stellenbosch University |
| description | Thesis (PhDFoodSc)--Stellenbosch University, 2023. |
| format | Thesis |
| id | oai:scholar.sun.ac.za:10019.1/128476 |
| institution | Stellenbosch University (South Africa) |
| language | English |
| last_indexed | 2026-06-10T12:43:16.997Z |
| license_str | Other — see source repository |
| provenance_str_mv | Harvested via OAI-PMH from SUNScholar — Stellenbosch University Repository |
| publishDate | 2023 |
| publishDateRange | 2023 |
| publishDateSort | 2023 |
| publisher | Stellenbosch : Stellenbosch University |
| publisherStr | Stellenbosch : Stellenbosch University |
| record_format | dspace |
| source_str | SUNScholar — Stellenbosch University Repository |
| spelling | oai:scholar.sun.ac.za:10019.1/128476 Approaching authenticity issues in processed meat products by using hyperspectral imaging and chemometrics Edwards, Kiah Williams, Paul James Manley, Marena Hoffman, Louwrens Christiaan Stellenbosch University. Faculty of AgriSciences. Dept. of Food Science. Meat inspection -- Law and legislation -- South Africa Food adulteration and inspection Hyperspectral imaging Chemometrics Meat -- Composition -- Analysis Processed foods -- Technological innovations Hamburgers -- Safety measures UCTD Thesis (PhDFoodSc)--Stellenbosch University, 2023. ENGLISH ABSTRACT: The aim of this dissertation was to present the South African and global meat sector with a rapid, accurate and automated technique for the classification and authentication of processed raw beef patties. This was achieved using near-infrared hyperspectral imaging (NIR-HSI) combined with image analysis, chemometric techniques, machine learning algorithms, optimal waveband selection and hierarchical/decision pathway modelling. Beef patties are assigned and labelled based on the composition of four main categories stipulated in the South African raw processed meat products regulations, namely premium ‘ground patty’; regular ‘burger patty’; ‘value-burger/patty’ and the ‘econo-burger’/’budget’. It was possible to distinguish between the four patty categories (800 patties; 200 patties per category) using NIR-HSI (HySpex SWIR-384; 952-2517 nm) and support vector machines classification (SVM-C; linear) models with accuracies ≥98.5%. The capability to detect and quantify adulteration in the fat content of raw beef patties was also investigated. The SVM-C (linear) models classified the fat content classes [extra lean (≤5%), lean (>5 to ≤10%), regular (>10 to ≤30%), out (>30%)] almost perfectly (98-100%), while support vector machines regression (SVM-R) achieved the best quantification results with a root mean square error of prediction (RMSEP) of 3.6%. This aforementioned technique was also investigated as an alternative authentication method for detecting and quantifying various adulterants [pork, lamb, ostrich, textured vegetable protein (TVP) and mechanically recovered meat (MRM)] in raw beef patties. It was possible to distinguish between the authentic- and adulterated patties with accuracies ≥98%. Once the species adulterants (lamb, pork and ostrich) were detected, with accuracies ≥97%, the adulteration level was predicted using the SVM-R models with RMSEP values of 5.14%, 3.71% and 3.37%, respectively for pork, lamb and ostrich substitution. The results also indicated that the SVM-C models could predict the different total meat (TM) content (98%) and MRM% (97%) classes of the patties. Despite the outstanding results, it was clear that classifying authentic- and adulterated classes was more successful than predicting the adulterant class or -concentration in processed beef burger patties. Following the encouraging success of raw beef patty analysis using the full NIR spectrum, waveband reduction and optimisation was conducted for the potential development of an affordable, rapid and accurate multispectral imaging (MSI) system for the classification and authentication of beef burger patties. Three waveband selection algorithms were explored, namely the use of PC loadings (PC-LD) (28 wavebands), waveband optimisation using the recursive feature elimination (RFE) algorithm (31 or 33 wavebands), and waveband selection based on variable importance in projection (VIP) scores (25 wavebands). A series of linear discriminant analysis (LDA), support vector machines classification (SVM-C) (rbf, linear, poly) and partial least squares discriminant analysis (PLS-DA) models, that consisted of the main levels (patty category classification) and sub-levels (fat content class and authenticity classification), were calculated for the full spectrum and the various reduced waveband sets. The results of all the waveband sets indicated that the 31 wavebands selected based on the SVM-C-RFE (952, 958, 969, 1056, 1110, 1121, 1187, 1192, 1214, 1296, 1312, 1345, 1350, 1863, 1873, 1884, 1890, 2151, 2168, 2173, 2277, 2348, 2408, 2413, 2424, 2435, 2446, 2457, 2468, 2511 and 2517 nm) approach would be favourable and allowed for the removal of 89.2% of the wavebands with only a 3% decrease in the overall average classification accuracy (97% to 94%). The decreased accuracies were observed in three of the four sub-categories. Overall, SVM-C-RFE allowed for the individual classification of 14 classes and shows promise for extending the implementation of HSI and/or MSI for the on-line authentication of processed meat products at an industrial scale. Hierarchical decision pathway modelling was also investigated as an alternative to separate individual model inspections for the simultaneous classification of multi-class and multi-level problems. NIR spectral imaging was used to simultaneously classify 14 classes (1600 patties; 400 patties per category) of beef burger patties using five SVM-C models constructed in a hierarchical decision pathway. Three separate hierarchical pathway models were investigated, one for the full spectrum (288 wavebands) and two based on the reduced waveband sets (31 and 39 wavebands). Overall, the 39 wavebands selected based on SVM-C-RFE were recommended for the classification and authentication of beef burger patties. The average classification accuracy of the hierarchical model for the 14 classes decreased from 97% (full spectrum) to 93% (SVM-C-RFE). This hierarchical pathway resulted in an overall accuracy of 97.4% for the main categories (4 classes) and 95.3% for the sub-categories (10 classes). A large number of the sub-category errors were due to the misclassification of fat content classes, as well as the difficulty distinguishing between adulterated and authentic classes of specific patty categories. Therefore, the classification of fat content (93.0%) and authenticity (93.5%) should ideally be improved before the method is implemented for industry applications. The results of this study demonstrated that NIR hyper- and multispectral imaging are promising analytical techniques for automated raw beef patty analyses. This also shows promise for extending the implementation of NIR spectral imaging for the on-line classification and authentication of processed meat products at an industrial scale. Furthermore, this study has the potential of providing an alternative technique to the current manual, destructive and time-consuming methods, thus contributing to the authenticity and fair-trade of processed meat products locally and internationally. AFRIKAANSE OPSOMMING: Die doel van hierdie verhandeling was om die Suid-Afrikaanse en globale vleissektor 'n vinnige, akkurate en outomatiese tegniek vir die klassifikasie en verifikasie van verwerkte rou beesvleis-patties aan te bied. Dit is bereik met behulp van naby-infrarooi hiperspektrale beelding (NIR-HSI) gekombineer met beeldanalise, chemometriese tegnieke, masjienleeralgoritmes, optimale golfbandkeuse en hiërargiese/besluitnemingspadmodellering. Beesvleis-patties word toegeken en geëtiketteer op grond van die samestelling van vier hoofkategorieë soos gestipuleer in die Suid-Afrikaanse rou verwerkte vleisprodukteregulasies, naamlik premium ‘ground patty’; regular ‘burger patty’; ‘value-burger/patty’ and the ‘econo-burger’/’budget’. Dit was moontlik om te onderskei tussen die vier pattie-kategorieë (800 patties; 200 patties per kategorie) met behulp van NIR-HSI (HySpex SWIR-384; 952-2517 nm) en ondersteuning-vektormasjiene klassifikasie (SVM-C; lineêre) modelle met akkuraatheid ≥98.5%. Die vermoë om vervalsing in die vetinhoud van rou beesvleis-patties op te spoor en te kwantifiseer, is ook ondersoek. Die SVM-C (lineêre) modelle het die vetinhoudklasse geklassifiseer [ekstra maer (≤5%), maer (>5 tot ≤10%), gereeld (>10 tot ≤30%), uit (>30%)] byna perfek (98-100%), terwyl ondersteuning-vektormasjiene regressie (SVM-R) die beste kwantifiseringsresultate behaal het met 'n wortelgemiddelde kwadraatfout van voorspelling (RMSEP) van 3.6%. Bogenoemde tegniek is ook ondersoek as 'n alternatiewe verifikasiemetode vir die opsporing en kwantifisering van verskillende egbrekers [vark, lam, volstruis, tekstuurgroenteproteïen (TVP) en meganies herwinde vleis (MRM)] in rou beesvleis-patties. Dit was moontlik om te onderskei tussen die outentieke en vervalste patties met akkuraatheid ≥98%. Sodra die spesie egbrekers (lam, vark en volstruis) opgespoor is, met akkuraatheid ≥97%, is die vervalsingsvlak voorspel met behulp van die SVM-R-modelle met RMSEP-waardes van onderskeidelik 5.14%, 3.71% en 3.37% vir vark-, lam- en volstruisvervanging. Die resultate het ook aangedui dat die SVM-C modelle die verskillende totale vleis (TM) inhoud (98%) en MRM% (97%) klasse van die patties kan voorspel. Ten spyte van die uitstekende resultate, was dit duidelik dat die klassifisering van outentieke en vervalste klasse meer suksesvol was as om die owerspelige klas of -konsentrasie in verwerkte beesburgerpatties te voorspel. Na aanleiding van die bemoedigende sukses van rou beesvleis-pattie analise met behulp van die volle NIR-spektrum, is golfbandvermindering en optimalisering uitgevoer vir die potensiële ontwikkeling van 'n bekostigbare, vinnige en akkurate multispektrale beeldstelsel (MSI) vir die klassifikasie en verifikasie van beesvleis-burgerpatties. Drie algoritmes vir golfbandseleksie is ondersoek, naamlik die gebruik van PC-ladings (PC-LD) (28 golfbande), golfbandoptimalisering met behulp van die rekursiewe funksie-eliminasie (RFE) algoritme (31 of 33 golfbande), en golfbandkeuse gebaseer op veranderlike belangrikheid in projeksie (VIP) tellings (25 golfbande). 'n Reeks lineêre diskriminante analise (LDA), ondersteuning vektormasjiene klassifikasie (SVM-C) (rbf, lineêr, poli) en gedeeltelike minste vierkante diskriminante analise (PLS-DA) modelle, wat bestaan uit die hoofvlakke (pattie kategorie klassifikasie) en subvlakke (vetinhoudsklas en egtheidsklassifikasie), is bereken vir die volle spektrum en die verskillende verminderde golfbandstelle. Die resultate van al die golfbandstelle het aangedui dat die 31 golfbande wat gekies is op grond van die SVM-C-RFE (952, 958, 969, 1056, 1110, 1121, 1187, 1192, 1214, 1296, 1312, 1345, 1350, 1863, 1873, 1884, 1890, 2151, 2168, 2173, 2277, 2348, 2408, 2413, 2424, 2435, 2446, 2457, 2468, 2511 en 2517 nm) benadering gunstig sou wees en toegelaat word vir die verwydering van 89.2% van die golfbande met slegs 'n afname van 3% in die algehele gemiddelde klassifikasie akkuraatheid ( 97% tot 94%). Die verminderde akkuraatheid is in drie van die vier subkategorieë waargeneem. Oor die algemeen het SVM-C-RFE die individuele klassifikasie van 14 klasse moontlik gemaak en toon belofte vir die uitbreiding van die implementering van HSI en / of MSI vir die aanlyn-verifikasie van verwerkte vleisprodukte op industriële skaal. Hiërargiese besluitnemingspadmodellering is ook ondersoek as 'n alternatief vir afsonderlike individuele modelinspeksies vir die gelyktydige klassifikasie van multiklas- en multivlakprobleme. NIR-spektrale beelding is gebruik om gelyktydig 14 klasse (1600 patties; 400 patties per kategorie) beesvleis-burgerpatties te klassifiseer met behulp van vyf SVM-C-modelle wat in 'n hiërargiese besluitnemingspad gebou is. Drie afsonderlike hiërargiese padmodelle is ondersoek, een vir die volle spektrum (288 golfbande) en twee gebaseer op die verminderde golfbandstelle (31 en 39 golfbande). Oor die algemeen is die 39 golfbande wat op grond van SVM-C-RFE gekies is, aanbeveel vir die klassifikasie en verifikasie van beesvleis-burgerpatties. Die gemiddelde klassifikasie-akkuraatheid van die hiërargiese model vir die 14 klasse het van 97% (volle spektrum) tot 93% (SVM-C-RFE) afgeneem. Hierdie hiërargiese pad het gelei tot 'n algehele akkuraatheid van 97.4% vir die hoofkategorieë (4 klasse) en 95.3% vir die subkategorieë (10 klasse). 'n Groot aantal van die subkategoriefoute was te wyte aan die verkeerde klassifikasie van vetinhoudklasse, sowel as die moeilikheid om te onderskei tussen vervalste en outentieke klasse van spesifieke pattie-kategorieë. Daarom moet die klassifikasie van vetinhoud (93.0%) en egtheid (93.5%) ideaal verbeter word voordat die metode vir bedryfstoepassings geïmplementeer word. Die resultate van hierdie studie het getoon dat NIR-hiper- en multispektrale beelding belowende analitiese tegnieke is vir outomatiese rou beesvleis-pattie ontledings. Dit toon ook belofte vir die uitbreiding van die implementering van NIR-spektrale beelding vir die aanlyn klassifikasie en verifikasie van verwerkte vleisprodukte op industriële skaal. Verder het hierdie studie die potensiaal om 'n alternatiewe tegniek vir die huidige handmatige, vernietigende en tydrowende metodes te verskaf en sodoende by te dra tot die egtheid en billike handel van verwerkte vleisprodukte plaaslik en internasionaal. Doctoral 2023-03-06T15:34:48Z 2023-08-30T13:11:53Z 2023-03 2023-03-06T15:34:48Z 2023-08-31T09:18:49Z 2023-03-06T15:34:48Z 2023-08-31T09:18:49Z 2023-03 Thesis https://scholar.sun.ac.za/handle/10019.1/128476 en Stellenbosch University application/pdf xviii, 179, 21 unnumbered pages : illustrations application/pdf Stellenbosch : Stellenbosch University |
| spellingShingle | Meat inspection -- Law and legislation -- South Africa Food adulteration and inspection Hyperspectral imaging Chemometrics Meat -- Composition -- Analysis Processed foods -- Technological innovations Hamburgers -- Safety measures UCTD Edwards, Kiah Approaching authenticity issues in processed meat products by using hyperspectral imaging and chemometrics |
| title | Approaching authenticity issues in processed meat products by using hyperspectral imaging and chemometrics |
| title_full | Approaching authenticity issues in processed meat products by using hyperspectral imaging and chemometrics |
| title_fullStr | Approaching authenticity issues in processed meat products by using hyperspectral imaging and chemometrics |
| title_full_unstemmed | Approaching authenticity issues in processed meat products by using hyperspectral imaging and chemometrics |
| title_short | Approaching authenticity issues in processed meat products by using hyperspectral imaging and chemometrics |
| title_sort | approaching authenticity issues in processed meat products by using hyperspectral imaging and chemometrics |
| topic | Meat inspection -- Law and legislation -- South Africa Food adulteration and inspection Hyperspectral imaging Chemometrics Meat -- Composition -- Analysis Processed foods -- Technological innovations Hamburgers -- Safety measures UCTD |
| url | https://scholar.sun.ac.za/handle/10019.1/128476 |
| work_keys_str_mv | AT edwardskiah approachingauthenticityissuesinprocessedmeatproductsbyusinghyperspectralimagingandchemometrics |