The effect of muscle type and ageing on Near Infrared (NIR) Spectroscopy classification of game meat species using a portable instrument

Dumalisile, Pholisa (2021-04)

Thesis (PhD)--Stellenbosch University, 2021.

Thesis

ENGLISH ABSTRACT: Meat and meat products represent a large proportion of the human diet as it is known to provide valuable proteins, and is a good source of minerals, particularly iron, and zinc. Because of its nutritional characteristics it tends to be a commodity of demand to consumers. Game meat offers even higher nutritional attributes than any other red meat category because of its low fat and high protein levels making game meat a highly priced product thereby causing it to be an appealing target for species substitution. Also, fraudsters prefer to use products that are easy to adulterate and difficult to detect. To mitigate the fraudulent substitution of meat products, food authentication and labelling is promoted. The conventional methods of authentication such as DNA based techniques are expensive and slow for the rapidly expanding meat trade. Near infrared (NIR) spectroscopy, a rapid non-destructive, environmentally friendly instrument is thought to be an alternative and cheap solution for on-site meat authentication purposes, although this technology has not yet been evaluated for its suitability to distinguish different South African game species and/or muscles. To evaluate the ability of NIR spectroscopy to distinguish between selected game species’ (impala (Aepyceros melampus), blesbok (Damaliscus pygargus phillipsi), springbok (Antidorcas marsupialis), eland (Taurotragus oryx), black wildebeest (Connochaetes gnou) and zebra (Equus quagga)) Longissimus thoracis et lumborum (LTL) muscle steaks, a handheld MicroNIR™ OnSite spectrophotometer was used in a spectral range of 908–1700 nm. After the spectral data was pre-treated with smoothing, SNV-Detrend, the PCA scores plot revealed two clear clusters separating the medium-sized antelopes and large-sized species. The waveband responsible for the separation as indicated by the loadings line plot situated at 1372 nm, was associated with fat. The developed classification models revealed that the steaks could be distinguished with linear discriminant analysis (LDA), soft independent modelling by class analogy (SIMCA) and partial least squares discriminant analysis (PLS-DA) at classification accuracies ranging from 68 - 100%, 67 - 100% and 70 - 96%, respectively. Also, NIR spectroscopy in combination with multivariate data analysis techniques was used to discriminate between different muscle steaks from longissimus thoracis et lumborum (LTL), infraspinatus (IS) and supraspinatus (SS), biceps femoris (BF), semitendinosus (ST) and semimembranosus (SM) of impala and eland species; and samples from fan fillet (FF), big drum (BD), triangle steak (TS), moon steak (MS) and rump steak (RS) of ostriches. Classification accuracies developed with PLS-DA models ranged from 85 to 100% throughout. It is interesting that good classifications accuracies were achieved when the muscles were grouped according to their anatomical locations, irrespective of the muscle used, PLS-DA models yielded accuracies of 97%, 81% and 92% for eland, impala and ostrich, respectively. Even though NIR spectroscopy in combination with multivariate data analysis techniques could successfully distinguish the different muscle types within animals, and muscles across different species, the instrument did fall short in discriminating the ageing periods of blesbok, eland, and ostrich muscles. However, it is postulated that there is still room for improvement when the device is coupled with machine learning. In summary, the handheld MicroNIR™ OnSite spectrophotometer demonstrated its capability in discriminating between different species of game meat indicating that the instrument could potentially be used in the authentication of game meat.

AFRIKAANSE OPSOMMING: Vleis en vleisprodukte verteenwoordig ‘n groot deel van die menslike dieet aangesien dit waardevolle proteïene verskaf, en dit is ‘n goeie bron van minerale, veral yster en sink. As gevolg van die voedingseienskappe is dit geneig om ‘n kommoditeit van aanvraag vir verbruikers te wees. Wildsvleis bied selfs hoër voedingseienskappe as enige ander rooivleis kategorie vanweë die lae vet en hoë proteïen vlakke, wat wildsvleis ‘n duur produk maak, wat veroorsaak dat dit ‘n aantreklike teiken is vir spesie-substitusie. Bedrieërs verkies ook om produkte te gebruik wat maklik is om te vervals en wat moeilik opgespoor kan word. Om die bedrieglike substitusie van vleisprodukte te verminder, word voedselverifikasie en etikettering aangemoedig. Die konvensionele verifikasiemetodes soos DNA-gebaseerde tegnieke is duur en stadig vir die vinnig-groeiende vleishandel. Naby infrarooi (NIR) spektroskopie, ‘n vinnige, nie-vernietigende, omgewingsvriendelike instrument, word beskou as ‘n alternatiewe en goedkoop oplossing vir ‘op-perseel’ vleisverifikasie doeleindes, alhoewel hierdie tegnologie nog nie geëvalueer is vir die geskiktheid om tussen verskillende Suid-Afrikaanse wildspesies en/of spiere te onderskei nie. Die vermoë van NIR spektroskopie om te onderskei tussen Longissimus thoracis et lumborum (LTL) spierskywe van spesifieke wildspesies (rooibok (Aepyceros melampus), blesbok (Damaliscus pygargus phillipsi), springbok (Antidorcas marsupialis), eland (Taurotragus oryx), swart wildebees (Connochaetes gnou) en zebra (Equus quagga)) is geëvalueer deur ‘n draagbare MicroNIR™ OnSite spektrofotometer te gebruik in ‘n spektrale reeks van 908-1700 nm. Nadat die spektrale data vooraf behandel is met SNV-Detrend gladmaaking, het die PCA tellingsplot twee duidelike groeperings getoon wat die mediumgrootte wildsbokke en die groter spesies van mekaar skei. Die golfband wat verantwoordelik is vir die skeiding wat aangedui is deur die beladingslynplot by 1372nm, is geassosieer met vet. Die ontwikkelde klassifikasiemodelle het aan die lig gebring dat daar tussen die steaks onderskeid getref kon word met lineêre diskriminante analise (LDA), sagte onafhanklike modellering volgens klasanalogie (SIMCA) en gedeeltelike kleinste kwadrate diskriminante analise (PLS-DA) met klassifikasieakkuraatheid tussen 68-100%, 67-100%, en 70-96%, onderskeidelik. NIR spektroskopie is ook in kombinasie met meerveranderlike dataontledingstegnieke gebruik om te onderskei tussen verskillende spiersnitte van longissimus thoracis et lumborum (LTL), infraspinatus (IS) en supraspinatus (SS), biceps femoris (BF), semitendinosus (ST) en semimembranosus (SM) van rooibok en eland spesies; en monsters van waaierfilet (FF), groot drom (BD), driehoek steak (TS), maan steak (MS) en kruis steak (RS) van volstruise. Klassifikasie-akkuraatheid wat met PLS-DA modelle ontwikkel is, het deurgaans gewissel van 85% tot 100%. Dit is interessant dat goeie klassifikasie akkuraatheid behaal is wanneer die spiere volgens hul anatomiese ligging gegroepeer is; ongeag die spiere wat gebruik is, het PLS-DA modelle akkuraathede van 97%, 81% en 92% behaal vir, onderskeidelik, eland, rooibok en volstruis. Alhoewel NIR spektroskopie in kombinasie met meervoudige data-analise tegnieke suksesvol kon onderskei tussen die verskillende spiertipes binne diere, en spiere oor verskillende spesies, het die instrument te kort geskiet wanneer dit gekom het by onderskeiding van verouderingstydperke van blesbok, eland en volstruis spiere. Daar word egter gepostuleer dat daar steeds ruimte is vir verbetering wanneer die instrument met masjienleer gekombineer word. Ten slotte, die draagbare MicroNIR™ OnSite spektrofotometer het die vermoë gedemonstreer om tussen verskillende wildsvleis spesies te onderskei, wat aandui dat die instrument moontlik gebruik kan word vir die verifikasie van wildsvleis.

Please refer to this item in SUNScholar by using the following persistent URL: http://hdl.handle.net/10019.1/109988
This item appears in the following collections: