Comparison of Hyperspectral Imaging and Spectrometers for Prediction of Cheeses Composition

Depósito de investigación de la Universidad de Sevilla https://idus.us.es/ “This is an Accepted Manuscript of an article published by Elsevier in FOOD RESEARCH INTERNATIONAL on May 2024, available at: https://doi.org/10.1016/j.foodres.2024.114242”. 1 Compari s o n of h y p er sp ectral imagi n g a nd sp ectrometer s for p re d ictio n of c h ee s es 2 com p o s itio n 3 4 M aria L u cimar da S ilva M edeiros a* ; Leila M oreira de Carvalho b ; M arta S uely M ad r uga b ; 5 Franci s c o J . Rodríguez- P ulido c , F ranci s co J . Heredia c ; Dougla s F ernande s Barbin a* 6 a Depa r t m ent of Food Enginee r ing and T echnology, School of Food Enginee r ing, University 7 of Campina s , Campina s , SP, B r a z il. 8 b Depa r t m ent of Food Enginee r ing, T echnology Cente r , Fede r al Unive rs ity of Pa r aiba, João 9 Pe ss oa, PB, B r a z il. 10 c Food C o lou r & Quality L abo r ato r y, Depa r tment of Nut r ition & Food Science, Universidad 11 de Sevilla, Facultad de Fa r macia, Sevilla, Spain. 12 *Co rr e s ponding autho r : [email protected], [email protected] 13 14 Ab s trac t 15 Arti s anal chee s e s are part of the heritage and identity of different countrie s or region s . In this 16 work, we inve s tigated the s pectral variability of a wide range of traditional Brazilian cheeses 17 and compared the performance of different s pectrometer s to di s criminate chee s e type s and 18 predict c o mpo s itional parameter s . Spectra in the vi s ible (vi s ) and near infrared (NIR) region 19 were collected, u s ing imaging ( vi s /NIR-H S I and NIR-H S I) and conventional ( NI R S) 20 s pectrometer s , and it wa s determi n ed the chemical compo s ition of s even type s of cheeses 21 produce d in Brazil. P rincipal com p onent analy s i s ( P CA) s howed that s pectral variability in 22 the vi s /NIR s pectrum i s related to difference s in color (yellowne ss index) and fat co n tent, 23 while in NIR there i s a greater influence of productive s tep s and fat content. P artial least 24 s quare s d i s criminant analy s i s ( P L S DA) model s ba s ed on s pectral information s howed g r eate r 25 accuracy than the model ba s ed on chemical compo s ition to di s criminate type s of traditional 1 26 Brazilian chee s e s . P artial lea s t s quare s ( P L S ) regre ss ion model s ba s ed on vi s /NIR-H S I, N I R S , 27 NIR-HS I data and H S I s pectro s c o pic data fu s ion (vi s /NIR + NIR) demon s trated excellent 28 performance to predict moi s ture content (R P D > 2.5), good ability to predict fat content ( 2.0 < 29 RPD < 2.5) and can be u s ed to di s c r iminate between high and low protein value s (~1.5 < R P D 30 < 2.0). The re s ult s obtained for imaging and conventional equipment are comparable and 31 s ufficiently accurate, s o that both can be adapted to predict the chemical compo s ition of the 32 Brazilian traditional chee s e s u s ed in thi s s tudy according to the need s of the indu s try. 33 34 Keyword s : arti s anal chee s e s ; denomination of origin; non-de s tructive technologie s ; visible - 35 near infrared (vi s /NIR) s pectro s copy; chemometric s ; data fu s ion. 36 37 1. I ntro du ctio n 38 Hi s torical a s pect s , regional characteri s tic s , manufacturing technologie s , microbial dive r sity 39 and dairy s pecie s contributed to the emergence of different chee s e s with u n ique 40 characte r i s tic s that reflect the cultu r al identity of certain population s . In Brazil, there i s a wide 41 type of traditional chee s e s that are part of the heritage and identity of different regions, 42 including the North ( M arajó chee s e), Northea s t (Butter and Coalho), S outh (Colonial and 43 Serrano) , S outhea s t (Araxá, Campo da s Vertente s , Cerrado, Cana s tra and S erro) and Cente r 44 (Caipira ) of the country ( M argalho et al., 2021). 45 Traditional Brazilian chee s e s have gained greater vi s ibility in recent year s , e s pecially through 46 the granting of Geographical Indication s (GI), award s in national and international 47 competition s ( s uch a s the World C h ee s e Award s ) and, above all, through the approval of new 48 regulatio ns . To date, five geographical indication s have been granted in the Indication o f 49 Origin (I P ) modality, which recognize s the traditional area of chee s e production, hi s to r ical 50 connotation and economic relevance ( S erro, Cana s tra, Colônia Witmar s um, M arajó and 2 51 Cerrado); and a GI in the Denomination of Origin (DO) modality, which recognizes the 52 s pecificitie s of the geographic en v ironment, including natural factor s ( s uitable climate and 53 native pa s ture s ) and human factor s , which provide the product with unique characteri s tics ( the 54 terroir, s uch a s S errano chee s e “Campo s de Cima da S erra”) (Bra s il, 1996a). In 2018, th r ough 55 new regulation s , the “Arte” s eal w a s created, which now allow s the retail s ale of artisanal 56 chee s e s made with raw milk throug h out the national territory (Bra s il, 2018). 57 The GI s and the “Arte” s eal connect characteri s tic chee s e s to their place of origin and, 58 through k now-how pa ss ed down t h rough generation s , recognize s their own identity and adds 59 value, w hich increa s e s the inte r e s t of other producer s in regulating their products. 60 Con s equently, it i s nece ss ary to k now the characteri s tic s of arti s anal chee s e s and identi f y 61 marker s that can a ss i s t in the certification of product s with geographical indication and in the 62 identification of fraudulent product s ( S anto s et al., 2017). 63 Different methodologie s ba s ed on chemical compo s ition ( F itztum et al., 2023; S anto s et al., 64 2017) and ga s chromatograph ( M argalho et al., 2021), M id-Infrared S pectro s copy (M I R ) , 65 Rever s e d - P ha s e High P erformance Liquid Chromatography (R P -H P LC) ( S ilva et al, 2023 ) , 66 and ind u ctively coupled pla s ma o ptical emi ss ion s pectrometer (IC P -OE S ) (Andrade et al., 67 2022) were applied to characterize and di s criminate chee s e s according to the type or region o f 68 producti o n. De s pite pre s enting promi s ing re s ult s , it i s nece ss ary to inve s tigate new techniques 69 that allow obtaining reliable, fa s t, and low-co s t re s ult s . 70 Techniq u e s ba s ed on vibrational s pectro s copy in the vi s ible/near infrared range (vi s /N I R ) 71 have con s olidated application s in t h e food indu s try (Ayvaz et al., 2021; S tocco et al., 2019; 72 Wiedemair et al., 2019; M arinoni et al., 2017; M adalozzo et al., 2015; Karoui et al., 2006; ) 73 and ha s been inve s tigated a s a t o ol for authentication. Vi s /NIR s pectro s copy provides a 74 s pectrum of the s ample, often called “fingerprint”, which can be u s ed to extract information 75 related to it s compo s ition. It i s characterized a s being non-de s tructive, avoiding the use o f 3 76 chemical s and allowing rapid characterization and mea s urement of s everal attributes 77 s imultaneou s ly at any s tage of the p roduction chain and directly on the s urface of the cheese. 78 However, the potential of vi s /NIR can be compromi s ed in heterogeneou s matrice s , s uch as 79 ripened chee s e s , a s conventional s pectrometer s provide information from a s mall portion o f 80 the s ample. To overcome thi s difficulty, it i s po ss ible to u s e vi s /NIR hyper s pectral imaging 81 (HSI), which combine s in a s ingle device the advantage s of conventional s pectro s cop y and 82 artificial vi s ion, allowing the s imultaneou s acqui s ition of information related to composition 83 ( s pectral information) and it s di s tribution within the s ample ( s patial information) (Amigo et 84 al., 2019). Con s idering the s hortage s of s tudie s in thi s field, there i s a need for fu r the r 85 inve s tigation comparing the application of hyper s pectral imaging and conventional 86 s pectrometer s to predict the compo s ition of a wider type of chee s e s and without the need f o r 87 s ample preparation. 88 Due to the large amount of information generated by hyper s pectral imaging and conventional 89 s pectro s copy, a data mining s tep (al s o known a s chemometric s ) become s nece ss ary to ext r act 90 only the relevant information from s pectra. S pectral pre-proce ss ing s tep s are performed to 91 remove/minimize the influence of unde s irable phenomena that affect the s pect r al 92 mea s urement, s uch a s light s cattering, particle s ize and morphology effect s and detecto r 93 artifact s . However, there i s no s tan d ard method, and a trial-and-error approach i s required f o r 94 a s pecific application. Additionally, a wavelength s election s tep can be performed to develop 95 s impler model s , reducing data proce ss ing time and allowing indu s trial application s ( P a s quini, 96 2018; Amigo, 2019). F inally, alg o rithm s are applied to inve s tigate s pectral variability and 97 develop model s to predict chemical compound s of intere s t. 98 Thu s , thi s s tudy compared the performance of NIR S and hyper s pectral imaging (vi s /NIR - H SI 99 and NIR - H S I) device s , a s well a s the fu s ion of H S I s pectro s copic data (vi s /NIR + N I R ) to 100 di s criminate and predict the chemical compo s ition of a wide type of Brazilian traditional 4 101 chee s e s . In thi s work, we inve s tigate the relation s hip of chee s e proce ss ing s tep s with s pect r al 102 variability, and the influence of in s trument technology (image and point mea s urement ) , 103 s pectral r egion (vi s and NIR) and s pectral preproce ss ing method s on the performa n ce o f 104 cla ss ification and predictive model s . 105 106 2. Material a nd met h o ds 107 2.1. Brazilia n tra d itio n al c h ee s e sa m p le s 108 Seventy-two s ample s of s even type s of Brazilian traditional chee s e s were obtained fr om 109 producers, s upermarket s and mar k et s located in three region s of Brazil were analyzed: 110 Northea s t, S outhea s t and S outh. The s ample s et (Table 1) included the main types o f 111 traditional Brazilian chee s e s and con s idered different production technology, curd heat 112 treatment (uncooked and cooked), curd melting, maturation (ripened or not ripened), chemical 113 compo s ition and s tructural characteri s tic s of the chee s e matrix ( s oft and hard chee s es ) to 114 addre ss a wider s ample variation and allow effective evaluation of the ability of vis/N I R 115 device s to predict chemical compo s ition. 116 117 Table 1. De s cription of the Brazilian traditional chee s e s included in the s ampling. Region Northeast South/ Southeast Cheese type (Identification) Coalho integral (CO) and light (CL) Butter cheese integral (BC) and light (BL) Minas Frescal (MF) Minas Araxá (AR) Minas Canastra (CT) Production aspects Production technology Enzymatic coagulation, cutting, stirring, whey drainage, cooking (45–55 °C), and pressing Spontaneous coagulation, desorption, washing with water and/or milk and melting (at 85°C for at least 15 min) with butteroil Enzymatic coagulation, supplemented or not with the action of lactic acid bacteria, whey drainage and packaging Coagulation with endogenous culture and industrial rennet, cutting, agitation, molding, and superficial salting. The cheese is turned daily during shelf maturation Coagulation with natural dairy culture and industrial rennet, cutting, stirring, molding, pressing, and salting Distinctive steps Not ripened Melted Not ripened Ripened (14 days) Ripened (14 days) Regulatory patterns Moisture: 36.0 –54.9%; FTS: 35.0 a 60.0% 1 Moisture: ≤54.9 %; FTS: 25.0 – 55.0% 1 Moisture: >55.0%; FTS: 25.0 – 44.9% 2 Moisture: ≤ 45.9% 3 Moisture: ≤ 45.9% 3 5 Minas Padrão Ripened (light) (MP) (20 days) Ripened (10 days) Enzymatic coagulation, complemented by Moisture: 36 a the action of lactic acid bacteria, heating 45.9%. FTS: (32–42°C), draining, pressing, and salting 42 a 57% 4 Coagulation with industrial rennet, heating Moisture: 36.0 Colonial (CN) (30–45°C), whey drainage, molding, a 45.9%; FTS pressing, and salting 45.0 a 59.9% 5 118 1 Brasil (2001), 2 Brasil (1997), 3 Minas Gerais (2008), 4 Brasil (2020), 5 Rio Grande do Sul (2023). 119 120 2.2. I ma g e a nd sp ectral d ata collectio n 121 2.2.1. Sam p le treatme n t a nd morp h ological feat u re s 122 To inve s tigate the s pectral changes a ss ociated with the chee s e proce ss , e s pecially ripening, 123 the cheese s were s ectioned horizontally (20 mm in height) and two cylinder of 25 mm in 124 diameter were removed from the center (internal s urface) toward s the rind (external s ur f ace ) , 125 with the aid of a s tainle ss - s teel s am p ler. 126 127 Fig. 1. Repre s entation of obtaining chee s e s ample s for s pectral acqui s ition (a) and the ty p es o f 128 traditional Brazilian chee s e s (b) included in the s ampling. 129 6 130 2.2.2. Porta b le N I R sp ectrometer (N I R S ) 131 Spectra were recorded u s ing a portable NIR s pectrophotometer ( M icroNIR P ro Lite 1700, 132 VIAVI, S anta Ro s a, California, USA), previou s ly calibrated, s canning the wavelength r ange 133 of 908 and 1676 nm ( s pectral re s olution of 6.2 nm). Reflectance s pectra were corrected by 134 mean s of a two-point calibration. F or that, a white reference s pectrum ( S pectralon, Lab s phe r e 135 Inc., North S utton, U S A) and a dark current s pectrum were acquired in each collection 136 s e ss ion. F inally, two s pectra were acquired at random location s on the outer s urface (totaling 137 144 s pectra) and two s pectra on the inner s urface (totaling 144 s pectra) of each c h eese 138 cylinder. 139 140 2.2.3. Hy p er sp ectral imagi n g (H S I) 141 Hyper s pectral image s of Brazilian traditional chee s e s were acquired in the vi s ible and nea r 142 infrared spectral region (vi s /NIR – H S I) u s ing a S pecim IQ camera ( S pectral Imaging Ltd., 143 Oulu, F inland) with a s pectral range of 397 to 1004 nm (512 × 512 pixel s and F WH M o f 7 144 nm). Thi s camera i s ba s ed on the pu s hbroom principle and ha s a mobile, portable, and 145 autonomou s de s ign (integrated operating s y s tem and control s ), which allow s the acqui s ition 146 of image s in different environment s . Thi s camera ha s a certified reflectance device made o f 147 teflon to perform the calibration s . Al s o, near-infrared (NIR – H S I) hyper s pectral image s we r e 148 acquired u s ing a laboratory s y s tem compo s ed of a Xenic s ® XEVA-U S B InGaA s camera ( 320 149 × 256 pixel s ; Xenic s Infrared S olution s , Inc., Leuven, Belgium), a s pectrograph ( S pecim 150 Im S pector N17E Enhanced; S pectral Imaging Ltd., Oulu, F inland) covering the s pectral r ange 151 between 884 – 1717 nm (3.25 nm s pectral re s olution), a mirror s canner ( S pectral Imaging 152 Ltd., O u lu, F inland) and a computer s y s tem with in s trumental acqui s ition s oftwa r e 153 SpectralDAQ v. 3.62 ( S pectral Imaging Ltd., Oulu, F inland). A “white reference” image ( W, 154 100% reflectance) wa s acquired f r om a white ceramic tile (Lab s phere Inc., North S utton, 7 155 U S A), a n d a “dark reference” image (B, 0% reflectance) wa s obtained with the light so u r ce 156 off and the camera covered with it s opaque cap. 157 The chee s e cylinder s were placed on a dark s urface and illuminated by two 70 W tungsten 158 iodine halogen lamp s ( P rilux ®, Barcelona, S pain) s eparated 50 cm and oriented at 45° fr om 159 the area of image. The di s tance between the camera and the s ample wa s 45 cm. The came r a 160 cover s an angle of 24º with a s pee d of 5.6 degree s / s ec. F or each chee s e cylinder, an image o f 161 the oute r s urface and one of the inner s urface s wa s recorded. After the calibration and 162 s egmentation proce ss e s , the average s pectra of the region of intere s t (ROI) were extracted 163 u s ing M atlab (R2019; M athwork s , Natick, U S A). 164 165 2.2.4. HS I sp ectro s co p ic d ata s et f us io n 166 Given t h e complementary nature of the information provided by the vi s ible (vi s ) and nea r 167 infrared ( NIR) s pectra, the average vi s /NIR-H S I (397 to 1004 nm) and NIR-H S I (884 to 1717 168 nm) s pectra were s ectioned at the 950 nm wavelength and unified into a new matrix (397 to 169 1717 nm ) to increa s e reach and improve cla ss ification and prediction model s . 170 171 2.3. Che m ical com p o s itio n , colorimetric p arameter s , a nd text u re 172 2.3.1. Chemical com p o s itio n 173 The s ample s were cru s hed until obtaining a homogeneou s s tate and then were s ubjected to the 174 following analy s e s (g/100 g of chee s e): moi s ture, fat, protein s , and a s h, according to the 175 A ss ociation of Official Analytical Chemi s t s protocol s (AOAC, 2012). After, fat was 176 calculated in dry matter (g/100g) for compari s on with the legi s lation and carbohydrate s we r e 177 calculated by s ubtracting the moi s ture, fat, protein, and a s h content s from the total 178 compo s ition (100%). 8 327 the s e chee s e s . In butter chee s e, the milk i s coagulated by the action of acid s and then 328 s ubjected to draining, wa s hing, and melting with butteroil. Acid coagulation cau s e s colloidal 329 calcium pho s phate s olubilization ( M a s otti et al., 2020) and wa s hing can promote the leaching 330 of mineral s into the whey, contributing to the reduction of mineral content. On the other hand, 331 in the production of CO chee s e, as well a s in CL, MP and MF chee s e s , calcium chloride is 332 added a s an ingredient to provide ideal condition s for milk coagulation (improves gel 333 firmne ss and reduce s coagulation time), re s ulting in an increa s e in mineral content (Koutina et 334 al., 2016). M ina s traditional AR a n d CT chee s e s had s tati s tically equal mineral contents ( p < 335 0.05) to chee s e s with added calcium chloride. Although the s e chee s e s do not have addition o f 336 calcium chloride, thi s increa s e may be related to difference s in the percentage of s alt ad d ed in 337 the technological proce ss and the concentration of s olid s during maturation. 338 The compo s ition interval s pre s ented in thi s s tudy are in accordance with tho s e reported by 339 M argalho et al. (2021) for 402 s ample s of 11 type s of Brazilian arti s an chee s e s and are close 340 to tho s e reported by author s who o b tained robu s t model s to predict compo s itional attributes in 341 a wide type of chee s e s (Ayvaz et al., 2021; S tocco et al., 2019). Thu s , it can be inferred that 342 the s ampling u s ed in thi s s tudy an d the model s developed from the s e data are repre s entative 343 of the compo s itional variability of Brazilian traditional chee s e s . 15 344 Table 2. Chemical compo s ition, in s trumental color parameter s and yellowne ss index (YI) of B r azilian t r aditional cheeses. R egio n Northea s t Ch eese sa mp les Coalho - I nteg r al ( CO ) C h e m ical com p ositio n (%) Mois tu re Fat 45.46bc ±2.62 23.34bc±2.17 F D M 42.73ab ±2.42 Protei n 24.73ab ±1.12 C ar b 2.58de± 1.46 As h 3.89ab ±0.66 C olor parameter s ( CI EL A B unit s ) L* C ab ha b 89.1ab ± 1.5 25.3ab ± 2.8 100.2ab ± 1.0 YI 40.6abc ± 4.7 South/ S o uthea s t Coalho - Light ( CL ) Minas Fr escal ( M F) Butte r cheese - I nteg r al ( BC ) Butte r cheese - Light ( BL ) Minas A r axá ( AR ) M inas Canast r a ( CT ) M inas P ad r ão light ( M P) Colonial ( CN ) 44.78bcd ±2.83 46.65b ±2.33 40.71cd±3.84 54.45a ± 0 .15 42.54bcd ±2.27 40.12d ±4.90 46.55b ±1.10 43.87bcd±2.14 21.98c±6.29 22.95bc±0.61 29.25a ±4.57 11.48d±0.12 26.78abc±1.73 28.12ab ±3.47 13.51d±3.37 26.29abc±3.27 39.38b ±9.41 43.12ab ±2.87 49.06a±4.66 25.19c±0.28 46.65a ±3.07 46.86a ±2.83 25.17c ±5.86 45.82a ±3.89 24.23b±1.77 22.74b ±2.43 24.56b±1.24 23.72b±0.24 23.11b±2.21 24.96ab ± 2.20 27.31a ±0.96 24.30b ±1.26 4.82b ± 1.39 4.42bc ± 0.90 3.36bcde±1.01 7.76a±0.32 3.93bcd±0.69 2.84cde ±0 . 94 8.66a ± 1.45 2.15e ±1.51 4.18a ±0.62 3.25bc ±0.21 2.11d±0.31 2.60cd±0.02 3.64ab±0.36 3.95ab ±0.86 3.98ab±0.27 3.89ab ±0.44 90.9a ± 0.6 90.8a ± 4.9 69.3d ± 4.9 67.4d ± 2.6 81.2bc ± 10.3 83.2abc ± 7.9 79.6c ± 1.7 83.8abc ± 1.9 22.5ab ± 2.2 20.4b ± 2.2 25.9ab ± 6.4 21.1ab ± 0.7 27.3a ± 6.9 19.7b ± 2.3 22.8ab ± 3.2 25.6ab ± 4.7 90.8d ± 0.5 98.4bc ± 5.3 105.0a ± 2.5 96.7bc ± 0.7 90.6d ± 8.3 95.9bcd ± 2.9 94.2cd ± 0.9 97.8bc ± 3.8 35.4bc ± 3.4 31.8c ± 4.5 51.6a ± 12.6 44.5abc ± 3.3 49.7ab ± 19.9 33.9c ± 5.7 40.2abc ± 5.6 43.3abc ± 9.2 Range 34.35 – 54.43 9.07 – 36.47 17.50 – 55.88 18.78 – 28.35 0.39 – 10.19 1.64 – 5.45 60.9 – 94.7 17.0 – 38.5 77.2 – 109.5 26.1 – 82.4 345 FDM = Fat in dry matter and Carb = Carbohydrate. Values in the same column followed by different letters are significantly different by ANOVA test (p < 0.05). 346 347 Table 3. Fatty acid s profile of Brazilian t r aditional chee s e s . R egio n C h eese sa mp les C14 : 0 C 16:0 C 16:1 C 18:0 Fatty aci d s profile (%) C 18:1 C 18:2 C 18:3 SF As M U F As P U F As N o rthea s t S outh/ Southea s t Coalho - I nteg r al ( CO ) Coalho - Light ( CL ) M inas Fr escal ( M F) Butter cheese - I nteg r al ( BC ) Butte r cheese - Light ( BL ) M inas A r axá ( AR ) M inas Canast r a ( CT ) M inas P ad r ão light ( M P) Colonial ( CN ) Range 10.66ab±1.10 10.65ab±0.76 11.58ab±1.79 9.77b±1.10 11.02ab±0.09 12.07a±0.85 11.95a±0.57 11.03ab±0.78 10.35ab±0.4 9 8.36 – 13.15 31.19bc±2.35 31.41bc±0.16 32.73abc±2.54 29.22c±1.62 31.85abc±0.01 36.08a±2.39 33.56ab±3.38 32.85abc±1.80 32.91abc±0.85 27.05 – 38.30 1.33b±0.70 1.50b±0.98 2.41ab±1.32 1.71ab±1.12 1.17b±0.04 1.93ab±1.02 2.82ab±1.31 3.60a±1.27 2.94ab±1.09 0.81 – 5.04 13.55ab±1.37 13.33ab±0.84 12.51abc±1.55 14.65a±1.65 13.77ab±0.07 11.69bcd±0.80 9.82d±1.53 10.55cd±0.56 12.02bcd±0.59 8.32 – 16.66 29.24ab±2.8 3 28.61ab±1.9 6 25.76bc±3.8 3 30.67a±2.97 29.33ab±0.0 4 22.60c±2.56 24.78bc±1.76 27.14abc±1.99 26.34abc±0.34 20.29 – 34.58 2.32ab±0.23 2.10ab±0.14 2.36ab±0.21 2.15ab±0.42 2.00ab±0.07 2.17ab±0.18 1.94ab±0.28 1.87b±0.47 2.49a±0.42 1.18 – 2.83 1.21ab±0.31 1.08abc±0.33 0.91abc±0.40 1.35a±0.32 1.15abc±0.20 0.54bc±0.38 0.70abc±0.30 0.79abc±0.56 0.52c±0.19 0.00 – 1.77 65.91b±2.99 66.59b±1.68 69.08ab±5.37 64.06b±2.92 66.35b±0.13 72.77a±2.94 68.86ab±2.37 66.61b±3.06 67.71ab±0.82 59.24 – 76.15 30.57b±2.6 7 30.12b±1.4 9 28.17ab±4. 82 32.38b±2.4 5 30.50b±0.0 0 24.53b±2.7 4 27.60ab±1. 47 30.74b±2.9 4 29.28ab±1. 30 21.29 – 36.59 3.53a±0.44 3.17abc±0.26 3.27abc±0.31 3.50ab±0.62 3.16abc±0.13 2.71bc±0.45 2.64c±0.34 2.65c±0.28 3.01abc±0.52 2.20 – 4.17 16 * 348 In Table 2, P olar coordinate s chroma (Cab) and hue (h ab ) are s hown in s tead of the Cartesian 349 coordinate s a* and b*, a s they explain color better from a p s ychophy s ical point of view. 350 Chroma in chee s e s i s almo s t entirely due to the influence of the b* component. In ad d ition, 351 indicating the hue value (h ab ) alone s how s whether the chee s e ha s a more yellowi s h or mo r e 352 orange appearance without the need to evaluate a* and b* s imultaneou s ly. F inally, the 353 yellowne ss index (YI), which depend s on b* and L*, ha s been s hown a s a common indicato r 354 of chee s e color in s cientific literature. Brazilian arti s anal chee s e s s howed s igni f icant 355 difference s (p < 0.05) for the in s tr u mental color parameter s : lightne ss (60.9 – 94.7), ch r oma 356 (17.0 – 38.5), hue (77.2 – 109.5) and yellowne ss index (26.1 – 82.4). S tandard deviation 357 value s s howed that there i s al s o a large variation in color parameter s for the s ame type o f 358 chee s e, w hich may be related to difference s in the production proce ss and s torage conditions. 359 The yellow color of the chee s e (h ab ~ 90) i s mainly due to the pre s ence of carotenoid s , s uch as 360 β-carote n e (C 40 H 56 ), lutein (C 40 H 56 O 2 ) and β-cryptoxanthin (C 40 H 56 O) (Gentili et al., 2013 ) . I n 361 thi s s tudy, a relation s hip wa s ob s erved between the brightne ss and the productive a s pects o f 362 the chee s e s , where the unripened chee s e s (CO and MF ) were characterized by lighter colo r s 363 (higher v alue s of lightne ss ), the ripened chee s e s by intermediate color s and the melted by 364 darker coloration s (lower lightne ss value s ). Regarding the yellowne ss index, which takes into 365 account the color parameter b* and brightne ss , it wa s ob s erved that BC i s more yellow, w hile 366 MF and AR chee s e s are more whiti s h. 367 Table 3 pre s ent s the re s ult s of the fatty acid profile of traditional Brazilian chee s e s , whe r e 368 s tati s tically s ignificant difference s are ob s erved (p < 0.05). In general, the s ample s consist 369 mainly of s aturated fatty acid s ( S FA s ) (59.24 – 76.15%), with empha s i s on myri s tic acid 370 (C14:0, 8 .36 – 13.15%), palmitic acid (C16:0, 27.05 – 38.30%) and s tearic (C18:0, 8 .32 – 371 16.66%), followed by monoun s aturated fatty acid s ( M U F A s ) (21.29 – 36.59%), e s pecially 372 oleic (C18:1, 20.29 – 34.58%). It i s po ss ible to ob s erve that chee s e s produced in the Northeast 17 373 region (CO, CL, BC, BL and MF) have a different fatty acid profile than chee s e s from the 374 South and S outhea s t region s (AR, CT, MP and CN), a s they have lower percentage s of SF As 375 and higher level s of P U F A s . The fatty acid profile of chee s e s i s directly related t o the 376 compo s ition of milk fat (rearing sys tem, animal feeding and time of year), a s well a s c h eese 377 proce ss i n g, e s pecially microbiota and maturation period ( J e s u s et al., 2023; 2021). Thu s , these 378 compo s itional difference s a ss ociated with the production proce ss and producing region can be 379 inve s tigated to certify the origin of the chee s e s . 380 381 3.2. S p ectral p rofile 382 The mean ab s orbance s pectra of B r azilian traditional chee s e s ( F ig. 2) are comparable to those 383 previou s ly reported for other chee s e type s ( M alegori et al., 2021; Rei s et al., 2022; M e d ei r os 384 et al., 2 0 23). It i s po ss ible to ob s erve different ab s orption band s in the vi s ible (450 nm ) and 385 near infrared (970, 1150, 1210, 12 3 0, 1390, 1450, 1500, 1526 and 1660-1690 nm) regions o f 386 the prep roce ss ed s pectra related to the main organic compound s pre s ent in chee s e. 18 387 388 Fig. 2. vi s /NIR and NIR s pectra of Brazilian traditional chee s e s raw, preproce ss ed with 389 Savitzky - Golay s moothing combined with S NV ( S G + S NV) and with the fir s t derivative o f 390 Savitzky - Golay (1 s t S G). 391 392 The ab s orption peak at 450 nm can be attributed to the pre s ence of carotenoid s re s pon s i b le f o r 393 the yellow color of chee s e s (Britton et al., 2004), s uch a s β-carotene, lutein a n d β - 394 cryptoxa n thin, and/or the pre s ence of riboflavin (vitamin B2) (Becker et al., 2003). The 395 ab s orption band s at 970 and 1450 nm corre s pond to s econd and O–H s tretching in the f i r st 396 overtone, characteri s tic of water. The ab s orption band s at 740, 1150, 1230, 1340-1390 and 397 1690 nm corre s pond to C–H s tretching at the fourth, s econd and fir s t overtone (–CH, –C H 2, – 398 CH3), p r e s ent in CH2 group s in the acid chain fatty acid s and their terminal CH3 grou p s, as 19 399 well a s CH and CH2 pre s ent in the glycerol fraction. The ab s orption band s at 1170 and 1660 400 nm are attributed to C–H vibratio ns in the s econd and fir s t overtone and are a ss ociated with 401 the pre s ence of un s aturated fatty acid s (–HC=CH–) in the aliphatic chain s of fat, e s pecially 402 oleic acid (C18 :1) (O s borne et al., 1993). The peak s at 910 and 1210 nm, attributed to C–H 403 vibration s at the third and fir s t overtone, and at 1500 and 1530 nm, attributed to N–H 404 vibration s at the fir s t overtone, are a ss ociated with protein content, s uch a s ca s ein ( F rank & 405 Birth, 1982; O s borne et al., 1993). 406 407 3.3. Princi p al Com p o n e n t s A n aly s i s (PCA) 408 Principal Component Analy s i s ( P C A ) ( F ig. 3) wa s applied to NIR S and H S I s pectra (vi s /N I R, 409 NIR and data fu s ion) to explore s pectral variation s among Brazilian traditional chee s e s . The 410 re s ult s a r e s hown according to three grouping s : chee s e type, production s tep s (ripening and 411 curd melting) and compo s ition. The influence of curd heating wa s al s o inve s tigated, but it was 412 not ob s e r ved s ample s eparation regarding thi s feature. 413 The two - dimen s ional repre s entation ( P C1 × P C2) of the P CA s core s s howed that the N I R - 414 HSI s pectra explained a higher percentage of chee s e variability (93.64%), followed by NI R S 415 data (92 . 41%), vi s /NIR-H S I (86.78 %) and the fu s ion of H S I s pectro s copic data (77.26 %) . 416 Although there i s a difference in the variability explained by the fir s t two component s (P C1 417 and PC2), the P CA s performed with the different data allow the vi s ualization of the s epa r ation 418 of s ome chee s e type s , e s pecially b u tter chee s e (BC) and M ina s P adrão light chee s e ( MP ). The 419 overlapping of the other chee s e s implie s that there are other factor s , in addition to the type o f 420 chee s e, t h at influence the s pectral variability of the chee s e s and, con s equently, the s eparation 421 of the s ample s . 20 422 423 424 Fig. 3. P CA s core plot of Brazilian traditional chee s e s according to type, production (ripening 425 and curd heating), and compo s itional a s pect s (fat content and yellowne ss index). 426 427 In the PCA performed with the vi s /NIR-H S I s pectral data (397- 1004 nm, 1 s t S G) ( F ig. 3 ) it is 428 po ss ible to ob s erve that the s pectral variability repre s ented by the s core s of fir s t principal 21 429 component wa s influenced to a greater extent by the color of the s ample s , where chee s e s with 430 lower yellowne ss index (YI ≤ 40 ) were characterized by negative s core s on P C1 and the 431 highe s t yellowne ss index (YI > 40 ) by po s itive s core s . Thi s re s ult i s in line with the loadings 432 plot ( F ig. S 2a), where it i s po ss ible to ob s erve that variation s in P C1, a s well a s in P C2, a r e 433 a ss ociated with the content of carotenoid s and/or riboflavin (440 nm) (Britton et al., 2004 ) . 434 The compo s itional a s pect s of the chee s e s , e s pecially the fat content, influenced the s eparation 435 of s ample s in P C3 s core s ( F ig. S 1 in s upplementary material), where chee s e s with lo w e r f at 436 content w ere characterized by negative s core s and higher fat chee s e s by po s itive s cores. The 437 bigge s t contributor s to the s eparation in thi s main component ( F ig. S 2a in s upplementa r y 438 material ) are a ss ociated with the content of carotenoid s and riboflavin (430, 500 and 530 nm ) 439 (Britton et al., 2004; Becker et al., 2 003) and the fat content (940 nm) (O s borne et al, 1993 ) . 440 In PCA model s performed with NIR S (1100 – 1600 nm, S NV + 2nd S G) and NIR-H SI data 441 (1100 – 1600 nm, S NV + 1st S G) ( F ig. 3) the fir s t principal component ( P C1) i s a ss ociated 442 with change s in protein s tructure (e s pecially ca s ein) due to chee s e proce ss ing. In non-ri p ened 443 chee s e s ( CO, CL and MF ), located in the po s itive region of P C1 s core s , the enzymatic action 444 of indu s trial rennet promote s limited proteoly s i s of k-ca s ein (cleavage of the P he 105 - M et 106 445 peptide bond) s eparating it into two macropeptide s (Herbert et al., 1999). In ripened cheeses 446 (AR, CT, MP and CN), di s tributed mainly between the negative and intermediate sco r es 447 region s , in addition to enzymatic coagulation, change s occur during ripening that b r eak 448 chee s e protein s into oligopeptide s , which can additionally be degraded into s horter pe p tides 449 and amino acid s (Boran et al. 2023). In melted chee s e s (BC and BL), s pread in the negative 450 part of P C1 s core s , the main a s pect s of production are acid coagulation promoted by the 451 natural microbiota of milk and the melting s tage. The latter ha s a greater impact on the 452 s tructure of the protein s , s ince melting s alt s (citrate s , polypho s phate s , or s odium bicarbonate ) 453 are u s ed that promote the peptization of ca s ein, s eparating it s large hydrophobic aggregates 22 454 into s maller unit s (Garcia et al., 2023). Thu s , it i s po ss ible to infer that the extent of changes 455 in protein s tructure increa s e s in the negative s en s e of P C1 s core s (not ripened < ripened < 456 melted). Thi s behavior wa s al s o ob s erved by Herbert et al. (1999) when s tudying the 457 influence of milk coagulation type s (acid, enzymatic and mixed) on tryptophan emission 458 fluore s cence s pectral data, where the fir s t principal component s eparated the s amples 459 according to modification s in the micellar s tructure. The s econd principal component 460 de s cribes the variability a ss ociated with the fat content, where it i s po ss ible to s eparate the 461 s ample s with the highe s t (full fat) and lowe s t (low fat) content in the negative and positive 462 part of PC2 s core s , re s pectively. The wavelength s ( F ig. S 2b and S 2c in s upplementa r y 463 material ) that contributed to the s eparation of the s ample s toward s P C1 and P C2 s cores a r e 464 a ss ociated with moi s ture content (1450 nm), fat (1140, 1170, 1210, 1320, 1390 and 1410 nm ) 465 and protein s (1190 and 1510 nm) (O s borne et al., 1993) 466 From an exploratory point of view, the fu s ion of H S I s pectro s copic data (vi s /NIR + NIR) ( 397 467 – 1600 nm, S NV + 1st S G) did not promote s ample s eparation beyond what wa s ob s erved f o r 468 the s epa r ate technique s . In the fir s t principal component ( P C1) s core s it i s po ss ible to obse r ve 469 a trend of s ample s eparation according to the s tructural modification s of the protei n s, as 470 ob s erved in the P CA performed with the NIR data, while the influence of the yello w ness 471 index i s al s o ob s erved, like the observed in the P CA for the vi s /NIR data. In the loadings o f 472 PCA applied to the data fu s ed ( F ig. S 3d in s upplementary material) it i s po ss ible to obse r ve 473 that the w avelength s that influenced the s eparation in P C1 and P C2 s core s are the s ame ones 474 reported in vi s /NIR-H S I and NIR-H S I s eparately, which are a ss ociated with the carotenoid 475 content ( 490 nm), water (970 nm), fat (1140, 1230, 1330 and 1410 nm) and proteins 476 (1500nm) (O s borne et al., 1993). 477 478 3.4. Clas s ificatio n mo d el s 23 2 2 556 re s ult i s expected, becau s e in addition to carotenoid s contributing to the yellowi s h colo r o f 557 chee s e s , they are s olubilized in the fat fraction. S imilarly, S tocco et al. (2019) ob s erved that 558 model s built with the vi s /NIR region (350 – 1830 nm) pre s ented s imilar performances to 559 tho s e based on the NIR region ( 1 100 – 1830 nm) (R P of 0.85 and R P D of 2.03), when 560 comparing the performance of two s pectrometer s and s pectrum interval s to predict t h e f at 561 content in chee s e s . All model s p r edicting fat content s howed R P D between 2.0 and 2.5, 562 indicating that they can be u s ed for s creening purpo s e s . 563 The fu s ion of the H S I s pectro s c o pic data s et (vi s /NIR + NIR) s howed the be s t re s ult f o r 564 predictin g protein content, with calibration and prediction error s clo s e to 0.9%. The 565 wavelength s that contributed to t h e performance of the model are a ss ociated with p r otein 566 content (1510 nm) and s tructure s a ss ociated with carotenoid s (508 nm) and fat (1145, 1225, 567 1395 and 1660 nm). It i s po ss ible to predict the protein content without lo ss of performance 568 (R P of 0.78 and error of 0.93%) u s i n g the NIR-H S I s pectra and to obtain approximate results 569 u s ing the other equipment (betwee n 0.57 and 0.73 and error s of ~1.0%). However, acco r ding 570 to the RPD value s , only model s ba s ed on the fu s ion of s pectro s copic data and NIR-H SI a r e 571 indicated for e s timating the protein content, a s they are able to di s tingui s h between hig h and 572 low level s of thi s compound (1.5 < R P D < 2.0). 573 Few s tudie s available in the s cientific literature have inve s tigated the u s e of dif f e r ent 574 s pectrometer s and s pectral range s in the vi s and NIR range s to predict chemical characte r istics 575 of a wide type of chee s e s . Thi s number i s even lower when looking for s tudie s that developed 576 model s from s pectra acquired wit h whole chee s e (without any s ampling, grinding or othe r 577 preparation). The di s cu ss ion, there f ore, focu s e s mainly on s tudie s that u s ed thi s la s t crite r ion 578 and wa s carried out in term s of R MS E P and R P D, s ince the coefficient of determination ( R 2 ) 579 i s influenced by the range of reference value s . 28 580 Wiedemair et al (2019) compared the performance of model s ba s ed on vi s /NIR (740 – 1070 581 nm) and NIR (800 – 2500 nm) sp ectra to predict the compo s ition of hard and s emi - ha r d 582 chee s e s ( n = 46). The s e author s r eported re s ult s in accordance to thi s current wor k and 583 ob s erved better performance for moi s ture prediction u s ing the NIR s pectrum (R MS E P o f 1.10 584 and R PD of 5.60). On the other hand, fat content wa s predicted with lower error (R MS E P o f 585 1.19 and R P D of 7.75) from vi s /N I R s pectra. S tocco et al. (2019) developed P L S models to 586 predict compo s itional characteri s tic s of a wide type of chee s e s (197 s ample s from 37 587 categorie s ) u s ing a portable vi s /NIR s pectrometer (350 - 1830 nm) and obtained performance 588 (R M SE P ) very clo s e to tho s e reported in thi s s tudy to predict fat ( 2.03%, 350 - 1830 nm ) and 589 protein s (1.57%, 850 - 1050 nm), and s lightly lower for moi s ture prediction (2.00%, 1100 - 590 1830 nm ) . Ayvaz et al (2021) when predicting the compo s ition of Ezine chee s e from diffe r ent 591 s pecie s ( n = 81) u s ing an F TIR s pectrophotometer (10000 – 4000 cm -1 ) obtained s u p e r io r 592 re s ult s f o r moi s ture (R P D of 3.38) and s imilar re s ult s for fat (R P D of 2.14) and protein s ( R P D 593 of 1.47). 594 The re s ult s obtained in thi s s tudy w ere lower than tho s e reported for predicting fat content in 595 Emmental chee s e (n = 91, R MS EP of 0.39% and R P D of 3.82) (Karoui et al., 2006), G r ana 596 Padano (n = 190, R MS E P of 1.19%) ( M arinoni et al., 2017) and ricotta (n = 19, R MS ECV o f 597 1.9%) (Madalozzo et al., 2015). The s maller error s obtained in the s e s tudie s are ju s tified by 598 the high homogeneity of the s ample s et, s ince only one category of chee s e wa s include d . The 599 re s ult s are al s o lower than tho s e obtained for model s ba s ed on s pectra acquired with 600 cru s hed/grated chee s e. Thi s re s ult i s expected, a s predicting chee s e compo s ition from the 601 outer s u r face s pectrum i s more challenging. Although grinding produce s better performances 602 (Ayvaz et al., 2021; Wiedemair et al., 2019; M arinoni et al., 2017), it confront s the non - 603 de s tructive character of vi s /NIR s pectro s copy and limit s it s application in the r apid 604 authentication of chee s e s with of o r igin or protected geographical indication. 29 605 606 3.6. Che m ical ma ps of moi s t u re, fat, a nd p rotei n 607 The be s t model s obtained for the vi s /NIR-H S I and NIR-H S I data were applied to each pixel 608 from ROI of the image s to vi s ualize the di s tribution of moi s ture, fat, and protein in the cheese 609 s ample s , obtaining a p s eudo-color image, or chemical map. F ig. 5 s how s s ome map s o f the 610 s patial di s tribution of the predicted chemical propertie s in different type s of chee s e. The 611 value s pre s ented in the lower corner of each chee s e cylinder corre s pond to the ab s olute e rr o r 612 (%), calculated by s ubtracting the reference value and that predicted by the model. 30 613 One of the greate s t advantage s of u s ing hyper s pectral imaging, compared to RGB imaging 614 and conventional s pectro s copy, i s that it i s po ss ible to ob s erve the concentration di s tri b ution 615 of the m o deled analyte. In the RGB image ( F ig. 5) one can ob s erve difference s in the colo r 616 and texture of the external s urface of the s ample s due to the pre s ence of different types o f 617 chee s e, while the compo s itional difference s are practically imperceptible. On the other hand, 618 hyper s pectral imaging allow s you t o clearly vi s ualize difference s in compo s ition by the colo r 619 inten s ity of each pixel in the chemical map. 620 Comparing the s patial di s tribution map s vi s /NIR-H S I and NIR-H S I it i s po ss ible to ob s e r ve a 621 difference in the homogeneity an d color inten s ity of the pixel s for the s ame s ample. This 622 ob s ervation may be related to the difference in re s olution between the image s and the 623 heterogeneou s di s tribution of analyte s in the s ample. De s pite thi s , there are no majo r 624 difference s between the re s ult s obtained by the reference method and tho s e predicted b y the 625 two device s . 31 626 627 628 Fig. 5. RGB image s and s patial distribution map s of moi s ture, fat and protein content based on vi s /NIR-H S I and NIR-H S I data. 32 629 630 It i s important to mention that the vi s /NIR-H S I and NIR-H S I model s (ba s ed on the sp ect r a 631 acquired on the external s urface) pre s ented a low average error to predict the compo s ition o f 632 the s am p le s from the image s of the internal part of the chee s e, with average ab s olute erro r s o f 633 0.32 and 0.00% for moi s ture, 1.01 and 0.31% for fat, and 0.33 and 0.06% for protein, 634 re s pectively. The s e re s ult s indicate that, even though s ome chee s e s have composition 635 gradients due to proteolytic, lipolytic and dehydration phenomena that occur du r ing 636 maturation, the information obtained from the external s urface i s s ufficient to predict its 637 compo s ition s ati s factory accuracy. Hence, the s e model s can be u s ed to predict the 638 compo s ition of Brazilian chee s e s u s ed in thi s s tudy and al s o enhanced including additional 639 type s of chee s e and can be u s eful to s tudy compo s itional change s that occur du r ing 640 maturation. 641 642 4. Conclu s io n 643 Thi s s tudy compared the fea s ibility of hyper s pectral imaging (vi s /NIR-H S I, NIR-H SI and 644 data fu s ion) and conventional ( N IR S ) s pectrometer s to characterize Brazilian traditional 645 chee s e s . P rincipal component anal ys i s ( P CA) s howed that the s pectral variability of Brazilian 646 traditional chee s e s in the vi s /NIR i s related to difference s in color and fat content, while in the 647 NIR there i s a greater influence of productive a s pect s and fat content. 648 The PL S -DA model s s howed that the s pectrum in the near infrared region (NIRS ) has 649 relevant information to di s criminate the type s of Brazilian arti s anal chee s e s , indicating that 650 thi s technique can be u s ed a s a u s eful tool in inve s tigating the authenticity of cheeses 651 according to the region of origin. 652 The PL S model s demon s trated that it i s po ss ible to efficiently predict the chemical 653 compo s ition (moi s ture, fat, and protein) of a wide type of chee s e s . Hyper s pectral imaging 33 654 equipment (vi s /NIR-H S I and NIRHS I) can be recommended, e s pecially when there i s g r eat 655 heterogeneity within the s ample and/or when there i s intere s t in s tudying and vi s ualizing 656 compo s itional change s , a s the general quantitative re s ult s obtained are comparable to the 657 technique punctual (NIR S ). The latter, in turn, i s more attractive from an economic and 658 operatio n al point of view, due to l o wer co s t, portability and s horter proce ss ing time. F i n ally, 659 and in thi s ca s e, the fu s ion of H S I s pectro s copic data (vi s /NIR + NIR) did not provide 660 s ignificant improvement of the predictive capability to ju s tify it s recommendation, because 661 de s pite providing better performance s to predict moi s ture and protein s , the s e are not 662 s ub s tantially greater to ju s tify it s use. 663 664 Declar a tio n of Com p eti n g In terest 665 The author s declare that they have no known competing financial intere s t s or pe r sonal 666 relation s hip s that could have appeared to influence the work reported in thi s paper. 667 668 Ack n owle d gme n t s 669 Thi s s tudy wa s financed in part by the Coordenação de Aperfeiçoamento de P e ss oal de Nível 670 Superior - Bra s il (CA P E S ) - F inance Code 001. M aria Lucimar da S ilva M edei r os 671 acknowledge s s cholar s hip funding from CN P q (300145/2022-5) and S antander international 672 mobility program for graduate s tu d ent s . P rof. Dougla s F ernande s Barbin i s CN P q resea r ch 673 fellow (308260/2021-0). 674 675 Refere n ce s 676 Andrade, B. M ., M argalho, L. P ., Bati s ta, D. B., Lucena, I. O., Kamimura, B. A., Balthaza r , 677 C. F., Brexó, R. P ., P ia, A. K. R., Co s ta, R. A. S ., Cruz, A. G., Granato, D., S ant’Ana, 678 A. S ., Luna, A. S ., & de Goi s , J . S . (2022). Chemometric cla ss ification of Brazilian 679 arti s anal chee s e s from different region s according to major and trace element s by IC P- 34 680 OE S . Jou r nal of Food Compo s ition and Analy s i s , 109, 104519. 681 http s ://doi.org/10.1016/j.jfca.2022.104519 682 Amigo, J. M . (2019). Hype rs pect r al and multi s pect r al imaging: s etting the s cene (pp. 3–16 ) . 683 http s ://doi.org/10.1016/B978-0-444-63977-6.00001-8 684 AOAC. (2012). Official Method s of Analy s i s (19th ed.). AOAC Internacional. 685 Ayvaz, H ., M orta s , M ., Dogan, M . A ., Atan, M ., Yildiz Tiryaki, G., & Karagul Yuceer, Y . 686 (2021). Near- and mid-infrared determination of s ome quality parameter s of chee s e 687 manufactured from the mixture of different milk s pecie s . Jou r nal of Food Science and 688 Technology, 58(10), 3981–3992. http s ://doi.org/10.1007/ s 13197-020-04861-0 689 Becker, E. M ., Chri s ten s en, J ., F rederik s en, C. S ., & Haugaard, V. K. (2003). F ront- F ace 690 Fluore s cence S pectro s copy and Chemometric s in Analy s i s of Yogurt: Rapid Analysis o f 691 Riboflavin. Jou r nal of Dai r y Science, 86(8), 2508–2515. 692 http s ://doi.org/10.3168/jd s . S0 022-0302(03)73845-4 693 Bezerra, T. K. A., Arcanjo, N. M . O., Garcia, E. F ., Gome s , A. M . P ., Queiroga, R. de C. R. 694 do E., S ouza, E. L., & M adruga, M . S . (2017). Effect of s upplementation with probiotic 695 lactic acid bacteria, s eparately or combined, on acid and s ugar production in goat 696 ‘coalho’ chee s e. L W T , 75, 710–718. http s ://doi.org/10.1016/j.lwt.2016.10.023 697 Boran, O . S ., S ulejmani, E., & Hayaloglu, A. A. (2023). Acceleration of proteoly s i s , flavou r 698 development and enhanced bioactivity in a model chee s e u s ing Kuflu chee s e s lurry: An 699 optimi s ation s tudy. Food Chemi s t r y, 412, 135495. 700 http s ://doi.org/10.1016/j.foodchem.2023.135495 701 Bra s il (1996). Regulamento s Técnico s de Identidade e Qualidade do s P roduto s Lácteo s . 702 Portaria n° 146 de 07 de março de 1996, P ub. L. No. 146, Diário Oficial da União. 703 Retrieved M arch 17, 2023 from: 704 https://www.defesa.agricultura.sp.gov.br/legislacoes/portaria-mapa-146-de-07-03- 705 1996,669.html 706 Bra s il (1996). Regula direito s e obrigaçõe s relativo s à propriedade indu s trial. Lei nº 9.279, de 707 14 de maio de 1996, P ub. L. No. 9.279, Diário Oficial da União (1997). Retrieved 708 february 17, 2024 from: http s ://www.planalto.gov.br/ccivil_03/lei s /l9279.htm 709 Bra s il (1997). Regulamento Técnic o para F ixação de Identidade e Qualidade de Queijo M inas 710 Fre s cal. P ortaria n° 352 de 0 4 de s etembro de 1997, P ub. L. No. 352, Diário Oficial da 711 União (1997). Retrieved march 17, 2023 from: 712 http s :// s idago.agrodefe s a.go.gov.br/ s ite/adicionai s proprio s /protocolo/arquivo s /409 8 53.p 713 df 714 Bra s il (2001). Regulamento s Técnico s de Identidade e Qualidade de M anteiga da Terra ou 715 M anteiga de Ganafa; Queijo de Coalho e Queijo de M anteiga. In s trução Normativa n o 716 30 de 26 de junho de 2001, P ub. L. No. 30, Diário Oficial da união. Retrieved 717 December 08, 2022 from: 718 https://pesquisa.in.gov.br/imprensa/jsp/visualiza/index.jsp?data=16/07/2001&jornal=1& 719 pagina=13&totalArquivos=219 720 Bra s il (2018). Di s põe s obre o proce ss o de fi s calização de produto s alimentício s de origem 721 animal produzido s de forma arte s anal. Lei n o 13.680, de 14 de junho de 2018. Retrieved 722 march 17, 2023 from: http://www.planalto.gov.br/ccivil_03/_ato2015- 723 2018/2018/lei/L13680.htm 724 Bra s il (2020). Identidade e o s requi s ito s de qualidade que deve apre s entar o produto 725 denominado queijo mina s padrão. In s trução Normativa n o 66 de 21 de julho de 2020, 726 Pub. L. No. 66, Diário Oficial da União. Retrieved march 17, 2023 from: 727 http s ://www.in.gov.br/en/we b /dou/-/in s trucao-normativa-n-66-de-21-de-julho-de- 2 020 - 728 268265894 35 729 Britton, G., Liaaen- J en s en, S ., & Pf ander, H. (2004). Ca r otenoid s Handbook. Birkhdu s e r 730 Ve r lag. 731 CIE, C. I. de l’Éclairage. (1978). Recommendation s on unifo r m colou r s pace s , colou r - 732 diffe r ence equation s , p s ychomet r ic colou r te r m s . Supplement No.2 to CIE 15-1971. 733 Bu r eau Central de la CIE P ari s . 734 Co s ta, J . R., P ereira, D. A., de P aula, I. L., de Abreu, L. R., P into, S . M ., Edward s , H. G. M., 735 Stephani, R., & de Oliveira, L. F . C. (2022). The ta s te of a champion: Characterization 736 of traditional chee s e s from the M ina s Gerai s region (Brazil) by Raman s pectro s copy and 737 micro s tructural analy s i s . Jou r nal of Food Compo s ition and Analy s i s , 112, 104704. 738 http s ://doi.org/10.1016/j.jfca.2022.104704 739 Fitztum, A. C., Lima, L. S ., J udacew s ki, P ., S alem, R. D. S ., de Dea Lindner, J ., Demiate, I . 740 M ., Zielin s ki, A. A. F ., Alberti, A., & Nogueira, A. (2023). Evaluation of traditional 741 s emi‐hard Brazilian chee s e s u s ing chemometric tool s to identify po ss ible denomination 742 fraud. Inte r national Jou r nal of Dai r y T echnology. http s ://doi.org/10.1111/1471- 743 0307.12954 744 Franci s , F . J ., & Clyde s dale, F . M . ( 1975). Food Colo r imet r y: T heo r y and Application s . A V I 745 Publi s hing. 746 Frank, J . F ., & Birth, G. S . (1982). A pplication of Near Infrared Reflectance S pectro s copy to 747 Chee s e Analy s i s . Jou r nal of Dai r y Science, 65(7), 1110–1116. 748 https://doi.org/10.3168/jds.S0022-0302(82)82319-9 749 Garcia, A., Alting, A., & Huppertz, T. (2023). Di s ruption of ca s ein micelle s by calcium 750 s eque s tering s alt s : F rom ob s ervation s to mechani s tic in s ight s . Inte r national Dai r y 751 Jou r nal, 142, 105638. https://doi.org/10.1016/j.idairyj.2023.105638 752 Gentili, A ., Caretti, F ., Bellante, S ., Ventura, S ., Canepari, S ., & Curini, R. (2013). 753 Comprehen s ive P rofiling of Carotenoid s and F at- S oluble Vitamin s in M ilk from 754 Di f ferent Animal S pecie s by LC-DAD- MS / MS Hyphenation. Jou r nal of Ag r icultural 755 and Food Chemi s t r y, 61(8), 1628–1639. http s ://doi.org/10.1021/jf302811a 756 Hayaloglu, A. A., & M c S weeney, P. L. H. (2014). P rimary biochemical event s during c h eese 757 ripening. In Dai r y Mic r obiology and Biochemi s t r y: Recent Development s . 758 10.1201/b17297-8. 759 Herbert, S ., Riaublanc, A., Bouchet, B., Gallant, D. J ., & Dufour, E. (1999). F luore s cence 760 Spectro s copy Inve s tigation of Acid-or Rennet-Induced Coagulation of M ilk. Jou r nal of 761 Dai r y Science, 82(10), 2056–2062. http s ://doi.org/10.3168/jd s . S 0022-0302(99)75446 - 9 762 Ibáñez, R. A., Govinda s amy-Lucey, S ., J aeggi, J . J ., J ohn s on, M . E., M c S weeney, P . L. H., & 763 Lucey, J . A. (2020). Low- and reduced-fat milled curd, directs alted Gouda chee s e: 764 Compari s on of lacto s e s tanda r dization of chee s emilk and whey dilution technique s. 765 Jou r nal of Dai r y Science, 10 3 (2), 1175–1192. https://doi.org/10.3168/jds.2019-17292 766 J e s u s F ilho, M ., Klein, B., Wagner, R., & Godoy, H. T. (2021). Key aroma compound s o f 767 Cana s tra chee s e: H S - SPM E optimization a ss i s ted by olfactometry and chemometrics. 768 Food Re s ea r ch Inte r national, 150, 110788. 769 https://doi.org/10.1016/j.foodres.2021.110788 770 J e s u s F ilho, M ., Klein, B., Quintão Teixeira, L. J ., S ilva, J . G. S ., P allone, J . A. L., Wagne r , 771 R., & Godoy, H. T. (2023). The influence of production unit s and s ea s on s on the 772 phy s icochemical characteri s tic s , mineral and fatty acid content, and texture profile o f 773 the arti s anal chee s e s from S e r ra da Cana s tra, Brazil. Jou r nal of Food Compo s ition and 774 Analy s i s , 123, 105589. http s ://doi.org/10.1016/j.jfca.2023.105589 775 Karoui, R., M ouazen, A. M ., Dufo u r, É., P illonel, L., S challer, E., P icque, D., de 776 Baerdemaeker, J ., & Bo ss et, J.-O. (2006). A compari s on and joint u s e of NIR and M I R 777 s pectro s copic method s for the determination of s ome parameter s in European Emmental 36