3.1. Instrumental Comparison on Canadian Colour Standard
From score 0 to score 6, the Spectro 1 Pro instruments had the highest
L*, and the Spectro 1 instruments showed the lowest
L* value (
Table 1). The difference between the lowest and the highest value ranged from 2 to 3 units, which indicated the lightness value from Spectro 1 Pro was higher than Spectro 1. For
a*, the Nix Pro II instruments displayed the highest values from score 0 to 6. The Spectro 1 instrument had the lowest
a* value from score 0 to 3, and the Spectro 1 Pro instrument had the lowest
a* value from score 4 to 6. For
b* value, the Spectro 1 instruments had the highest values for score 0 and score 1 but had the lowest value from score 4 to score 6. For score 3, the Minolta instruments presented the highest
b* value of 11.41, and the Spectro 1 Pro instruments had the lowest
b* value of 8.51. From score 4 to 6, the Nix Pro II instruments produced the highest
b* value. The major differences observed among instrument models may be due to their technical specifications. The Nix Pro II instruments use a different native illuminant (D
50, 2°), with the illuminant D
65 10° setting being mathematically converted. Spectrophotometers, such as the Minolta CM 700 and the Spectro 1 instruments, use reflected light to quantify energy across the light spectrum. The reflected energy amount is then converted to colour value. Colourimeters, such as the Nix instruments, use a detector to measure the energy which passes through the filter (between the light source and sample) and reflects from the sample surface [
16]. Additionally, each instrument uses different algorithms to calculate colour value, which may not indicate the changes on Canadian pork colour standard in the same degree.
In addition to descriptive statistics of colour measurements, the Bland-Altman plot (
Figure 1) revealed that for
L* value, both Nix and Spectro series instruments presented relatively less bias (~1), and the range of agreement was narrower, compared to
a* and
b* value. For
a* value, the Nix Pro II had the greater bias (−5.24), compared to other instruments, and this bias was negative. For
b* value, the Nix series instruments presented less bias (0.36–0.48), compared to Spectro 1 series (2.51–2.62). When
b* value ranged from 5–10, Nix series instruments tended to have greater bias to Minolta measurements compared to the Spectro 1 series. The mean comparison results also indicated that the instruments from the same manufacturer produced similar bias (Nix vs. Spectro 1). The only exception was for the
a* value generated by Nix Pro II. These differences in colour measurements, especially for
a* and
b* value, may be due to the different calculation mechanisms of reflected surface light. In addition, the Nix Pro II device uses native D
50 illuminant at a 2° observer angle, which is then mathematically converted by the instrument to D
65 at 10° [
17,
18]. While the Spectro 1 instruments are designed to obtain different values from shiny and matte surfaces (“visual colour”), the Spectro 1 Pro instruments account for this difference and provide the same results for both surfaces (“actual colour”). Overall, all instruments tended to generate equal increments or decrements for lower colour scores.
The RSD results of intra-day analysis were all under 5%, and ranged from 0.01 to 0.03%, 0.05 to 0.18%, and 0.08 to 0.46%, for
L*,
a*,
b* trait, separately (
Table 2). For inter-day analysis, RSD results for
L*,
a*,
b* were all under 1%, except for
b* from the Nix Pro II instruments (1.11%) and the Spectro 1 Pro instruments (1.59%). These RSD values ranged from 0.07 to 0.48%, 0.21 to 0.98%, and 0.50 to 1.59%, for
L*,
a*,
b* traits, respectively. RSD results of intra-model ranged from 0.21 to 0.62%, 0.41 to 2.35%, and 0.53 to 2.84%, for
L*,
a*,
b*, respectively. Among all instruments, the Minolta and the Nix QC were the only two instruments with all RSD values below 1%. According to the manufacturers, the colour difference
(∆E00) value of short-term measurement and inter-instrument agreement for the Spectro 1 series instruments are 0.05, and 0.2–0.5, respectively [
17], and for the Nix series instruments are 0.1, and 0.30–0.75, respectively [
18]. Apart from these product specifications, one Spectro 1 Pro instrument had unexpected calibration issues during the measurement process, which may have slightly altered the RSD results. Typically, accepted RSD values are lower than 5%, and the observed low RSD values in the current study suggest the data had low variability [
19]. Although all instruments had RSD values < 5%, some instruments had lower values (<2%), which would be beneficial if implementing an instrument for colour measurement with high accuracy, repeatability and reproducibility in systems that require greater precision, such as research environments or quality-based export markets with low levels of acceptable sorting error.
3.2. Instrumental Comparison on Retail Meat Samples
The average
L* values of fresh pork samples from the five instrument models ranged from 44 to 55, and the Nix and Spectro 1 instruments were significantly different (
p < 0.05) from the Minolta instruments (
Table 3). Minolta instruments presented the highest
L* values, and Nix Pro II instruments presented the lowest
L* values. The average
a* values ranged from 4.3 (Spectro 1 Pro) to 7.7 (Nix Pro II), and the Spectro 1 instruments were not significantly different (
p > 0.05) from Spectro 1 Pro instruments. The average
b* values ranged from 8.1 (Nix Pro II) to 13.5 (Minolta). Nix Pro II instruments were not significantly different from Nix QC instruments, and Spectro 1 instruments were not significantly different from Spectro 1 Pro instruments (
p > 0.05).
Although instruments generated significantly different
L* coordinate values when compared with Canadian colour standard scores (
Table 1), the difference between maximum and minimum values were lower than the differences observed on meat samples. This may be due to the different aperture diameter of each instrument and, therefore, the percentage of meat surface being measured. The larger aperture size could also infer a comparatively greater susceptibility to edge-losses, indicating light on the meat surface would be not reflected, and was considered absorbed (Holman et al., 2015). In turn, the reflectance would interfere with
L* (lightness). Meanwhile, meat surface is not homogeneous, unlike the surface of a Canadian colour standard. The presence of two-tone colours, connective tissues, and intramuscular fat on the meat surface may result in inconsistent measurement results [
20,
21]. Particular to the current study, darker tenderloins were also mixed and measured by all instruments, which could lead to the larger variance on colour coordinates.
For whole muscle measurements, a strong correlation (|r| = 0.8 to 0.9) was observed for the average
L* and
a* value for the Nix and Spectro 1 instruments with the Minolta instruments (
Table 4). For
b* value, the correlation between Nix and Minolta instruments was close to 0.7, > 0.8 for the Spectro 1 instruments, but lower for the Spectro 1 Pro instruments (|r| = 0.32). Similar results were observed for central area measurements, indicating that a narrower surface colour range (central area), and the number of measurements had very limited effect on correlation results. Additionally, the lack of effect between whole muscles and central areas could be due to the central areas being adequately representative of variability in the muscles in this study. In theory, more replicates or measurements would decrease the standard error of mean [
19]. In practice, based on the size of surface area, replicates of measurement could vary from 3 to 30 [
6]. In a previous study, seven measurements could contribute to minimize the SEM (standard error of predicted mean) value when assessing beef colour via Nix Pro instruments [
8]. In the current study, correlation coefficients of novel instruments to Minolta instruments did not present differently between four and eight measurements.
Correlation is a statistical approach for determining if and how closely two variables are related. However, a strong correlation does not inherently suggest that the two methodologies are in accord [
22]. The Bland-Altman plot (
Figure 2) based on whole muscles revealed that for
L* value, Spectro 1 series instruments presented relatively less bias (3.07–5.10), and the range of agreement was narrower, compared to Nix series instruments (bias: 9.54–11.25). The plot of differences from central areas presented similar results (
Figure 3). But all four hand-held instruments presented significant (
p < 0.05) strong correlation (|r| > 0.8) to Minolta. At the
L* value range of 60–70, Spectro 1 and Nix instruments had lower bias when compared to Minolta. For
a* value, the Nix Pro II had the greater bias (−2.61), compared to other instruments. When
b* value ranged from 5–15, all instruments tended to have less bias to Minolta measurements. However, Spectro 1 Pro presented greater agreement bias, when
b* value ranged from 10–20, when compared to the rest of recently developed instruments. The calibration protocols might also alter the measurements. Minolta, Nix QC, and Spectro 1 series instruments use calibration tiles, while Nix Pro II instruments use built-in software calibration. Different calibration techniques may lead to inconsistencies of colour coordinates [
9]. At this point, it may not be conclusive to determine why these instruments produced different colour coordinates for pork samples in the current study. As the benchmark classification instrument, Minolta instruments have often been used to estimate standard colourimetric thresholds. The difference in absolute number value of colour coordinates from all instruments must be taken into account to avoid false positive or negative results before establishing alternative colour thresholds [
10]. The colour acceptability threshold equated to the corresponding colour coordinate should be adjusted, depending on which specific instrument was used.
The measuring mechanism of the surface colour might impact the determination of
b* value. As discussed in the previous section, Spectro 1 Pro instruments account for the shine of a surface. This feature may have had a large impact on the correlation of
b* value to the Minolta instruments. It might have been more important to investigate
L* and
a* values when instrumental meat colour was measured, as the actual meat colour was a combination of
L*,
a*, and
b* values. The higher correlation of the
L* and
a* value between novel instruments and Minolta, compared to correlations of
b* values, may suggest that the
L* and
a* value may assess the lightness and red colour of fresh meat as
b* values express the colour in the yellow-blue region of the spectrum, and the actual meat colour may present less colour variability in the yellow-blue region [
23]. Additionally, while the Minolta was equipped with a xenon lamp as the light source, the Nix and Spectro 1 instruments use LED lamps. The different types of light sources displayed different emission spectra, which would affect how the colour was subjectively perceived. Red colour generally ranges from 600 to 700 nm in wavelength, and LED light displayed lower relative intensity (%) than xenon light in this region, which could result in different colour readings from instruments [
24].
Since fat deposits and water residue on the surface may alter the instrument measurement of surface colour
b* value [
9], instruments designed to account for or ignore these differences would provide different results for this variable. In this study, the marbling content in tenderloin samples was negligible. Although heavy marbling regions on loin chops were excluded before dividing the meat surface into different locations for measurement, the trace amount of marbling might not be completely avoided, which would result in different number values on colour coordinates from different instruments. The strong correlation and mean difference comparison of novel instruments on colour coordinates (except
b* from the Spectro 1 pro instrument) to the Minolta instruments revealed the potential of these instruments in developing alternative pork classification systems based on colour traits.
3.3. Instrumental Measurements and Subjective Standards
For whole muscle measurements, the correlation coefficients (|r|) between Canadian subjective standards and
L* values from Nix instruments were >0.8 and were close to 0.8 from Spectro 1 series instruments and the |r| from Minolta spectrophotometer was <0.7 (
Table 5). The correlation between Canadian subjective standards and
a* value from Nix instruments and Spectro 1 instruments ranged from 0.6–0.7. Minolta instruments presented lower correlation (|r| = 0.47). Correlation between Canadian subjective standards and
b* value from Nix instruments and Spectro 1 instruments was between ~0.3–0.4, and Minolta instruments had a lower correlation (|r| = 0.14).
For the central area measurements, the results were similar to whole muscle measurements (
Table 5), with the exception that Minolta had increased correlations (|r|) on central area measurements for
L* (0.68–0.76) and
a* values (0.47–0.59). Similarly, the correlation of all instruments to the Japanese subjective standard were very similar (
Table 6) to those with the Canadian subjective standards. The current study of Minolta instrument correlations presented stronger coefficients for
L* and
a* values, but very poor coefficients for
b* value. Interestingly, for
b* value, Spectro 1 instruments presented negative correlations, but Spectro 1 Pro instruments presented positive correlations. As discussed, the difference may be due to the instrumental calculation of the surface shine. The Canadian colour standard is divided into seven colour scores, which vary from pale pink (score 0) to dark red (score 6). Similarly, Japanese colour standards are divided into six colour scores. The visual appraisal from human eyes may not be indicative of different colour coordinates, as the subjective red colour in meat is a combination of
L*,
a*, and
b* values [
23]. The
b* values had a small range when compared to the ranges in the
L* and
a* values (
Table 1), which could have had less correlation to human score assignments.
In general, visual colour assessments are performed by trained personnel, and despite regular calibration and proper training, human errors are difficult to avoid [
25]. However, the above hand-held instruments are also frequently operated by humans. In order to minimize human error, there is a need to study alternative innovative colour assessment instruments that can be automated. Although small hand-held devices like those evaluated in this study present potential for automation due to their compact size and connectivity capabilities, current versions may present difficulties for adaptation to in-plant use. For example, Spectro 1 series instruments automatically initiate self-correction procedures after approximately 10 measurements, which increase the single scanning time to up to 60 s. This feature guarantees the adequate performance of the instrument but presents limitations to their efficiency in meat processing lines, where an efficient scanning time is critical to meet processing speed. Another future application for these hand-held devices may be the establishment of acceptability thresholds for fresh and displayed meat for various settings, such as wholesale stores and retail stores. Furthermore, the importance of accurate and efficient assessment of
a* values is emphasized, due to its association with fresh colour stability during storage [
26,
27]. Additionally, other important pork quality traits, such as pH and drip loss, may also correlate with instrumental colour measurement, thereby highlighting the need for a more comprehensive pork quality assessment system that could be established in the future.