TWI684763B - Method for detecting meat quality using nir spectrum - Google Patents

Abstract

The present disclosure provides a method for detecting meat quality using NIR spectrum. The method comprises following steps: (a) obtain a plurality of qualified meat samples and a meat analyte; (b) scan the qualified meat samples by NIR spectrometer; (c) analyze the spectral data of step (b) to get a qualified meat characteristic waveform, the waveform includes the range form 479-644 nm, 1353-1450 nm and/or 1846-1998 nm; (d) analyze the meat analyte as steps (b) and (c) to get a meat analyte characteristic waveform; and (e) compare the qualified meat characteristic waveform and the meat analyte characteristic waveform.

Description

Meat quality testing method using near infrared spectroscopy

The present disclosure relates to a method for detecting the quality of meat using near infrared spectroscopy, in particular to a method for identifying normal and bad roasted meat.

The operation that causes the substance to cause physical and chemical changes in a dry and hot state at a higher temperature is called roasting. It is commonly used in the production of livestock products (floss, meat crisps). Roasting has a denaturing effect on the protein in the raw materials, a gelatinizing effect on the starch, and a coking effect on the sugar, and it can also dry the raw materials, produce golden yellow color and unique aroma, and also have a bactericidal effect.

However, in the mass production process of the food industry, the roasted meat often suffers from burnt black particles and burnt black lumps due to uneven heating, which affects the quality of the product. In the food industry, the experience and sensory identification of on-site operators are used as the standard for production process adjustment and product quality judgment, which consumes manpower and time costs. According to this, the traditional method of identifying the quality of roasted products can no longer keep up with the needs of the industry, and may hinder the upgrading of the industry.

One purpose of this disclosure is to provide a meat quality testing method that responds to the industry 4.0 policy direction (intelligent integrated sensory control system), with the core of automated testing to achieve automatic and rapid monitoring.

To achieve the above purpose, the present disclosure provides a method for detecting meat quality using near infrared spectroscopy. The method includes the following steps:

(a) Obtain multiple normal meat samples and one meat sample to be tested;

(b) Scan these normal meat samples with a near-infrared spectrometer to collect atlases of these normal meat samples;

(c) Analyze the maps collected in step (b) to obtain the characteristic profile of normal meat, wherein the characteristic profile of normal meat includes the wavelength range of 479-644 nm, 1353-1450 nm and/or 1846-1998 nm Map

(d) processing the meat sample to be tested in steps (b) and (c) to obtain a characteristic profile of the meat to be tested; and

(e) Compare the characteristic profile of the meat to be tested with the characteristic profile of the normal meat.

In an embodiment, the method for analyzing the graph in step (c) includes a standard normal variable (SNV) and a second derivative.

In one embodiment, the normal meat sample and the meat sample to be tested in step (a) are roasted meat.

In one embodiment, the roasted meat is meat floss or meat crisp.

In one embodiment, the average number of scans of normal meat samples in step (b) is 25-30.

In one embodiment, in step (d), the average number of scans of the meat sample to be tested is 25-30 times.

In one embodiment, the method of comparison in step (e) is to obtain a normal threshold range of the characteristic profile of the normal sample, and then compare whether the characteristic profile of the sample to be tested falls within the normal sample threshold range.

In one embodiment, if the characteristic contour of the meat to be tested does not fall within the normal threshold range, the meat sample to be tested is bad meat.

In an embodiment, if the characteristic contour of the meat to be tested falls within the normal threshold range, the meat sample to be tested is normal meat.

In one embodiment, the normal threshold range and threshold are calculated by the qualitative analysis function of VISION software.

In one embodiment, in step (a), the color of the meat sample is used as a reference to determine whether it is a normal meat sample.

In one embodiment, the undesirable meat products include burnt black meat and/or burnt black meat.

According to the present disclosure, another method for detecting meat quality using near infrared spectroscopy is provided. The method includes the following steps:

Scan a meat product to be tested with a near infrared spectrometer to obtain a map of the meat product to be tested;

Analyzing the spectrum to obtain a characteristic profile, wherein the characteristic profile includes a spectrum in the wavelength range of 479-644 nm, 1353-1450 nm and/or 1846-1998 nm; and

Compare the characteristic profile with a predetermined normal meat profile.

In order to make the above and other aspects of the disclosure more clear and understandable, the embodiments are specifically described below and described in detail in conjunction with the accompanying drawings.

According to different experimental techniques and applications, the infrared light region can be divided into three bands: near-infrared region (wavelength 750-2,500 nm, wave number 13,158-4,000 cm ^{-1 )} , and the energy transition type is the frequency doubling of OH, NH, CH absorption), mid-infrared region (wavelength 2,500-25,000 nm, the wave number 4,000-400 cm ^-1, molecular vibration type level transitions and rotation) and the far-infrared region (wavelength 25,000-300,000 nm, the wave number 400-10 cm ^{- 1. The} type of energy level transition is molecular rotation). In the past, the mid-infrared region has been the most researched and applied, mainly used for qualitative analysis and structural analysis. The development of near-infrared light is limited by the difficulty of analysis of the spectrum. With the advancement of multivariate statistical software systems, there has been a lot of research.

The use of near infrared spectroscopy (Near infrared spectroscopy, NIRS) to analyze compounds is a non-destructive analysis method. The main object of the analysis of the near infrared spectrometer is the organic component. If the inorganic component forms a bond with the organic component or water, the structure with C-H, O-H, N-H and other sensing functional groups can also be measured. These functional groups can absorb specific energy and cause bending, elongation and vibration of the bonding within the functional group, so the absorption band of the spectrum exhibits different absorbance values according to the type and concentration of the functional groups unique to each component, which can be inferred The type and content of the compound achieve the purpose of qualitative and quantitative analysis, which is an indirect measurement method.

The present disclosure provides a meat quality detection method using near infrared spectroscopy. In this method, the characteristic band (characteristic profile) of a normal meat product is first established. This characteristic band is used for comparison with other meat products to be tested to detect whether the meat product to be tested is a defective product. This method includes the following steps:

(a) Obtain normal meat samples and meat samples to be tested;

(b) Scan normal meat samples with a near-infrared spectrometer to collect atlases of normal meat samples;

(c) Analyze the map collected in step (b) to obtain the characteristic profile of normal meat. The characteristic profile of this normal meat product includes a spectrum of wavelengths in the 479-644 nm, 1353-1450 nm and/or 1846-1998 nm bands;

(d) processing the meat sample to be tested in steps (b) and (c) to obtain the characteristic profile of the meat to be tested; and

(e) Compare the characteristic profile of the meat to be tested with the characteristic profile of normal meat.

If the characteristic profile of a normal meat sample is known, the step (a) of obtaining a normal meat sample can be omitted, and the meat to be tested is directly scanned with a near-infrared spectrometer to obtain its map; then the map of the meat to be tested is analyzed to obtain The characteristic contour of the meat to be tested; then compare the characteristic contour of the meat to be tested with the characteristic contour of the known normal meat.

In this embodiment, the following instruments are used to analyze the quality of meat. This disclosure is about a meat quality inspection method using near infrared spectroscopy (including part of visible light). In order to establish a model for this meat quality detection method, the detection data of the color difference meter is used as the numerical standard for normal/defective products, and the spectral information of the mid-infrared spectrometer is used to evaluate and confirm the discriminating functional groups of the spectral characteristic profile.

Use the instrument:

Mid-infrared spectrometer (Frontier, Perkin Elmer, USA) with a wavenumber range of 4,000-650 cm ^-1 and a wavelength range of 2,500-15,385 nm; near-infrared spectrometer (NIRSystemsXDS, Methrohm, Switzerland) with a wavelength range of 400-2,500 nm, Aperture size is 17.25 mm; color difference meter (Spectrophotometer CM-5, KONICA MINOLTA).

Sample Preparation:

In this embodiment, pork floss in roasted meat is used as a meat sample. However, the detection method of the present disclosure is not limited to this, and other types of roasted meat such as meat crisps can also be applied. In this embodiment, pork floss is first collected and classified into normal samples and bad samples. Due to over-roasted meat products, coking will occur, and normal samples and defective products are distinguished by color and firmness. As shown in Figure 1, the normal sample group includes normal products obtained from food processing plants, standard color, upper color (dark) and lower color (light) pork floss. The bad samples were selected by the personnel of the food processing plant based on visual judgment (dark color, pork pine with excessive heating), followed by another step according to the difference in firmness and color status, it was divided into burnt black particles and burnt black clumps ( Particles that are too large are called clumps). For each sample, a colorimeter is used as the numerical standard for color depth. The values are shown in Table 1 below. In practical applications, the thresholds for normal samples and defective products of this category can be adjusted by themselves for different types, brands, batches or processes of meat.

Table 1 Normal and bad samples colorimeter test values L value a value b value WI (whiteness) Normal sample Normal goods 34.7±0.5 11.8±0.1 15.7±0.2 31.8±0.4 Lower limit color (light) 35.0±0.6 10.3±0.1 14.5±0.2 32.6±0.6 standard color 33.6±0.5 12.2±0.3 14.8±0.4 30.8±0.5 Upper limit color (dark) 30.7±0.3 11.0±0.2 12.8±0.2 28.7±0.3 Bad sample Scorched black particles 29.4±1.3 7.3±0.4 10.6±0.9 28.2±1.2 Scorched black lump 26.9±1.2 5.6±0.5 8.8±0.7 26.2±1.1

The following exemplary experiments will mix bad samples and normal samples in different proportions (based on weight), and then mix them by stirring to confirm that the detection method of the present disclosure can indeed identify bad meat samples.

Spectral detection and data processing:

For each sample in Table 1 (additional 2 other commercially available normal products), use a mid-infrared spectrometer to collect spectral data in an Attenuated Total Reflectance (ATR) mode. The number of scans per map is 4 and each sample The number of repetitions is 15-30. After the baseline is processed, the absorption value is acquired, and then the calculation is performed according to the selected band.

Each sample was also scanned with a near-infrared spectrometer. After measuring 4 grams of each sample, the sample was placed in a quartz cup and scanned after compaction. The number of replicates for each sample was 25-30. After obtaining the near-infrared spectrum, the standard data (Standard Normal Variate, SNV) and second derivative are used to pre-process the spectral data to correct the situation of spectral scattering and baseline drift.

Spectral system evaluation and characteristic band selection:

The near-infrared spectrum of each sample in Table 1 and its processing results are shown in Figures 2 and 3. Normal samples of pork floss include commercially available normal products, standard colors, upper color limit (dark color), lower color limit (light color), and defective products (scorched particles/clumps). There is a significant absorption in the visible range (approximately 470 nm) in the original spectrum of Figure 2. In the near-infrared range, there is a difference in the waveform of the 1900-2000 nm band. In order to remove the baseline shift in order to observe the change of the spectrum, the data is further mathematically converted, and the test samples can be observed at 426-608, 1358-1409, 1457-1529, 1693-1773, 1817-1971, 2037-2075, 2220-2254 And 2288-2374 nm are slightly different in these wavelengths. Therefore, as shown in Figures 3(A), (B), and (C), this disclosure selects 479-644 nm, 1353-1450 nm, and 1846-1998, where the variation of the radiation is more obvious and the normal samples can be more concentrated. nm is the three representative characteristic bands (characteristic profile, selected from the corresponding bands in Figure 2, processed by quadratic differentiation and enlarged to confirm the difference).

Comparison of the identification effects of different characteristic bands:

Please refer to Tables 2 and 3 below, and mix the normal sample and the bad sample of Table 1 into the sample to be tested at different weight ratios to evaluate the identification effect of different characteristic bands on the bad sample. Among them, Table 2 is normal samples mixed with unclassified bad samples, and Table 3 is normal samples mixed with classified bad samples (scorched black lumps and black particles). The causes of burnt black lumps and burnt black particles are mainly caused by unevenness in the roasting (heating) of meat. In the definition of this specification, the difference between the two lies in the particle size (refer to Figure 1), that is, the size of the pyro-black agglomerates is larger than the pyro-black particles. After the normal sample and the bad sample are uniformly mixed according to the weight ratios in Tables 2 and 3, scan with a near infrared spectrometer using the sampling method described in paragraph [0046], collect the spectrum and analyze it (data pre-processing). The characteristic bands (characteristic profiles) in the three wavelength ranges of 479-644 nm, 1353-1450 nm and 1846-1998 nm were taken from the spectrum and compared with the characteristic bands of normal samples.

The comparison method uses the qualitative analysis function of statistical software such as VISION 4.1.1.54 software, and the parameter settings in Qualification Method (such as threshold type and value...). After entering multiple normal samples, the threshold range of normal samples can be obtained. If the characteristic contour of the sample to be tested does not fall within the set threshold range of the normal sample, it means that the sample to be tested does not belong to the normal sample. Conversely, if the characteristic contour of the sample to be tested falls within the set threshold range of the normal sample, the sample to be tested belongs to the normal sample.

Table 2 The accuracy rate of different characteristic wave bands to the test samples of poorly mixed samples (not classified) Normal sample: bad sample (bad sample ratio) characteristic band (profile) 1: 4 (80%) 1: 2 (67%) 1:1 (50%) 2:1 (33%) 4:1 (20%) Visible light + near infrared region 479-644+1353-1450+1846-1998 nm 100 100 100 100 100 Near infrared region 1353-1450+1846-1998 nm 100 100 100 100 100 1353-1450 nm 100 100 100 100 100 1846-1998 nm 100 100 100 100 100 Visible area 479-644nm 93.7 86.7 67.7 36.7 23.3 *Unit is percentage, n=30

Table 3 The accuracy rate of different characteristic bands for the samples of poorly mixed samples (classified) to be tested Poor sample ratio characteristic band (profile) 10% 5% 1% Bad samples of mixed coke black particles Visible light + near infrared region 479-644 +1353-1450+1846-1998 nm 70 40 23.3 Near infrared region 1353-1450+1846-1998 nm 63.3 33.3 10 1353-1450 nm 40 10 3.3 1846-1998 nm 60 33.3 6.6 Visible area 479-644nm 26.7 6.7 6.7 Bad samples of mixed coke black lumps Visible light + near infrared region 479-644 +1353-1450+1846-1998 nm 93.3 93.3 80 Near infrared region 1353-1450+1846-1998 nm 93.3 80 80 1353-1450 nm 86.7 56.7 13.3 1846-1998 nm 93.3 80 80 Visible area 479-644nm 23.3 36.7 3.3 *Unit is percentage, n=30

It can be seen from Table 2 that three representative characteristic bands (profiles) of 479-644 nm, 1353-1450 nm and 1846-1998 nm are used at the same time, or two near infrared regions of 1353-1450 nm and 1846-1998 nm are combined/used separately All the bands can correctly identify test samples containing more than 20% of bad samples. If the content of bad samples is low (10%, 5%, 1%), the highest discrimination is achieved by using three representative characteristic bands at the same time. The bad samples of focused black mass can achieve a discrimination rate of more than 80%, and the correct judgment It is found to be a bad sample, which proves that the characteristic band selected in this disclosure has the identification effect. As for the poor samples of pyro black particles, although this method has poor discrimination in its low mixing ratio, due to the small size of the pyro black particles, samples with a small amount of pyro black particles may still be sold as normal products when they are actually shipped. The method can still effectively and quickly determine bad samples that cannot be shipped. Therefore, the present disclosure provides a meat quality inspection method using near-infrared spectroscopy (and some visible light spectroscopy), which uses a lower-cost and fast-testing near-infrared spectrometer, and statistical methods to inspect specific spectral characteristic bands (profiles) , You can quickly identify roasted meat of poor quality. Compared with the traditional artificial naked eye inspection, this method can be combined with an automated production line to achieve Industry 4.0.

Qualitative discussion and evaluation of characteristic bands (profiles):

This disclosure selects three representative characteristic bands of 479-644 nm, 1353-1450 nm, and 1846-1998 nm. Among them, visible light 479-644 nm can mainly show the red to blue chromaticity of the sample, which can correspond to Table 1. The color difference meter detects the a value and the b value in the numerical value. Near-infrared light 1353-1450 nm covers the second octave of CONH ₂ , CH and OH, and is used to indicate organics such as protein, fat and moisture; near-infrared light 1846-1998 nm covers the first octave of CONH ₂ and OH, It is used to indicate organic substances such as protein and moisture. The above two near-infrared light bands cover the absorption range of CONH ₂ , indicating that protein should be an important changing substance. The near infrared light 1846-1998 nm band covers the absorption of the -OH functional group, showing that the moisture content in the test sample is one of the influencing factors. Due to the humid climate in Taiwan, the unsealed meat products such as meat floss can easily absorb the moisture in the air. Therefore, the samples that have not been sealed and stored and the moisture content changes are not suitable for this band identification.

To confirm that the above characteristic bands (profiles) are related to protein, a variety of pork pine samples in Table 1 were scanned with a mid-infrared spectrometer. The results showed that the normal and bad samples of pork pine were at the wave number (2917, 2850 cm ^-1 ), (1622 , 1535 cm ^-1 ) and (1046, 990 cm ^-1 ) have differences in absorption values. Convert the absorption values of the above three sets of wavenumber spectra to proportional values, and add the results of mixing the spectral data of different proportions of defective products to confirm the key discrimination parameters. The results are shown in Figure 4. Figure 4 shows that each sample (standard color, upper color limit, lower color limit and those with different proportions of defective products) is first subjected to mid-infrared light detection to obtain the original spectrum, and from the spectrum data to obtain 2917, 2850, 1622, 1535, 1046, The absorption value of 990 cm ^-1 wave number, and then convert the value into a proportional value. Wherein the value of the absorbent 2917 cm ^-1 / 2850 cm ^-1 absorption of a value of 4 (A) to FIG, cm ^-1 absorption value of 1622/1535 cm ^-1 absorption of a value of FIG. 4 (B) in FIG, 1046 cm ^-1 absorption value The absorption value of /990 cm ^-1 is shown in Figure 4(C), which shows that the group 1620/1535 cm ^-1 has the best discriminating ability. The ratio value has a positive correlation with the proportion of defective products, which is significantly different from the normal sample. 1622 and 1535 cm ^-1 respectively represent C=O and NH or NC (=O) bonds, which are the main bonds in protein structure, so it can be confirmed that the spectral changes of near infrared light mainly come from proteins.

Although this disclosure has been described above using embodiments, the above embodiments are only examples and are not intended to limit this disclosure. Those with ordinary knowledge in this field should be able to make reasonable changes and adjustments based on the content of the above disclosure, so the scope of protection of this disclosure shall be subject to the scope of the patent application attached later.

Figure 1 is the appearance photos of the test samples (roasted pork floss); Figure 2 is the original infrared spectrum of the meat samples of Table 1 for near infrared detection; Figure 3 is the meat samples of Table 1 Near-infrared light detection after data pre-processing (standard normal variables and quadratic differential) spectrum chart; Figure 4 is a comparison of the specific band discrimination ability of the mid-infrared spectrum chart.

Claims (12)

A meat quality detection method using near infrared spectroscopy includes the following steps: (a) obtaining a plurality of normal meat samples and a meat sample to be tested; (b) scanning the normal meat samples with a near infrared spectrometer and collecting the Atlases of some normal meat samples; (c) Analyze the atlases collected in step (b) to obtain a normal meat feature profile, where the normal meat feature profile includes wavelengths in the range of 479-644nm, 1353-1450nm and/ Or 1846-1998nm band spectrum; (d) processing the meat sample to be tested in steps (b) and (c) to obtain a characteristic contour of the meat to be tested; and (e) comparing the characteristic contour of the meat to be tested Compared with the characteristic contour of the normal meat product, a comparison method is to obtain a normal threshold range of the characteristic contour of the normal sample, and then compare whether the characteristic contour of the sample to be tested falls within the threshold range. The method as described in item 1 of the patent application scope, wherein the method for analyzing the maps in step (c) includes a standard normal variable (Standard Normal Variate, SNV) and a second derivative. The method as described in item 1 of the patent application scope, wherein the normal meat samples and the meat samples to be tested in step (a) are roasted meat. The method as described in item 3 of the patent application scope, wherein the roasted meat is meat floss or meat crisp. The method as described in item 1 of the patent application scope, wherein in step (b), the average scanning times of the normal meat samples are 25-30 times. The method as described in item 1 of the patent application scope, wherein in step (d), the average number of scans of the meat sample to be tested is 25-30 times. According to the method described in item 1 of the patent application scope, if the characteristic contour of the meat to be tested does not fall within the normal threshold range, the meat sample to be tested is bad meat. The method as described in item 7 of the patent application scope, wherein the unhealthy meat includes burnt black meat and/or burnt black meat. According to the method described in item 1 of the patent application scope, if the characteristic contour of the meat to be tested falls within the normal threshold value range, the meat sample to be tested is normal meat. The method as described in item 1 of the patent application scope, wherein the normal threshold range and the threshold are calculated by the qualitative analysis function of VISION software. The method as described in item 1 of the patent application scope, wherein in step (a), the color of the meat samples is used as a reference to determine whether it is a normal meat sample. A meat quality detection method using near-infrared spectroscopy includes the following steps: scanning a meat to be tested with a near-infrared spectrometer to obtain a map of the meat to be tested; analyzing the map to obtain a characteristic contour, wherein the characteristic contour includes Atlases in the wavelength range of 479-644nm, 1353-1450nm and/or 1846-1998nm; and comparing the characteristic profile with the characteristic profile of a preset normal meat by comparing the characteristic profile of the preset normal meat A normal threshold range of the feature contour, and then compare whether the feature contour falls within the threshold range.

Images