Non-Destructive Detection of Egg Freshness Based on Raman Spectroscopy

Abstract

Disclosed is a method for non-destructive detection of egg freshness based on Raman spectroscopy technology, which belongs to the field of food detection. Partial least squares regression models are built using Raman spectroscopic data and measured values of physicochemical indexes for egg freshness, which can be used to predict egg freshness based on Raman spectrum of egg shell surface. The Raman spectroscopic data are collected in the waveband of 100-3000 cm.sup.−1. The physicochemical indexes used in the invention include the Haugh unit, the albumen pH, the air chamber diameter and the air chamber height. By using the partial least squares model, values of physicochemical index for egg freshness can be obtained from Raman spectra collected on egg shell surfaces, thus achieving the goal of non-destructive detection of egg freshness.

Claims

1. A method for non-destructive detection of egg freshness, comprising: (1) acquiring Raman spectral data on surfaces of sample eggs; (2) measuring values of a physicochemical index for egg freshness of the sample eggs; (3) building a partial least squares regression (PLSR) model using the acquired Raman spectral data and the measured values of the physicochemical index; and (4) using the PLSR model and a Raman spectrum of an egg to obtain a value of the physicochemical index for the egg without breaking the egg.

2. The method of claim 1, wherein the physicochemical index for egg freshness is Haugh unit, albumen pH, air chamber diameter or air chamber height.

3. The method of claim 1, wherein a position for acquiring Raman spectral data is egg top, egg bottom or egg waist.

4. The method of claim 3, wherein the Raman spectral data used for building the PLSR model are Raman spectral data acquired at a single position.

5. The method of claim 3, wherein the Raman spectral data used for building the PLSR model are average data of Raman spectral data acquired at the three positions.

6. The method of claim 1, further comprising pretreating the Raman spectral data before building the PLSR model.

7. The method of claim 6, wherein the pretreatment method is one or more methods selected from the group consisting of Savitzky-Golay smoothing, normalization, first derivative, second derivative, baseline correction, standard normal variable transformation, multiplicative scatter correction and denoise.

8. The method of claim 7, wherein the pretreatment method is first derivative when building a PLSR model using the Raman spectral data and values of air chamber diameter or air chamber height.

9. The method of claim 7, wherein the pretreatment method is second derivative when building a PLSR model using the Raman spectral data and values of Haugh unit or albumen pH.

10. The method of claim 1, wherein parameters for acquiring the Raman spectral data are as follows: excitation wavelength is 785 nm, acquisition waveband is 100-3000 cm.sup.−1, acquisition time is 5 s, the acquisition is performed for 3 times, and distance from a probe to the egg surface is 6 mm.

Description

BRIEF DESCRIPTION OF DRAWINGS

[0032] FIG. 1 is a flow diagram of a method for non-destructive detection of egg freshness based on Raman spectroscopy.

[0033] FIG. 2 is a schematic diagram of an overall average Raman spectrum of an egg sample.

DETAILED DESCRIPTION

[0034] Specific embodiments of the invented method are described in detail with regard to the drawings and the Examples. The specific embodiments are presented here for illustrative purposes only, and are not meant to limit the scope of the invention, which is defined by the claims as presented. Various changes and modifications to the disclosed methods known to those skilled in the art shall be covered within the scope of protection.

[0035] Haugh unit: a measure of egg freshness that is calculated according to the formula, HU=100×1 g(H+7.57−1.7×G.sup.0.37), wherein HU is the Haugh unit; H is the egg albumen height, in mm; and G is the mass of a whole egg, in gram.

[0036] Measurement of Albumen pH: an egg albumen pH value is measured by a pH meter after the egg albumen is separated into a flask and is uniformly stirred by a glass rod.

[0037] Measurement of Air chamber diameter and air chamber height: the air chamber diameter and the air chamber height are measured by a vernier caliper from the inside of an egg after the egg is broken and a complete air chamber is remained; and the unit of the air chamber diameter and the air chamber height is mm.

EXAMPLE 1

Building of Raman Spectrum Prediction Models

[0038] (1) Egg samples: Firstly, a batch of freshest eggs was bought. The eggs selected in the present example was HY-LINE brown eggs, the chicken age was 200 days, the eggs were stored in a 20° C. and 40% RH constant-temperature and constant-humidity incubator, and a portion of the eggs were taken out for data acquisition every 3 days.

[0039] (2) Acquisition of Raman spectrum of each sample: Raman spectral data of the top, the bottom and the waist line of the sample was acquired by a portable Raman spectrometer, two points were selected at each position, and each point was repeatedly tested for 3 times. Raman detection parameters were as follows: the integral time was 5 s, scanning was performed for 3 times, the acquisition waveband was the full waveband of 100-3000 cm−1, and the distance from a probe to the egg shell surface was 6 mm. Corresponding Raman spectral data was measured.

[0040] (3) Detection of the reference value of each physicochemical index of each sample: The Haugh unit of the eggs was detected by an egg product analyzer. The albumen pH was measured by a pH meter. The air chamber diameter and the air chamber height were measured by a vernier caliper. Reference values of the Haugh unit, the egg albumen pH, the air chamber diameter and the air chamber height were respectively acquired.

[0041] (4) Data modeling analysis: By combining with chemometrics, the measured Raman spectrum intensity data of the eggs in the full waveband of 100-3000 cm.sup.−1 was utilized to build partial least squares regression (PLSR) models, i.e., Raman spectrum prediction models of egg freshness physicochemical indexes, with the reference value of each of the physicochemical indexes respectively.

[0042] The Raman spectrum prediction models were partial least squares regression models built by using the partial least squares regression method. 80% of samples in test samples were used as a correction set, and 20% of samples were used as a prediction set (where the correction set was samples for building a model structure and parameters, and the prediction set was samples for evaluating the model robustness and prediction performance). It should be ensured that the data range of the prediction set are within the data range of the correction set, and additionally, the difference between average values of the correction set and the prediction set was not too big. Different-set data of the example was shown in Table 1.

TABLE-US-00001 TABLE 1 Different-set statistical data of each physicochemical index Mini- Maxi- Standard Sample mum mum Average deviation Index Sample quantity value value value SD Haugh Correction 100 42.7 82.7 62.2 7.1 unit set Prediction 25 49.4 79.4 61.5 7.4 set Egg Correction 100 8.78 9.45 9.26 0.15 albumen set pH Prediction 25 8.81 9.45 9.27 0.15 set Air Correction 100 15.47 44.91 32.01 5.82 chamber set diameter Prediction 25 18.20 43.17 30.20 5.21 set Air Correction 100 2.19 22.12 8.96 4.28 chamber set height Prediction 25 2.22 18.74 9.05 4.17 set

[0043] In this example, an average value of the Raman spectra acquired from all positions of the egg was used as an overall average spectrum to build full waveband PLSR models with each of four physicochemical indexes. Results were shown in Table 2.

TABLE-US-00002 TABLE 2 Modeling results of original Raman spectrum and each physicochemical index Main component Index number Rc RMSEC Rp RMSEP Haugh unit 6 0.664 5.326 0.520 6.541 Egg albumen pH 6 0.832 0.083 0.846 0.081 Air chamber 6 0.767 3.734 0.822 3.629 diameter Air chamber 3 0.630 3.324 0.822 2.494 height

[0044] In Table 2, Rc represents the correlation coefficient of the correction set; RMSEC represents the root-mean-square error of the correction set; Rp represents the correlation coefficient of the prediction set; and RMSEP represents the root-mean-square error of the prediction set.

[0045] From Table 2, it can be seen that the correlation coefficient of the model built by the unpretreated average Raman spectrum and the egg albumen pH value could reach 0.6 or higher, and could reach 0.8 or higher under better conditions. This shows that the model had good prediction performance. Additionally, the correlation coefficients of the models built with other three physicochemical indexes were all 0.6 or higher. If the spectrum is pretreated, it is possible to obtain a model with more stable and better prediction performance.

EXAMPLE 2

Influence of Different Raman Spectrum Pretreatments on Modeling

[0046] Referring to Example 1, the average value of all Raman spectra acquired from three detection positions was used as the overall average Raman spectrum, and the overall average Raman spectrum was respectively subjected to the following pretreatment: Savitzky-Golay smoothing (SG), normalization (NL), first derivative (1.sup.st Der), second derivative (2.sup.nd Der), baseline correction (BL), standard normal variable transformation (SNV), multiplicative scatter correction (MSC) and denoise (Denoise). Taking the Haugh unit value as an example, full waveband PLSR modeling results were as shown in Table 3.

TABLE-US-00003 TABLE 3 Influence of different Raman spectrum pretreatment methods on model performance Main Pretreatment component mode number Rc RMSEC Rp RMSEP Raw 6 0.664 5.326 0.520 6.541 SG 6 0.632 5.524 0.504 6.495 NL 6 0.646 5.439 0.659 5.714 1.sup.st Der 5 0.874 3.463 0.790 4.566 2.sup.nd Der 6 0.908 2.985 0.842 4.026 BL 6 0.648 5.427 0.517 6.521 SNV 6 0.491 5.085 0.629 5.899 MSC 6 0.699 5.097 0.630 5.890 Denoise 6 0.925 2.708 0.325 7.363

[0047] Further, the effects of the pretreatment methods was determined on models between the overall average Raman spectrum and each physicochemical index. Using only one pretreatment method, an optimal Raman spectrum pretreatment method of each physicochemical index model was obtained. For the Haugh unit and albumen pH value modeling, the optimal Raman spectrum pretreatment method was 2.sup.nd Der, and for the air chamber diameter and the air chamber height modeling, the optimal pretreatment method was 1.sup.st Der. If more than one of the pretreatment methods are combined, the modeling performance may be better.

EXAMPLE 3

Influence of Different Raman Measuring Positions on Modeling

[0048] This example tested the effect of Raman spectrum acquisition position on the modeling performance. The Raman spectrum acquisition positions were selected to be the egg top, the egg bottom and the egg waist. The egg top refers to a small end position of the egg. The egg bottom refers to a big end position of the egg. The egg waist refers to a middle line position of the egg. Two spots were selected for each acquisition position, and three Raman spectra were acquired at each spot.

[0049] The average value of the Raman spectra acquired from each acquisition position are used as representative Raman spectrum of the respective position. The representative Raman spectrum of the three acquisition positions were respectively pretreated by the optimal pretreatment method as determined in Example 2. Results of full waveband PLSR models built with the four physicochemical indexes at different positions were as shown in Table 4.

TABLE-US-00004 TABLE 4 Performance of PLSR Models of representative Raman spectra at different positions Pretreatment Index method Top Bottom Waist Average Haugh unit 2.sup.nd Der Rc 0.944 0.934 0.930 0.908 Rp 0.925 0.751 0.794 0.842 Egg albumen 2.sup.nd Der Rc 0.945 0.936 0.936 0.940 pH Rp 0.935 0.948 0.927 0.941 Air chamber 1.sup.st Der Rc 0.903 0.854 0.870 0.872 diameter Rp 0.915 0.914 0.882 0.902 Air chamber 1.sup.st Der Rc 0.828 0.764 0.815 0.817 height Rp 0.830 0.829 0.834 0.826

[0050] The best testing methods for the HY-LINE brown eggs of 200-day-old chicken, includes: the Raman spectrum was acquired from the egg top; the Raman spectrum was pretreated by 2.sup.nd Der or 1.sup.st Der method; full waveband PLSR models of 100-3000 cm.sup.−1 were built using the pretreated Raman spectral data and measured values of the egg freshness physicochemical indexes. As shown from Table 4, the PLSR models achieved good predictive performance. For the model of the Haugh unit value, Rc was 0.944 and Rp was 0.925. For the model of the albumen pH value, Rc was 0.945 and Rp was 0.935. For the model of the air chamber diameter, Rc was 0.903 and Rp was 0.915. For the model of the air chamber height, Rc was 0.828 and Rp was 0.830.

COMPARATIVE EXAMPLE 1

[0051] When performing non-destructive measurement of other egg freshness indexes (such as yolk index) of the HY-LINE brown eggs or other types of eggs based on Raman spectroscopy, one can refer to the methods and procedures provided herein.

[0052] Referring to operation steps of Example 1, Raman spectral data of egg samples was collected through steps (1)-(2), and the measured values of the physicochemical indexes in the step (3) were replaced with values of the yolk index.

[0053] The process of measuring the yolk index was as follows: an egg was broken onto a plane; and the height and the diameter of the yolk were measured by a vernier caliper. The yolk index was calculated by the formula, yolk index=yolk height/yolk diameter (the unit of the yolk height and the yolk diameter is mm).

[0054] (4) Data modeling analysis: the collected Raman spectral data of the egg in the full waveband of 100-3000 cm.sup.−1 was utilized to build partial least squares regression models with the measured values of yolk index.

[0055] Before any pretreatment, the Rc for the PLSR model between the average Raman spectrum and yolk index only reached 0.366. After the spectrum pretreatment, the Rc for the best PLSR model between the average Raman spectrum and yolk index reached 0.569. It thus showed that PLSR model based on the Raman spectrum is not suitable for modeling the yolk index.