A non-destructive rapid detection method for fish meat texture of different freshness

Abstract

The present invention is applicable to the technical field of nondestructive detection of food quality and safety, and provides a nondestructive rapid detection method for the texture of fish meat with different freshness, comprising the following steps: preparing fish meat samples with different freshness, so that the modeled fish meat samples contain primary, secondary and tertiary freshness; using a texture analyzer to test the elasticity, hardness and chewiness indexes of fish meat with different freshness; using near infrared and Raman spectroscopy to collect near infrared spectrum and Raman spectrum information of fish meat respectively; using CARS and SiPLS methods to extract features from the spectral information of fish meat, and then using the OSC method for preprocessing on this basis, performing intermediate data fusion on the processed spectra, and establishing PLSR prediction models for elasticity, hardness and chewiness respectively; finally, using the established prediction model to test the texture characteristics of fish meat with unknown freshness. The detection method of the present invention improves the accuracy of nondestructive detection of fish meat texture, and is of great significance for the rapid evaluation of fish meat quality.

Description

Technical Field

[0001] The invention relates to the technical field of non-destructive food testing, and in particular to a non-destructive rapid testing method for fish meat textures of different freshness. Background Art

[0002] Fish is rich in high-quality protein, vitamins, unsaturated fatty acids and other nutrients, and is an important food source for humans. However, due to its rich nutrition and water content, fish is very prone to spoilage during storage, which not only affects the economic value of fish, but also affects its taste and even causes food safety problems. The texture of fish is one of the key indicators for evaluating the freshness and quality of fish during storage, and its changes will directly affect its market value and consumers' eating experience.

[0003] Food texture is defined as "all rheological and structural (geometric and surface) properties of a product that can be perceived by mechanical, tactile, and appropriate visual and auditory receptors". The texture of fish meat (including hardness, elasticity, chewiness, etc.) is an important attribute for evaluating the taste and value of fish products. At the same time, texture is also a universal characteristic related to the freshness of fish products. Texture analysis can provide information about the physical and chemical changes that occur in fish meat during storage, such as protein denaturation, water loss, and microbial growth. Related literature research also shows that the texture parameters (elasticity, cohesion) are highly correlated with quality parameters (TVB-N), so the texture test can better reflect the freshness of aquatic products. Therefore, for the fish industry, it is of great significance to test the texture of fish meat. The texture quality characteristics of food can be evaluated by descriptive sensory evaluation (subjective) or instrumental analysis (objective). Descriptive sensory evaluation requires a team trained in sensory evaluation methods, which often leads to long time, cumbersome procedures and low repeatability. Compared with subjective evaluation, objective instrument-based measurements are less affected by human factors, reducing the differences between measurement results. Total texture analysis (TPA) has the advantages of high reliability and high precision, and is widely used to analyze food texture. However, TPA relies on the use of a texture analyzer, which results in time-consuming operation, contact with samples, and is destructive to samples. It also requires professional equipment and skilled operators, and cannot achieve on-site rapid testing of sample texture.

[0004] Raman and near-infrared spectroscopy are molecular spectroscopy techniques that can collect chemical information of samples without destroying them, and have been used for non-destructive testing of food chemical indicators. However, Raman and near-infrared spectroscopy have certain limitations for non-destructive testing of physical properties of food, such as hardness, elasticity, shear force, etc. This is mainly reflected in the difficulty of obtaining comprehensive information about sample properties using a spectral technology, resulting in poor detection accuracy. In addition, due to the complexity of spectral information and high noise, how to pre-process the spectrum and mine texture-related features from a large amount of spectral data is the key to improving detection accuracy. Therefore, due to the heterogeneity of samples and the complexity of processing methods, the accurate and objective evaluation of fish texture has certain challenges and is difficult to meet the needs of the modern food industry for fast, accurate, and non-destructive detection methods. Summary of the invention

[0005] In order to make up for the deficiencies of the prior art, the present invention provides a non-destructive rapid detection method for the texture of fish meat with different freshness.

[0006] The present invention is achieved through the following technical solution: Based on the limitations of spectral rapid non-destructive detection technology for fish meat texture detection, the present invention provides a non-destructive rapid detection method for fish meat texture of different freshness, aiming to solve the problem of poor performance of existing non-destructive detection of fish meat texture.

[0007] The present invention is achieved by a non-destructive rapid detection method for fish meat texture of different freshness, characterized in that it comprises the following steps:

[0008] S1: Remove the head, scales and internal organs of fresh fish, take the fish meat from the back, cut it into fish fillets of about 30 mm × 30 mm × 10 mm, put the fish fillets into plastic ziplock bags, store them at different temperatures for a certain period of time, and obtain fish meat samples of different freshness;

[0009] S2: Use a physical property analyzer to measure the texture of the fish fillets in S1, using the full texture analysis mode to collect the elasticity, hardness, and chewiness parameters of the fish fillets;

[0010] S3: collect near infrared spectra and Raman spectra of the fish fillets in S2, collect multiple spectra from different parts of the fish fillets, and take the average value;

[0011] S4: The collected near-infrared spectra and Raman spectra are first normalized, and then the spectral features are extracted according to different texture indicators. The extracted spectral features are then preprocessed, and finally the near-infrared and Raman spectral data are fused at an intermediate level to obtain a near-infrared-Raman spectral intermediate fusion data set;

[0012] S5: dividing the near infrared-Raman spectroscopy fusion data set in step S4 into a training set and a prediction set, and establishing partial least squares regression (PLSR) prediction models for elasticity, hardness, and chewiness, respectively;

[0013] S6: Use the PLSR model established in S5 to predict the texture properties of unknown fish fillets.

[0014] Preferably, for the preparation of fish samples of different freshness in step S1, the amount of fish samples is greater than 100, and the TVB-N content of the fish is tested to ensure that the fish fillets of different freshness cover the first level of freshness, the second level of freshness and the third level of freshness, and the storage time end point of the fish fillets is set until a distinct odor can be smelled;

[0015] Preferably, the determination conditions of the overall texture of fish meat in step S2 are as follows: a TA / 5S spherical probe is used, the range of the force sensor is 100 N, the probe rises to a surface height of 15 mm, the deformation is 50%, the detection speed is 120 mm / min, the starting force is 0.15 N, the detection interval is 2 s, and each sample fillet is measured 3 times to take the average value;

[0016] Preferably, in step S3, the wavelength range of the near infrared spectrum is 908-1676 nm, each fish sample is scanned 20 times, and the average value is taken. The test laser of the Raman spectrum is 785 nm, the test time is 5-10 s, and the spectral range is 200-3200 cm -1 , three test points were selected for each fish sample, and the average value of the tested Raman spectra was taken. The samples collected by near-infrared and Raman spectra corresponded one by one to the samples tested for texture.

[0017] Preferably, in step S4, the acquired near infrared and Raman spectral data are subjected to mean normalization processing. Specifically, in the mean normalization process, the value of each spectral data point is subtracted from the average value of the data set, and then divided by the range of the data set value (maximum value minus minimum value). The near infrared and Raman spectral peaks are limited to between 0 and 1, and the normalized near infrared and Raman spectra are subjected to intermediate data fusion, that is, the near infrared and Raman spectral characteristic bands obtained by feature extraction are directly merged to obtain a near infrared-Raman spectral intermediate fusion data set.

[0018] Preferably, the spectral feature extraction method in step S4 is a competitive adaptive reweighting algorithm (CARS), a collaborative interval partial least squares method (SiPLS), etc., and the order of spectral processing is to first extract features and then preprocess the spectrum.

[0019] For CARS, the basis is to dynamically adjust the weights of features and recursively select them to achieve the goal of the optimal feature set. First, the initial weights of all features are set equal. In each iteration i, the number of features selected is r i Determined by the exponential decay formula:

[0020]

[0021] Among them, a and k are constants calculated based on the number of iterations n and the total number of features D. After each round of iteration, the feature weights are updated according to the coefficients of the PLSR model. Finally, cross-validation is used to calculate the root mean square error (RMSE) of the model after each iteration, and the feature with the smallest RMSE is selected as the optimal feature set.

[0022] For SiPLS, partial least squares regression (PLS) is applied to the segmented spectral intervals to accurately evaluate the information contribution of each interval. First, the input spectral data matrix is ​​segmented according to a fixed interval length, and each interval contains the characteristics of 5 continuous wavelengths. PLS is performed independently for each spectral interval, and the PLS model is set to 2 components. Then, the mean square error (MSE) of the PLS model for each interval is calculated using 10-fold cross validation. The score is the negative mean of the mean square error obtained by cross validation. Finally, an array containing the MSE score of each interval and the starting wavelength of the corresponding interval is returned.

[0023] Preferably, for the prediction of fish meat elasticity and chewiness, the spectral feature extraction method is CARS to extract near-infrared spectrum and Raman spectrum, and for the prediction of fish meat hardness, the spectral feature extraction method is SiPLS to extract near-infrared spectrum and CARS to extract Raman spectrum; then the extracted spectral features are preprocessed with orthogonal scattering correction (OSC), which uses nonlinear iterative partial least squares to iteratively decompose orthogonal signals, and the maximum number of iterations is 100.

[0024] Preferably, the fish meat sample in step S1 is large yellow croaker, seabass, black carp, grass carp, black carp, turbot, silver carp, bighead carp, etc.

[0025] Due to the adoption of the above technical solution, the present invention has the following beneficial effects compared with the prior art:

[0026] (1) The present invention realizes non-contact detection of fish meat texture by performing spectral scanning on unknown fish meat samples without destroying the fish meat samples. The method has the advantages of being rapid, non-destructive, pollution-free, and easy to operate, and provides a good basis for realizing rapid evaluation of fish meat quality.

[0027] (2) This method uses two molecular spectroscopy methods, Raman and near-infrared, and combines the advantages of different spectral technologies to complement each other's advantages and disadvantages, thus achieving an organic integration of different spectral technologies and significantly improving the accuracy of the texture prediction model.

[0028] (3) This method extracts the characteristic bands of near-infrared and Raman spectra of different texture indicators and combines them with a machine learning model to achieve simultaneous detection of the elasticity, hardness, and chewiness of fish meat through the same spectral information of fish meat, thereby improving the efficiency of texture detection.

[0029] Additional aspects and advantages of the present invention will become apparent from the following description or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0031] Figure 1 : Texture change trends of large yellow croaker stored at 4℃ for different days, (a) elasticity, (b) chewiness, (c) hardness;

[0032] Figure 2 :(a) Near infrared and (b) Raman spectra of large yellow croaker stored at 4℃ for different days;

[0033] Figure 3 : PLSR prediction model of texture of large yellow croaker based on near infrared and Raman spectroscopy data fusion, (a) elasticity, (b) chewiness, (c) hardness;

[0034] Figure 4 :Near infrared spectroscopy data of CARS feature selection, (a) elasticity, (b) chewiness, (c) hardness;

[0035] Figure 5 : Near infrared spectroscopy data of SiPLS feature selection, (a) elasticity, (b) chewiness, (c) hardness, (df) is the corresponding MSE score; DETAILED DESCRIPTION

[0036] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0037] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0038] Combine the following Figures 1 to 3 The non-destructive rapid detection method of fish meat textures of different freshness according to the embodiment of the present invention is specifically described.

[0039] Embodiment 1:

[0040] This embodiment provides a non-destructive rapid detection method for fish meat texture of different freshness, comprising the following steps:

[0041] (1) Preparation of fish meat samples of different freshness: The head, scales and internal organs of fresh large yellow croaker were removed, and the fish meat from the back was cut into fish fillets of approximately 30 mm × 30 mm × 10 mm in size to obtain 120 large yellow croaker fillets. The fillets were placed in plastic ziplock bags and stored at 4°C for up to 14 days to obtain large yellow croaker meat samples covering the first, second and third levels of freshness.

[0042] (2) Determination of fish meat texture: The texture of fish fillets stored at different times was determined using the TMS-TOUCH physical property analyzer. The full texture analysis mode was used to collect the elasticity, hardness, and chewiness parameters of the fish fillets. The test used a TA / 5S spherical probe, the force sensor range was 100 N, the probe rose to a height of 15 mm from the surface, the deformation was 50%, the detection speed was 120 mm / min, the starting force was 0.15 N, and the detection interval was 2 s. Each sample fish fillet was measured three times and the average value was taken. The results are shown in the table below. Figure 1 As shown;

[0043] (3) Fish spectral testing: The near-infrared spectrum and Raman spectrum of the fish fillets after texture testing were collected in sequence. Multiple spectra were collected from different parts of the fish fillets. The wavelength collection range of the near-infrared spectrum was 908-1676 nm. Each fish sample was scanned 20 times. The test laser of the Raman spectrum was 785 nm, the test time was 10 s, and the spectral range was 200-3200 cm -1 , 3 test points were selected for each fish sample, and the near infrared and Raman spectra of each piece of fish were averaged. The samples collected by near infrared and Raman spectra corresponded to the samples tested for texture one by one. The results are shown in Figure 2 shown.

[0044] (4) Spectral preprocessing: The collected near-infrared spectra and Raman spectra were subjected to OSC preprocessing and compared with smoothing, first derivative (FD), standard normal variable (SNV), and mean normalization (Mean Nor) preprocessing.

[0045] (5) Model construction and evaluation: The near-infrared and Raman spectroscopy fusion data sets were divided into a training set: prediction set = 4:1 based on the five-fold cross-validation method. A partial least squares regression model for elasticity, hardness, and chewiness was established. 2 c ,RMSEC,R 2 cv ,RMSECV,R 2 p , RMSEP were used to evaluate and verify the model, see Table 1 and Table 2.

[0046] Table 1 Comparison of texture test results of large yellow croaker based on near infrared spectroscopy combined with different pretreatment methods

[0047]

[0048] Table 2 Comparison of texture test results of large yellow croaker based on Raman spectroscopy combined with different pretreatment methods

[0049]

[0050] As can be seen from Tables 1 and 2, the prediction performance of near infrared spectroscopy and Raman spectroscopy without pretreatment for the elasticity, chewiness and hardness of large yellow croaker is poor. In comparison, the prediction effect of near infrared spectroscopy is better than that of Raman spectroscopy. After OSC pretreatment, the prediction performance is significantly improved, R 2 p >0.8, indicating that OSC pretreatment can effectively reduce irrelevant background noise or interference while enhancing relevant chemical signals. However, the prediction performance of Raman spectra after pretreatment is still poor, R 2 p <0.5, which may be attributed to the fact that Raman spectra contain less information in the samples of this study, or that the pretreatment technology fails to fully highlight the spectral features related to the texture of large yellow croaker. This indicates that the texture prediction effect of large yellow croaker using a single near-infrared or Raman spectrum is poor.

[0051] Embodiment 2:

[0052] The present embodiment provides a non-destructive rapid detection method for the texture of fish meat with different freshness, wherein the steps of preparing fish meat samples with different freshness, determining the texture of fish meat, and testing the fish meat spectrum are the same as those in Example 1, except that in this embodiment, CARS and SiPLS feature extraction are performed on the collected Raman spectra, and compared with the feature extraction of the continuous projection algorithm (SPA), and then a PLSR prediction model for the texture of large yellow croaker is established. The results are shown in Tables 3 and 4.

[0053] Table 3 Comparison of texture test results of large yellow croaker based on near infrared spectroscopy combined with different feature extraction methods

[0054]

[0055] Table 4 Comparison of texture test results of large yellow croaker based on Raman spectroscopy combined with different feature extraction methods

[0056]

[0057] As shown in Tables 3 and 4, the CARS method performs particularly well in processing Raman spectroscopy data, especially in predicting elasticity, chewiness, and hardness, where the R 2 P The values ​​all exceed 0.75. This achievement is attributed to the fact that CARS can effectively screen out the characteristic displacements that contribute significantly to the predicted target, while significantly reducing noise and improving the signal-to-noise ratio.

[0058] from Figure 4 and Figure 5 It can be seen that for NIR spectral data, the CARS method improved the prediction effect by selecting a small number of key wavelengths (13 for elasticity, 19 for chewiness, and 79 for hardness). In contrast, the SiPLS method achieved better results in hardness prediction by selecting wavelengths with low error intervals (45 characteristic wavelengths, MSE < 0.65), but the results in the prediction of elasticity and chewiness were relatively mediocre, despite selecting more wavelengths (75). This shows that in NIR spectroscopy analysis, CARS provides higher feature selection accuracy, while SiPLS is more suitable for finding broad correlations in the data.

[0059] In summary, CARS and SiPLS each have their own unique advantages and application scenarios. CARS is particularly applicable in Raman spectroscopy analysis and can effectively improve prediction accuracy and reliability; while SiPLS performs better when selecting near-infrared spectral features under specific conditions.

[0060] Embodiment 3:

[0061] The present embodiment provides a non-destructive rapid detection method for the texture of fish meat with different freshness, wherein the steps of preparing fish meat samples with different freshness, determining the texture of fish meat, and testing the fish meat spectrum are the same as those in Example 1, except that in the present embodiment, CARS or SiPLS feature extraction is performed on the collected near-infrared and Raman spectra, and then OSC preprocessing is performed, and the spectra after feature extraction and preprocessing are subjected to intermediate data fusion to establish a PLSR prediction model for the texture of large yellow croaker. The results are shown in Table 5.

[0062] Table 5 Comparison of the results of PLSR prediction model for texture of large yellow croaker based on near infrared and Raman spectroscopy fusion

[0063]

[0064] As shown in Table 5, the PLSR prediction model for the elasticity, chewiness and hardness of large yellow croaker adopts the fusion of near infrared and Raman spectroscopy data. After the spectra are preprocessed by CARS+CARS OSC, CARS+CARS OSC and SiPLS+CARS OSC, the prediction results of elasticity, chewiness and hardness of large yellow croaker are R 2 p >0.85, which is significantly higher than the prediction results of single near-infrared and Raman spectra, and also better than the model results after the fusion of the original near-infrared spectrum and Raman spectrum without extraction. Figure 3 It can be seen that this method has a good fitting effect on the elasticity, chewiness and hardness of large yellow croaker.

[0065] Based on the results of Example 1, Example 2 and Example 3, it can be seen that the near-infrared and Raman spectral data proposed in the present invention can achieve rapid non-destructive detection of fish meat texture with different freshness after spectral feature extraction, OSC spectral preprocessing and intermediate spectral data fusion, and the accuracy of the prediction results is significantly higher than the prediction results of traditional single near-infrared spectrum and Raman spectrum, which shows that the method of the present invention has good beneficial effects and significant creativity, and has a good application prospect in the actual application of fish meat texture with different freshness.

[0066] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the scope of protection of the invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on these embodiments, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in this field can still combine, add, delete or make other adjustments to the features in the various embodiments of the present invention according to the circumstances without conflict, without making creative work, so as to obtain different other technical solutions that do not deviate from the concept of the present invention in essence, and these technical solutions also belong to the scope of protection of the present invention.

Claims

1. A non-destructive rapid detection method for the texture of fish meat with different freshness, characterized in that: The steps include: S1: Remove the head, scales and internal organs of fresh fish, take the back fish meat, cut it into fish fillets of 30 mm × 30 mm × 10 mm, put the fish fillets into plastic ziplock bags, store them at different temperatures for a certain period of time, and obtain fish meat samples of different freshness; S2: The texture of the fish fillets in S1 was measured using a physical property analyzer, and the full texture analysis mode was adopted to collect the elasticity, hardness and chewiness parameters of the fish fillets; the conditions for the determination of the full texture of fish meat were: a TA / 5S spherical probe was used, the range of the force sensor was 100 N, the height of the probe rising to the surface was 15 mm, the deformation was 50%, the detection speed was 120 mm / min, the starting force was 0.15 N, the detection interval was 2 s, and each sample fillet was measured 3 times and the average value was taken; S3: collect near infrared spectra and Raman spectra of the fish fillets in S2, collect multiple spectra from different parts of the fish fillets, and take the average value; S4: The collected near-infrared spectrum and Raman spectrum are first normalized. In the mean normalization process, the value of each spectral data point is subtracted from the average value of the acquired near-infrared spectrum and Raman spectrum data, and then divided by the range of the acquired near-infrared spectrum and Raman spectrum data values, where the range is the maximum value minus the minimum value; the normalized near-infrared and Raman spectra are subjected to intermediate data fusion, that is, the characteristic bands of the near-infrared and Raman spectra obtained by feature extraction are directly merged to obtain a near-infrared-Raman spectrum intermediate fusion data set; Then, spectral features are extracted for different texture indicators, and the extracted spectral features are preprocessed. The spectral feature extraction methods are competitive adaptive reweighting algorithm CARS and collaborative interval partial least squares method SiPLS. The spectral processing order is feature extraction first, and then spectrum preprocessing. For CARS, the basis is to dynamically adjust the weights of features and recursively select them to achieve the goal of the optimal feature set; first, all features are initially weighted equally, and in each iteration i, the number of features selected is r i Determined by the exponential decay formula: Among them, a and k are constants calculated based on the number of iterations n and the total number of features D. After each round of iteration, the feature weights are updated according to the coefficients of the PLSR model. Finally, cross-validation is used to calculate the root mean square error RMSE of the model after each iteration, and the feature with the smallest RMSE is selected as the optimal feature set; For SiPLS, partial least squares regression (PLS) is applied to the segmented spectral intervals to accurately evaluate the information contribution of each interval. First, the input spectral data matrix is ​​segmented according to the fixed interval length, and each interval contains the characteristics of 5 continuous wavelengths. PLS is performed independently for each spectral interval, and the PLS model is set to 2 components. Then, the mean square error (MSE) of the PLS model for each interval is calculated using 10-fold cross validation. The score is the negative mean of the mean square error obtained by cross validation. Finally, an array containing the MSE score of each interval and the starting wavelength of the corresponding interval is returned. Finally, the near infrared and Raman spectroscopy data are fused at the intermediate level to obtain the near infrared-Raman spectroscopy intermediate fusion data set; S5: Divide the near infrared-Raman spectroscopy fusion data set in step S4 into a training set and a prediction set, and establish partial least squares regression PLSR prediction models for elasticity, hardness, and chewiness, respectively; for the prediction of fish elasticity and chewiness, the spectral feature extraction method is CARS to extract near infrared spectrum and Raman spectrum, and for the prediction of fish hardness, the spectral feature extraction method is SiPLS to extract near infrared spectrum and CARS to extract Raman spectrum; then, the extracted spectral features are preprocessed by orthogonal scatter correction OSC, and the orthogonal signal is decomposed and iterated using nonlinear iterative partial least squares in the process, and the maximum number of iterations is 100; S6: Use the PLSR model established in S5 to predict the texture properties of unknown fish fillets.

2. A non-destructive rapid detection method for fish meat texture of different freshness according to claim 1, characterized in that: For the preparation of fish samples of different freshness in step S1, the amount of fish samples is greater than 100. By testing the TVB-N content of the fish, it is ensured that the fish fillets of different freshness cover the first level of freshness, the second level of freshness and the third level of freshness. The storage time end point of the fish fillets is set until a noticeable odor can be smelled.

3. The non-destructive rapid detection method of fish meat texture of different freshness according to claim 1, characterized in that: In step S3, the wavelength range of the near infrared spectrum is 908-1676 nm. Each fish sample is scanned 20 times and the average value is taken. The test laser of the Raman spectrum is 785 nm, the test time is 5-10 s, and the spectral range is 200-3200 cm -1 , three test points were selected for each fish sample, and the average value of the tested Raman spectra was taken. The samples collected by near-infrared and Raman spectra corresponded one by one to the samples tested for texture.

4. The non-destructive rapid detection method of fish meat texture of different freshness according to claim 1, characterized in that: The fish samples in step S1 include large yellow croaker, seabass, black carp, grass carp, black carp, turbot, silver carp and bighead carp.

Images