Method for detecting tenderness and water-holding capacity of cold fresh meat of grazing sheep based on hyperspectral imaging

Abstract

The application discloses a grazing sheep chilled meat tenderness and water retention detection method based on hyperspectral imaging, relates to the field of nondestructive testing technology of agricultural and livestock products, and comprises the following steps: acquiring a hyperspectral image of grazing sheep chilled meat, establishing a calibration set in combination with shear force and water retention experimental data, extracting a region of interest and performing spectral pretreatment, adopting characteristic wavelength screening and full-waveband comparative analysis to construct a multi-index prediction model based on partial least squares regression, finally adopting a WT-PLSR model to predict and output tenderness, adopting an SG-PLSR model to predict and output cooking loss, and adopting a DT-PLSR model to predict and output drip loss, so that synchronous nondestructive quantitative detection of chilled meat tenderness and water retention indexes based on hyperspectral information is realized.

Description

Technical Field

[0001] This invention relates to the field of non-destructive testing technology for agricultural and livestock products, specifically a method for detecting the tenderness and water retention of chilled fresh meat from grazing sheep based on hyperspectral imaging. Background Technology

[0002] Free-range sheep mutton boasts a rich and natural flavor, and is marketed as chilled fresh meat. The tenderness and water retention of this chilled meat are key indicators for evaluating its edible quality and processing characteristics. Tenderness, primarily characterized by shear force, depends on the muscle fiber structure and connective tissue state, directly impacting the meat's texture and chewiness. Water retention is typically evaluated through cooking loss, drip loss, and centrifugal loss, reflecting the muscle tissue's ability to retain moisture.

[0003] Currently, the testing of mutton tenderness and water retention still mainly relies on traditional physicochemical methods. Tenderness is measured using a shear force meter, while water retention is measured through methods such as cooking loss and dripping water loss. In recent years, non-destructive testing methods that fuse image and spectral information using hyperspectral imaging technology have been widely applied in the field of non-destructive testing of agricultural and livestock product quality.

[0004] Considering that fat deposition in grazing sheep is significantly affected by pasture quality, seasonal changes, and supplementary feeding measures, resulting in considerable variability, meat tenderness and water retention fluctuate markedly. Furthermore, after slaughter, chilled meat undergoes thorough maturation, with moderate breakdown of muscle proteases and gradual softening of connective tissue, further altering meat quality. These two factors collectively contribute to changes in the spectral characteristics of chilled meat from grazing sheep. Simultaneously, the complex and highly heterogeneous structure of mutton makes its original spectrum susceptible to interference from noise, scattering, and redundant information. Therefore, directly applying general mutton tenderness and water retention prediction models based on hyperspectral imaging technology to detect chilled meat from grazing sheep yields low accuracy. Summary of the Invention

[0005] (a) Technical problems to be solved This invention provides a method for detecting the tenderness and water retention of chilled fresh meat from grazing sheep based on hyperspectral imaging, which can achieve simultaneous, non-destructive, quantitative detection of tenderness and water retention indicators of chilled fresh meat based on hyperspectral information.

[0006] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: a method for detecting the tenderness and water retention of chilled fresh meat from grazing sheep based on hyperspectral imaging, comprising the following steps: Fresh meat samples from grazing sheep were selected, and the tenderness and water retention of the same batch of samples were subjected to physicochemical tests. Among them, the shear force value was measured by a shear force meter to characterize the tenderness, and the water retention was quantified by cooking loss test and drip loss test. The chilled meat sample was subjected to non-destructive spectral acquisition to obtain a hyperspectral image containing spatial and continuous spectral information. In each of the hyperspectral images, a valid region of interest is selected. After removing the background and non-target tissue regions, the spectra of the pixels in each region of interest are extracted, and the spectra of the pixels in the region of interest are averaged to obtain the valid spectral data of the corresponding region. Different spectral preprocessing methods are used for the effective spectral data corresponding to the regions of interest for tenderness, cooking loss, and drip loss. Specifically, wavelet transform is used for the effective spectral data corresponding to tenderness; a smoothing algorithm is used for the effective spectral data corresponding to cooking loss; and derivative removal transform is used for the effective spectral data corresponding to drip loss. Based on the effective spectral data of each quality indicator after preprocessing, a continuous projection algorithm is introduced to screen the high-dimensional full-band spectral variables, extract the characteristic wavelength combinations that are highly correlated with each chilled fresh meat quality indicator, and construct the optimal prediction model for different quality indicators based on the partial least squares regression method. In the actual testing phase, hyperspectral images of the chilled fresh meat sample to be tested are acquired, and effective region of interest extraction, effective spectrum generation and corresponding spectrum preprocessing are performed in sequence. After extracting the characteristic wavelength variables, they are input into the optimal prediction model of the corresponding quality index, and the prediction output of the tenderness and water retention of chilled fresh meat is performed.

[0007] In some feasible embodiments, fresh meat from grazing sheep is placed in a cold storage at 0℃-4℃ for 48 hours to remove acidity. After the carcass is divided, the longissimus dorsi muscle between the 12th and 13th ribs is taken as the chilled fresh meat sample.

[0008] In some feasible embodiments, the following quality indicators are determined for each of the chilled meat samples from the same batch: Mechanical shearing tests were conducted on chilled fresh meat samples using a shear force tester, and the maximum shear force value during the shearing process was recorded as the tenderness value. After weighing the sample, a standard cooking treatment was carried out. After cooking and cooling, the sample was weighed again. The cooking loss rate was calculated by the difference in mass before and after cooking, which was used to characterize the water retention capacity of the sample during the heating process. The sample was suspended or left to stand under constant environmental conditions, and the drip loss rate was calculated after a preset time to characterize the degree of loss of free water in the sample. The average value of various quality indicators is taken as the final quality reference value.

[0009] In some feasible embodiments, after the various quality indicators of the chilled meat samples are determined, the chilled meat samples of the same batch are placed sequentially on the hyperspectral imaging system acquisition platform, and the chilled meat samples are scanned line by line to acquire the reflectance spectral information of the samples in the continuous band range.

[0010] In some feasible embodiments, after obtaining the original hyperspectral image of the chilled meat sample, the main body region of the chilled meat is first identified based on the spatial distribution of the sample in the hyperspectral image to determine the region of interest. After locating the region of interest, pixel-by-pixel spectral extraction is performed on all pixels within each region of interest, and each pixel forms an independent spectral curve. By analyzing the spectral differences of each pixel within the region of interest, effective spectra with significant response differences are selected, and a spectral matrix of all regions of interest within the sample is constructed; where each row corresponds to a region of interest, and each column corresponds to an effective spectral wavelength variable.

[0011] In some feasible embodiments, after obtaining the spectral matrix corresponding to each chilled meat sample, differential spectral preprocessing is performed on different quality indicators. For the effective spectrum corresponding to the tenderness index, wavelet transform is used to decompose each effective spectrum into sub-signals of different scales. After removing high-frequency noise, the effective spectral features related to tissue structure changes are retained, forming enhanced spectral features corresponding to the tenderness index.

[0012] In some feasible embodiments, a smoothing algorithm is used to suppress random noise and retain the overall absorption peak shape for the effective spectrum corresponding to the cooking loss index through local polynomial fitting, thereby forming the spectral characteristics corresponding to the cooking loss.

[0013] In some feasible embodiments, derivative transformation is used for the effective spectrum corresponding to the dripping loss index. The first derivative is used to represent the band change trend, and the second derivative is used to highlight the local absorption peak differences. After eliminating part of the baseline drift, the spectral characteristics corresponding to the dripping loss are formed.

[0014] In some feasible embodiments, after obtaining the preprocessed spectral inputs of various quality indicators of chilled meat samples, corresponding prediction models are established for tenderness, cooking loss, and drip loss, respectively, and the processing is as follows: Using the spectral features processed by wavelet transform as the input variable and the shear force reference value as the output variable, a tenderness prediction model is established by partial least squares regression to reflect the linear mapping relationship between shear force and spectral features. Using the spectral characteristics processed by the smoothing algorithm as input variables and the cooking loss reference value as output variables, a partial least squares regression model for predicting cooking loss water retention is constructed to reflect the mapping between cooking loss and spectral smoothing characteristics. Using the spectral characteristics after derivative transformation as input variables and the reference value of drip loss as output variables, a partial least squares regression model for predicting drip loss water retention is constructed to reflect the mapping between drip loss and local absorption peak variation characteristics.

[0015] In some feasible embodiments, during model training, the training set data is used to fit the parameters of the prediction models corresponding to various quality indicators, and the prediction performance of the models is evaluated based on the test set data. By comparing the coefficient of determination, root mean square error and prediction performance index, the predictive ability of each model for the corresponding quality indicators is verified.

[0016] (III) Beneficial Effects: Compared with the prior art, this invention has the following beneficial effects: This invention acquires hyperspectral images of chilled fresh meat from grazing sheep, establishes a calibration set by combining experimental data on shear force and water retention, and after ROI (region of interest) extraction and spectral preprocessing, constructs a multi-index prediction model based on PLSR (partial least squares regression) using characteristic wavelength screening and full-band comparative analysis to achieve non-destructive quantitative detection of tenderness and water retention. Attached Figure Description

[0017] Figure 1 This is a schematic flowchart of the method for detecting the tenderness and water retention of chilled fresh meat from grazing sheep based on hyperspectral imaging, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the selected chilled meat sample and the ROI region set on it, provided by the method for detecting the tenderness and water retention of chilled fresh meat from grazing sheep based on hyperspectral imaging in an embodiment of the present invention. Figure 3 This is a schematic diagram of the ROI regions selected for the chilled meat sample in the method for detecting the tenderness and water retention of chilled fresh meat from grazing sheep based on hyperspectral imaging provided in this embodiment of the invention. Figure 4 This is a schematic diagram of the spectra obtained from each ROI region without preprocessing in the method for detecting the tenderness and water retention of chilled fresh meat from grazing sheep based on hyperspectral imaging provided in the embodiments of the present invention. Figure 5 The method for detecting the tenderness and water retention of chilled fresh meat from grazing sheep based on hyperspectral imaging provided in this embodiment of the invention; a schematic diagram of the spectrum after performing SG preprocessing; Figure 6The method for detecting tenderness and water retention of chilled fresh meat from grazing sheep based on hyperspectral imaging provided in this embodiment of the invention; a schematic diagram of the spectrum after performing WT preprocessing; Figure 7 The method for detecting the tenderness and water retention of chilled fresh meat from grazing sheep based on hyperspectral imaging provided in this embodiment of the invention; a schematic diagram of the spectrum after DT preprocessing; Figure 8 This is a schematic diagram of the characteristic wavelength region selected for shear force in the method for detecting the tenderness and water retention of chilled fresh meat from grazing sheep based on hyperspectral imaging provided in an embodiment of the present invention. Figure 9 This is a schematic diagram of the characteristic wavelength region selected for cooking loss in the method for detecting tenderness and water retention of chilled fresh meat from grazing sheep based on hyperspectral imaging, as provided in an embodiment of the present invention. Figure 10 This is a schematic diagram of the characteristic wavelength region selected for drip loss in the method for detecting the tenderness and water retention of chilled fresh meat from grazing sheep based on hyperspectral imaging, provided in an embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0019] It should be noted that, where there is no conflict, the features in the embodiments of the present invention can be combined with each other.

[0020] Combination Figures 1 to 10 The method shown is a method for detecting the tenderness and water retention of chilled fresh meat from grazing sheep based on hyperspectral imaging. This method acquires hyperspectral images of chilled fresh meat from grazing sheep, establishes a calibration set by combining shear force and water retention experimental data, extracts ROI (region of interest) and performs spectral preprocessing, and constructs a multi-index prediction model based on PLSR (partial least squares regression) by using characteristic wavelength screening and full-band comparative analysis to achieve non-destructive quantitative detection of tenderness and water retention.

[0021] Fresh meat samples from grazing sheep were selected as the research object, and the tenderness and water retention of the same batch of samples were measured by physicochemical methods.

[0022] Specifically, multiple grazing sheep were selected, slaughtered, and their carcasses were placed in a 0℃ environment for 48 hours to remove acid. The longissimus dorsi muscle between the 12th and 13th ribs, weighing about 300g, was taken, wrapped in aluminum foil, placed in a sealed bag, and stored at 0℃-4℃ before being transported back to the laboratory.

[0023] When measuring the shear force of the samples, the measurement site is the longissimus dorsi muscle between the 12th and 13th ribs of the sheep. During the measurement, three pieces of meat of uniform size (5cm × 1cm × 1cm) are taken from each sample using a knife. These pieces are packaged in a self-sealing bag and heated in a water bath until the center temperature of the meat sample reaches 70℃. After the temperature of the meat sample returns to room temperature, the shear force instrument is used to measure the sample. To improve the accuracy of the measurement results, each sample is measured three times and the average value is taken as the final data.

[0024] When determining the cooking loss of the samples, a uniformly sized sample of the longest back muscle, weighing approximately 20g, was cut from the chilled fresh meat sample of a pasture-raised sheep using a scalpel, taking care to avoid areas such as fat and fascia. The initial weight of each meat sample was weighed using an electronic balance and recorded. The weighed meat sample was then placed in an inflatable resealable bag, and the bag was sealed to prevent moisture loss into the water bath during cooking. The resealable bag containing the meat sample was placed in a 72℃ constant temperature water bath, ensuring that the sample was completely submerged in the water. After cooking for 30 minutes, the sample was removed, and after cooling to room temperature, it was weighed again and the data was recorded. Then, the cooking loss of the sample was calculated.

[0025] When determining the water loss from a sample, first cut the sample into 5cm×3cm×2cm pieces with a scalpel, avoiding areas such as fat and fascia. After absorbing the surface moisture with absorbent paper, weigh and record the initial mass of the sample. Then, tie one end of the sample with a thin thread and hang it in a sealed bag using a hook, ensuring that the sample does not contact the bottom or side walls of the sealed bag. Seal the bag to prevent external moisture from entering. Then, place the sample in a 4℃ refrigerator for 24 hours. After absorbing the surface moisture with absorbent paper, weigh and record the final mass.

[0026] Finally, the results of the measurements were averaged to obtain a stable quality label dataset.

[0027] The above-mentioned chilled meat samples were non-destructively spectrally acquired using a hyperspectral imaging system to obtain three-dimensional hyperspectral image data containing spatial and continuous spectral information, thereby forming a cube of the original spectral response data of the samples in the visible to near-infrared band.

[0028] Specifically, a visible to near-infrared hyperspectral imaging system was used to acquire spectral images of chilled fresh meat samples from grazing sheep, such as... Figure 3As shown, the acquisition wavelength range is 400-1000 nm, totaling 300 bands, with a spectral resolution of 2.5 nm. The hyperspectral imaging system mainly consists of a hyperspectral camera (FigSpec FS-1X, Zhejiang), a charge-coupled device camera (CP-821N TYPE, Zhejiang), a 150W fiber optic halogen lamp white light source (7158 TYPE, Holland), and an electrically controlled translation stage (FIG-13 TYPE, Zhejiang).

[0029] To ensure the stability of the light source illumination, interference from ambient light should be minimized during the data acquisition process. Before each sample measurement, the hyperspectral imaging system should be powered on and warmed up for approximately 30 minutes to avoid measurement errors caused by changes in ambient temperature. When acquiring spectral data from chilled fresh meat samples from grazing sheep, the system parameters were set as follows: camera exposure time was 12 ms; the electronically controlled translation stage movement speed was set to 0.5 mm / s to ensure the integrity of the spectral images and data quality.

[0030] In hyperspectral images, effective regions of interest (ROIs) are selected based on the actual spatial distribution information of the meat sample. In some embodiments, ENVI software can be used to analyze each hyperspectral image, selecting 10 pixel regions from the sample as ROIs. Each region is approximately 100×100 pixels in size (with allowable fine-tuning of ±10 pixels). A schematic diagram of the selected ROIs and their corresponding spectra is shown below. Figure 2 and Figure 3 As shown.

[0031] The average spectral reflectance of all pixels within each ROI is calculated as the final spectral data for each pasture-raised sheep chilled meat sample, used for subsequent modeling. A schematic diagram of the original spectral curve plotted based on this average data is shown below. Figure 3 The background and non-target tissue regions were removed, and the pixels within the ROI were subjected to spectral extraction and convergence processing to construct the original spectral sample dataset corresponding to the chilled meat tissue.

[0032] To eliminate scattering effects, noise interference, and baseline drift in the raw spectral data after ROI extraction, different spectral preprocessing methods were employed to construct feature representation spaces for different quality indicators. These included:

[0033] Wavelet transform (WT) was used to perform multi-scale decomposition and denoising on the spectral data corresponding to the tenderness index, resulting in enhanced spectral features. The original spectrum before processing is shown below. Figure 4 As shown, the WT spectral processing results are as follows: Figure 5 As shown.

[0034] The spectral data corresponding to the cooking loss index were smoothed using a smoothing algorithm (SG) to enhance the overall spectral trend information. The spectral processing results are as follows: Figure 6 As shown.

[0035] First / second derivative transform (DT) was applied to the spectral data corresponding to the dripping loss index to enhance the differences in local absorption characteristics. The spectral processing results are as follows: Figure 7 As shown.

[0036] Through the above processing, three types of differentiated spectral feature input subspaces are formed. Based on the different preprocessed spectral data, the continuous projection algorithm (SPA) is introduced to filter high-dimensional spectral variables, extract feature wavelength combinations that are highly correlated with the quality indicators of chilled fresh meat, realize dimensionality reduction and redundant information removal of spectral data, and thus construct a representative set of low-dimensional feature variables.

[0037] Different characteristic wavelengths were obtained for different quality indicators of chilled fresh meat. For shear force, a total of 7 characteristic wavelengths were screened, located at 381.97 nm, 390.56 nm, 392.70 nm, 443.89 nm, 507.16 nm, 639.96 nm, and 798.98 nm, respectively. For cooking loss, a total of 8 characteristic wavelengths were screened, located at 384.12 nm, 422.62 nm, 619.37 nm, 932.35 nm, 970.71 nm, 978.79 nm, 980.81 nm, and 986.87 nm, respectively. For dripping loss, a total of 6 characteristic wavelengths were screened, located at 388.41 nm, 392.70 nm, 401.27 nm, 473.50 nm, 637.90 nm, and 792.90 nm, respectively. (See...) Figure 8-10 .

[0038] Before establishing a predictive model for the tenderness and water retention of chilled fresh meat from grazing sheep, it is necessary to divide the spectral data of all samples. The samples were randomly divided into a calibration set and a test set at a 3:1 ratio to ensure the scientific rigor of model training and validation. Ultimately, the calibration set contained 788 samples, and the test set contained 262 samples. Reference values ​​for shear force, cooking loss, and drip loss for each sample were recorded (see Tables 1-3) and mapped one-to-one with the divided sample sets, providing fundamental data for model construction.

[0039] Table 1. Results of shear force measurement Table 2 Results of cooking loss determination Table 3. Results of drip loss measurement Following the preprocessing described above, in this embodiment of the invention, five modeling methods—PLSR, Ada, RF, XGBoost, and LightGBM—were employed to model the tenderness, cooking loss, and drip loss of chilled fresh meat from grazing sheep across the entire spectral band. The training set data was used to fit the parameters of each model, and the predictive performance of the models was evaluated based on the test set data. By comparing indicators such as the coefficient of determination (R²), root mean square error (RMSE), and predictive performance index (RPD), the predictive ability of each model for the corresponding quality indicators was verified, thereby determining the final optimal combination of prediction models. The training results are shown in Tables 4-8 below.

[0040] Table 4. Modeling results of PLSR models under different preprocessing methods across the entire wavelength. Table 5. AdaBoost modeling results under different preprocessing methods across the entire wavelength range. Table 6. RF modeling results under different preprocessing methods across the entire wavelength. Table 7. Modeling results of XGBoost models under different preprocessing methods across the entire wavelength. Table 8. Modeling results of LightBGM model under different preprocessing methods across the entire wavelength. Based on the above tables, the PLSR model constructed using WT spectral preprocessing showed the best performance in predicting the shear force of chilled meat from grazing sheep, with an Rc² of 0.8943 and an RMSEC of 0.3252 on the training set, an Rp² of 0.8878 on the test set, and an RMSEP of 0.3256. For cooking loss, the PLSR model constructed using SG spectral preprocessing performed best, with an Rc² of 0.9004 and an RMSEC of 0.0056 on the training set, an Rp² of 0.8898 on the test set, and an RMSEP of 0.0059. For dripping loss, the PLSR model constructed using DT spectral preprocessing performed best, with an Rc² of 0.9172 and an RMSEC of 0.0037 on the training set, an Rp² of 0.8998 on the test set, and an RMSEP of 0.0039.

[0041] In summary, the optimal partial least squares regression (PLSR) models constructed for the characteristic wavelengths extracted from different quality indicators are as follows: For the shear force (tenderness) index, a WT-PLSR prediction model was constructed based on the WT preprocessed spectrum; For the cooking loss index, an SG-PLSR prediction model was constructed based on SG pretreatment spectra; For the drip loss index, a DT-PLSR prediction model was constructed based on DT preprocessed spectra.

[0042] The above method is used to establish the optimal spectral expression and regression mapping relationship for different quality indicators.

[0043] In the actual testing phase, hyperspectral images of the chilled fresh meat samples to be tested were acquired, and ROI extraction and extraction and preprocessing of the effective spectra of the corresponding quality indicators were performed in sequence. Among them, the WT-PLSR model was used to predict tenderness; the SG-PLSR model was used to predict cooking loss; and the DT-PLSR model was used to predict dripping loss.

[0044] Considering that water retention is a composite indicator, namely cooking loss plus dripping loss, in this embodiment of the invention, the contribution weights of the two types of indicators are normalized based on the regression coefficients obtained from the partial least squares regression model, and a weighted fusion is constructed to achieve a comprehensive prediction output of the water retention of chilled meat. Ultimately, this achieves simultaneous, non-destructive, quantitative detection of the tenderness and water retention of chilled meat based on hyperspectral information.

[0045] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. The scope of protection of the present invention is defined by the claims. Similarly, any equivalent structural changes made based on the description and drawings of the present invention should also be included within the scope of protection of the present invention.

Claims

1. A method for detecting the tenderness and water retention of chilled fresh meat from pasture-fed sheep based on hyperspectral imaging, characterized in that, The execution steps include the following: Fresh meat samples from grazing sheep were selected, and the tenderness and water retention of the same batch of samples were subjected to physicochemical tests. Among them, the shear force value was measured by a shear force meter to characterize the tenderness, and the water retention was quantified by cooking loss test and drip loss test. The chilled meat sample was subjected to non-destructive spectral acquisition to obtain a hyperspectral image containing spatial and continuous spectral information. In each of the hyperspectral images, a valid region of interest is selected. After removing the background and non-target tissue regions, the spectra of the pixels in each region of interest are extracted, and the spectra of the pixels in the region of interest are averaged to obtain the valid spectral data of the corresponding region. Different spectral preprocessing methods are used for the effective spectral data corresponding to the regions of interest for tenderness, cooking loss, and drip loss. Specifically, wavelet transform is used for the effective spectral data corresponding to tenderness; a smoothing algorithm is used for the effective spectral data corresponding to cooking loss; and derivative removal transform is used for the effective spectral data corresponding to drip loss. Based on the effective spectral data of each quality indicator after preprocessing, a continuous projection algorithm is introduced to screen the high-dimensional full-band spectral variables, extract the characteristic wavelength combinations that are highly correlated with each chilled fresh meat quality indicator, and construct the optimal prediction model for different quality indicators based on the partial least squares regression method. In the actual testing phase, hyperspectral images of the chilled fresh meat sample to be tested are acquired, and effective region of interest extraction, effective spectrum generation and corresponding spectrum preprocessing are performed in sequence. After extracting the characteristic wavelength variables, they are input into the optimal prediction model of the corresponding quality index, and the prediction output of the tenderness and water retention of chilled fresh meat is performed.

2. The method for detecting the tenderness and water retention of chilled fresh meat from grazing sheep based on hyperspectral imaging according to claim 1, characterized in that, Fresh meat from grazing sheep was placed in a cold storage at 0℃~4℃ for 48 hours to remove acid. After the carcass was divided, the longissimus dorsi muscle between the 12th and 13th ribs was taken as the chilled fresh meat sample.

3. The method for detecting the tenderness and water retention of chilled fresh meat from grazing sheep based on hyperspectral imaging according to claim 2, characterized in that, The following quality indicators were measured for each of the chilled meat samples from the same batch: Mechanical shearing tests were conducted on chilled fresh meat samples using a shear force tester, and the maximum shear force value during the shearing process was recorded as the tenderness value. After weighing the sample, a standard cooking treatment was carried out. After cooking and cooling, the sample was weighed again. The cooking loss rate was calculated by the difference in mass before and after cooking, which was used to characterize the water retention capacity of the sample during the heating process. The sample was suspended or left to stand under constant environmental conditions, and the drip loss rate was calculated after a preset time to characterize the degree of loss of free water in the sample. The average value of various quality indicators is taken as the final quality reference value.

4. The method for detecting the tenderness and water retention of chilled fresh meat from pasture-raised sheep based on hyperspectral imaging according to claim 3, characterized in that, After completing the determination of various quality indicators of the chilled meat samples, the chilled meat samples of the same batch were placed sequentially on the hyperspectral imaging system acquisition platform, and the chilled meat samples were scanned line by line to acquire the reflectance spectral information of the samples in the continuous band range.

5. The method for detecting the tenderness and water retention of chilled fresh meat from grazing sheep based on hyperspectral imaging according to claim 1, characterized in that, After obtaining the original hyperspectral image of the chilled meat sample, the main area of ​​the chilled meat is first identified based on the spatial distribution of the sample in the hyperspectral image to determine the region of interest. After locating the region of interest, pixel-by-pixel spectral extraction is performed on all pixels within each region of interest, and each pixel forms an independent spectral curve. By analyzing the spectral differences of each pixel within the region of interest, effective spectra with significant response differences are selected, and a spectral matrix of all regions of interest within the sample is constructed; where each row corresponds to a region of interest, and each column corresponds to an effective spectral wavelength variable.

6. The method for detecting the tenderness and water retention of chilled fresh meat from grazing sheep based on hyperspectral imaging according to claim 5, characterized in that, After obtaining the spectral matrix corresponding to each chilled meat sample, differential spectral preprocessing is performed on different quality indicators. For the effective spectrum corresponding to the tenderness index, wavelet transform is used to decompose each effective spectrum into sub-signals of different scales. After removing high-frequency noise, the effective spectral features related to tissue structure changes are retained, forming enhanced spectral features corresponding to the tenderness index.

7. The method for detecting the tenderness and water retention of chilled fresh meat from grazing sheep based on hyperspectral imaging according to claim 6, characterized in that, For the effective spectrum corresponding to the cooking loss index, a smoothing algorithm is used to suppress random noise through local polynomial fitting, while preserving the overall absorption peak shape, thus forming the spectral characteristics corresponding to the cooking loss.

8. The method for detecting the tenderness and water retention of chilled fresh meat from pasture-raised sheep based on hyperspectral imaging according to claim 7, characterized in that, For the effective spectrum corresponding to the dripping loss index, derivative transformation is used. The first derivative is calculated to represent the band change trend, and the second derivative is used to highlight the local absorption peak differences. After eliminating part of the baseline drift, the spectral characteristics corresponding to the dripping loss are formed.

9. The method for detecting the tenderness and water retention of chilled fresh meat from grazing sheep based on hyperspectral imaging according to claim 8, characterized in that, After obtaining the preprocessed spectral inputs of various quality indicators of chilled meat samples, corresponding prediction models were established for tenderness, cooking loss, and drip loss, respectively, and the following processing was performed: Using the spectral features processed by wavelet transform as the input variable and the shear force reference value as the output variable, a tenderness prediction model is established by partial least squares regression to reflect the linear mapping relationship between shear force and spectral features. Using the spectral characteristics processed by the smoothing algorithm as input variables and the cooking loss reference value as output variables, a partial least squares regression model for predicting cooking loss water retention is constructed to reflect the mapping between cooking loss and spectral smoothing characteristics. Using the spectral characteristics after derivative transformation as input variables and the reference value of drip loss as output variables, a partial least squares regression model for predicting drip loss water retention is constructed to reflect the mapping between drip loss and local absorption peak variation characteristics.

10. The method for detecting the tenderness and water retention of chilled fresh meat from pasture-raised sheep based on hyperspectral imaging according to claim 1, characterized in that, During model training, the parameters of the prediction models corresponding to various quality indicators are fitted using the training set data, and the prediction performance of the models is evaluated based on the test set data. By comparing the coefficient of determination, root mean square error, and prediction performance index, the predictive ability of each model for the corresponding quality indicators is verified.

Images