A method for detecting the degree of woodiness of chicken breast meat based on hyperspectral imaging technology

Abstract

A method for detecting the degree of chicken breast lignification based on hyperspectral imaging technology belongs to the field of nondestructive testing of agricultural products, which includes four grades of normal, slight, moderate and severe lignification of chicken breast by palpation, collection of hyperspectral images of chicken breast with different degrees of lignification in the 400-1000 nm and 1000-2000 nm bands, extraction of reflectivity information through the region of interest, processing of spectral information of chicken breast by Autoscale, SNV, OSC, Smoothing and 1st different pretreatment methods, construction of PLS-DA and SVM discrimination models based on full wavelength, and selection of the best pretreatment method. Three methods of SPA, CARS and UVE are used to select characteristic wavelengths, effective variables of single band are selected by comparing modeling results of different characteristic wavelengths, and then spectral data fusion is used to realize the application of hyperspectral technology in the classification of chicken breast lignification.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of non-destructive testing of agricultural products, and particularly relates to a method for detecting the degree of woodiness of chicken breast based on hyperspectral imaging technology. BACKGROUND

[0002] Chicken breast is a meat product that is very popular among consumers. It has the nutritional characteristics of "one high and three low", which means high protein content and low fat, calorie and cholesterol content. With the upgrading of consumption structure, the consumption of chicken is also increasing. In order to improve the quality of chicken breast and meet the growing demand of consumers, breeding scientists are actively breeding excellent varieties. In this process, the rapid growth of chicken muscle leads to a series of poultry breast diseases such as woodiness chicken breast (WB). Although eating woodiness chicken breast has no harmful effect on the human body, the taste of woodiness chicken breast is worse than that of normal chicken breast. In addition, the protein content of WB decreases and the fat content increases. The decrease of nutritional value and the decrease of taste acceptability affect the purchase of consumers and bring serious losses to the chicken industry. Therefore, improving the detection efficiency of woodiness chicken breast, integrating the development of woodiness degree and quality detection technology of chicken breast has important practical significance and actual demand for improving the quality and safety of chicken.

[0003] At present, the detection of WB mainly adopts palpation method, which relies on personal subjective feeling and is easily affected by environmental factors such as temperature, humidity and light, and cannot achieve accurate classification. Therefore, it is necessary to develop a rapid, accurate and reliable method to detect the degree of woodiness of chicken breast to meet the needs of the agricultural and food industries.

[0004] Previous studies have shown that spectral technology can be used for the detection of woodiness chicken breast, but these technologies do not achieve fine classification of the degree of woodiness. There is still a large development space for further classification and detection of the degree of woodiness of chicken breast. Hyperspectral imaging technology (HSI) combines image and data, and can obtain spectral information and image information of the sample at the same time, realizing the integration of image and spectrum. Therefore, hyperspectral imaging technology can detect the external characteristics of the sample through image information and obtain the internal information of the sample through spectral information, but there is no research on using different waveband hyperspectral fusion strategy to detect the degree of woodiness of chicken breast. SUMMARY

[0005] In view of the problems in the background art, the application designs a method for detecting the degree of woodiness of chicken breast based on hyperspectral imaging technology, which aims to provide a method for detecting the degree of woodiness of chicken breast based on hyperspectral imaging technology with fast detection speed and 100% correct discrimination rate.

[0006] The technical solution of the application is as follows:

[0007] A method for detecting the degree of chicken breast lignification based on hyperspectral imaging technology, comprising the following steps:

[0008] S1, preparing samples: divide and select chicken breast samples by visual inspection and palpation, sample and remove surface fascia, and store at 4°C.

[0009] S2, hyperspectral image acquisition and correction: use visible-near infrared hyperspectral imaging system and short-wave infrared hyperspectral imaging system to collect hyperspectral images of samples in the 400-1000nm visible-near infrared and 1000-2000nm short-wave infrared wavelength ranges, and perform black and white correction on the collected hyperspectral images to eliminate redundant information. White correction uses a white polytetrafluoroethylene plate with a reflectivity of 99.99%, and black correction is achieved by covering the lens with a cover. The final hyperspectral corrected image is obtained by the following formula:

[0010]

[0011] Wherein: R is the hyperspectral corrected image, R0 is the original hyperspectral image, B is the standard black correction image, and W is the standard white correction image.

[0012] S3, spectral information extraction: use image segmentation method to extract typical spectral data of chicken breast samples, select the head position of chicken breast as the region of interest, the size of each region is 900(30x30) pixels, use Matlab software(MATLAB R2016a, Mathworks company, USA) to identify ROI and extract spectral data in ROI, average all spectra of each pixel point in ROI, and take the average value as the spectrum of each chicken breast sample.

[0013] S4, spectral pretreatment method and feature variable selection: the pretreatment methods mainly include Autoscale(automatic standardization), SNV(standard normal variable method), OSC(orthogonal signal correction method), Smoothing(smoothing method), 1-st(first derivative method) and the like. After processing the spectral data using different pretreatment methods, three methods of SPA(sequential projection algorithm), CARS(competitive adaptive reweighted algorithm) and UVE(uninformative variable elimination) are used to select characteristic wavelengths, PLS-DA and SVM discriminant models based on full wavelength and characteristic wavelength are established respectively, the stability and accuracy of different models are compared, and effective characteristic wavelengths are screened.

[0014] S5, spectral data fusion strategy: for the above-mentioned 43 characteristic wavelengths of UVE in the visible-near infrared band: 412.07, 416.11, 433.69, 437.77, 451.41, 492.78, 505.3, 509.49, 510.89, 516.48, 519.28, 526.29, 527.7, 529.1, 530.51, 533.32, 534.73, 581.47, 582.89, 584.32, 585.75, 587.17, 588.6, 590.03, 591.46, 605.76, 641.7, 708.39, 739.02, 740.48, 741.94, 743.4, 747.79, 750.71, 760.94, 765.33, 837.12, 850.31, 860.57, 886.93, 897.17, 951.18, 990.40 nm and 11 characteristic wavelengths obtained by SPA in the short-wave infrared band: 994.80, 1013.63, 1032.69, 1104.47, 1158.33, 1427.91, 1670.41, 1830.30, 1866.00, 1894.34, 1984.74 nm, the characteristic wavelength spectral data in the two bands are fused. When performing data fusion, the preprocessed and characteristic wavelength screened spectral data are spliced into a matrix, and a SVM fusion discrimination model for classifying the degree of chicken breast woodification is established from the feature fusion level.

[0015] S6, establishing a discrimination model: based on the same discrimination model establishment method of full spectrum, characteristic band and spectral data fusion, all chicken breast samples are divided according to the proportion, the model is established, and the performance of the model is evaluated by the discrimination accuracy of the modeling set and the prediction set.

[0016] Preferably, the chicken breast in S1 is divided into four categories according to the hardness of the chicken breast when touched: normal chicken breast (the whole chicken breast is very soft, the surface is smooth and delicate, and the two ends of the chicken breast naturally droop when held in the hand), slightly woodified chicken breast (hardness mainly concentrates in the top end area, and there is still a certain drooping feeling at the bottom when held in the hand), moderately woodified chicken breast (hardness mainly concentrates in the top end and middle area, the bottom is softer than the top end, and the bottom can be seen as a convex shape), and severely woodified chicken breast (the whole touch feels hard, there is no drooping feeling, and the bottom can be clearly seen as a convex part).

[0017] Preferably, the normal (WB) chicken breast, slightly (MILD) woodified chicken breast, moderately (MOD) woodified chicken breast and severely (SEV) woodified chicken breast samples in S1 are 60 each.

[0018] Preferably, in S2, the hyperspectral imaging system is turned on 30 minutes in advance for preheating before the sample hyperspectral image is collected. After the light source is stabilized, a scanning test is performed. After the surface of the chicken breast is wiped clean of moisture, the chicken breast is placed on the moving platform for image collection.

[0019] Preferably, in S2, the visible-near-infrared hyperspectral imaging system comprises a computer, a camera, an imaging spectrometer, halogen light sources, and a moving platform. The camera is a high-resolution CCD camera with a single-wavelength image pixel of 804x534. The imaging spectrometer has a spectral resolution of 2.8 nm and is used to obtain image information and spectral information of the chicken breast sample. The halogen light sources are two, which are fixed on the two sides of the camera at a 45° projection angle at a distance of 30 cm from the sample on the moving platform. The distance between the camera lens and the sample on the moving platform is 26 cm. The halogen light source intensity is set to 45 W, the exposure time is set to 3 ms, and the moving speed is 7 mm / s.

[0020] Preferably, in S2, the short-wave near-infrared hyperspectral imaging system comprises a computer, a camera, an imaging spectrometer, halogen light sources, and a moving platform. The camera is a CL camera with a single-wavelength image pixel of 320x472. The imaging spectrometer has a spectral resolution of 6.2 nm. The halogen light sources are fixed at an oblique upper position at a 45° projection angle at a distance of 31 cm from the sample on the moving platform. The distance between the camera lens and the sample on the moving platform is 26 cm. The halogen light source intensity is set to 255 W, the exposure time is set to 3.5 ms, and each sample is scanned at a moving speed of 17 mm / s.

[0021] Preferably, in S3, the head end that undergoes lignification first is selected as the region of interest, which has good sample representativeness and can fully reflect the characteristics of chicken breast lignification.

[0022] Preferably, in S5, the data is standardized before the model is established to eliminate the dimension. The spectral data is limited to 0-1 using the normalization preprocessing method, and the formula is as follows:

[0023]

[0024] wherein X' is the normalized data, X is the original data, X max is the maximum value of the data, and X min is the minimum value of the data.

[0025] Preferably, in S6, all chicken breast samples are divided according to the ratio of modeling set: prediction set = 3:1, and PLS-DA (partial least squares discriminant analysis) and SVM (support vector machine) algorithms are used to establish a chicken breast lignification grade discrimination model.

[0026] The beneficial effects of the present application are: (1) in the woodification chicken breast grade discrimination model developed based on full-waveband spectral information, the SVM model with 400-1000 nm waveband pretreated by SNV and 1000-2000 nm pretreated by Autoscale has the highest discrimination accuracy in single data model, but the data volume is large and the model running speed is slow; (2) after selecting characteristic wavelengths using SPA, CARS and UVE three algorithms, the model closest to full spectral data can be established using about 10% spectral data; 43 characteristic variables are reserved in the 400-1000 nm waveband using UVE algorithm, and the accuracy rates of the modeling set and the prediction set are 98.3% and 88.3% respectively; 11 effective variables are reserved in the 1000-2000 nm waveband using SPA algorithm, which can reduce the operation amount and improve the operation efficiency, and the accuracy rates of the modeling set and the prediction set are 96.7% and 95.0% respectively, which has good prediction accuracy; (3) the accuracy rates of the modeling set and the prediction set of the model based on visible-near infrared hyperspectral and short-wave infrared hyperspectral characteristic data fusion are 98.9% and 96.7% respectively, and the discrimination accuracy of normal chicken breast and severely woodification chicken breast has reached 100%. BRIEF DESCRIPTION OF DRAWINGS

[0027] Figure 1 The flow chart for detecting the woodification grade of chicken breast based on hyperspectral imaging technology in the present application.

[0028] Figure 2 The visible-near infrared hyperspectral detection system diagram in the present application.

[0029] Wherein: 1, camera, 2, imaging spectrometer, 3, lens, 4, halogen light source, 5, computer, 6, mobile platform, 7, sample.

[0030] Figure 3 The short-wave infrared hyperspectral detection system diagram in the present application.

[0031] Wherein: 1, camera, 2, imaging spectrometer, 3, lens, 4, halogen light source, 5, computer, 6, mobile platform, 7, sample.

[0032] Figure 4 The spectral characteristic diagram of different grades of woodification chicken breast in 400-1000 nm and 1000-2000 nm.

[0033] Wherein: a, original spectrum (400-1000 nm), b, average spectrum (400-1000 nm), c, original spectrum (1000-2000 nm), d, average spectrum (1000-2000 nm). DETAILED DESCRIPTION

[0034] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all the other embodiments obtained by a person of ordinary skill in the art without creative effort belong to the scope of protection of the present application.

[0035] Embodiment 1

[0036] A method for detecting the degree of woodiness of chicken breast meat based on hyperspectral imaging technology, comprising the following steps:

[0037] S1, preparing samples: 60 normal chicken breast meat samples, 60 slightly woodiness chicken breast meat samples, 60 moderate woodiness chicken breast meat samples and 60 severe woodiness chicken breast meat samples are selected by visual inspection and palpation.

[0038] In this embodiment, the samples are purchased from Jiangsu Yike Food Group Co., Ltd. Meat Food Chicken Slaughterhouse. After sampling, the samples are quickly transported to the laboratory at 4 DEG C using a refrigerated box. After removing the surface muscle membrane from each sample, the sample is individually packaged in a disposable plastic bag and labeled, and stored at 4 DEG C.

[0039] S2, hyperspectral image acquisition and correction:

[0040] The spectral information of the chicken breast meat is collected, and a hyperspectral image is collected for each sample.

[0041] The visible-near infrared hyperspectral imaging system used in this embodiment is shown in Figure 1 The system uses a high-resolution CCD camera 1 with 804x534 single-wavelength image pixels, and an imaging spectrometer 2 with a spectral resolution of 2.8 nm. The two halogen light sources 4 are fixed at a distance of 30 cm from the sample at a projection angle of 45 DEG on both sides of the CCD camera 1. The distance between the camera lens 3 and the sample 7 is 26 cm, the light source intensity is set to 45 W, the exposure time is set to 3 ms, and the moving speed is 7 mm / s. The short-wave near-infrared hyperspectral imaging system used is shown in Figure 2 The system uses a CL camera 1 with 320x472 single-wavelength image pixels, and an imaging spectrometer 2 with a spectral resolution of 6.2 nm. The halogen light source 4 is fixed at a distance of 31 cm from the sample at a projection angle of 45 DEG on the upper side, and the distance between the camera lens 3 and the sample 7 is finally determined to be 26 cm. The light source intensity is set to 255 W, the exposure time is set to 3.5 ms, and the moving speed is 17 mm / s. Before spectral acquisition, the hyperspectral imaging system is turned on for preheating 30 minutes in advance. After the light source is stable, the scanning test is performed. After wiping off the water on the surface of the chicken breast meat, it is placed on the moving platform for image acquisition.

[0042] White correction uses a white polytetrafluoroethylene plate with a reflectivity of 99.99%, black correction is achieved by covering the lens with a machine cover, and the final hyperspectral correction image is obtained by the following formula:

[0043]

[0044] Wherein: R is the hyperspectral correction image, R0 is the original hyperspectral image, B is the standard black correction image, and W is the standard white correction image.

[0045] S3, spectral information extraction: in this embodiment, the image segmentation method is used to extract the typical spectral data of chicken breast samples. The head position of the chicken breast is taken as the region of interest, the size of each region is 900 (30x30) pixels, the recognition of ROI and the extraction of spectral data in ROI are carried out by using Matlab software (MATLAB R2016a, Mathworks company, USA), and the average value is taken as the spectrum of each chicken breast sample.

[0046] S4, spectral pretreatment method and variable selection: in this embodiment, the pretreatment methods mainly include Autoscale, SNV, OSC, Smoothing, 1-st, etc. After the spectral data is processed by different pretreatment methods, SPA, CARS and UVE are used to select characteristic wavelengths, and full wavelength and characteristic wavelength discrimination models are established respectively, and the stability and accuracy of different models are compared.

[0047] S5, spectral data fusion strategy: for the above-mentioned 43 characteristic wavelengths of UVE in the visible-near infrared band: 412.07, 416.11, 433.69, 437.77, 451.41, 492.78, 505.3, 509.49, 510.89, 516.48, 519.28, 526.29, 527.7, 529.1, 530.51, 533.32, 534.73, 581.47, 582.89, 584.32, 585.75, 587.17, 588.6, 590.03, 591.46, 605.76, 641.7, 708.39, 739.02, 740.48, 741.94, 743.4, 747.79, 750.71, 760.94, 765.33, 837.12, 850.31, 860.57, 886.93, 897.17, 951.18, 990.40 nm and 11 characteristic wavelengths obtained by SPA in the short-wave infrared band: 994.80, 1013.63, 1032.69, 1104.47, 1158.33, 1427.91, 1670.41, 1830.30, 1866.00, 1894.34, 1984.74 nm, the characteristic wavelength spectral data of the two bands are fused. When performing data fusion, the preprocessed and characteristic wavelength selected spectral data are spliced into a matrix, and a SVM fusion discrimination model for classifying the degree of chicken breast woodification is established from the feature fusion level.

[0048] S6, establish discrimination model: the discrimination model establishment method based on full spectrum, characteristic band and spectral data fusion is consistent. All samples are divided according to the ratio of modeling set: prediction set = 3:1, and PLS-DA and SVM algorithm is used to establish the model, and the performance of the model is evaluated by the discrimination accuracy of the modeling set and the prediction set.

[0049] Table 1 and Table 2 are respectively the performance parameters of the optimal discrimination model based on 400-1000 nm band and 1000-2000 nm band; Table 3 is the classification result of woodified chicken breast based on characteristic band; Table 4 is the performance parameter of the discrimination model of woodified chicken breast established based on spectral data fusion.

[0050] Table 1 modeling results of woodified chicken breast grade classification based on 400-1000 nm band (including application of different spectral pretreatment methods)

[0051]

[0052] Table 2 modeling results of woodified chicken breast grade classification based on 1000-2000 nm band (including application of different spectral pretreatment methods)

[0053]

[0054] Table 3 Modeling results of woodiness chicken breast classification based on characteristic wavelengths (including applying different characteristic wavelength screening methods)

[0055]

[0056] Table 4 Woodiness chicken breast SVM classification models based on visible-near infrared hyperspectral and short wave infrared hyperspectral data fusion (including modeling set and prediction set)

[0057]

[0058] This example discusses the spectral characteristics of different grades of woodiness chicken breast. As shown in Figure 1, with the increase of woodiness grade, the spectral reflectance intensity of the sample gradually changes from low to high in a gradient manner, and relatively obvious spectral absorption peaks appear at 420 nm, 550 nm, 760 nm and 970 nm, 1190 nm, 1420 nm and 1940 nm. Figure 4

[0059] This example includes the influence of different spectral pretreatment methods on the performance of the woodiness chicken breast grade identification model constructed under the full waveband of 400-1000 nm, as shown in Table 1. The modeling set accuracy of the model established by the five pretreatment methods is higher than that of the original spectrum. Among them, the SNV pretreatment method increases the modeling set and prediction set accuracy by 12.8% and 5.0%, respectively, and is the best in the PLS-DA model. In the SVM model, using the SNV pretreatment method, the underfitting condition is improved to some extent, and the accuracy of the modeling set and the prediction set is 98.3% and 88.3%, respectively, which is 9.4% and 8.3% higher than that of the PLS-DA model established by the same SNV pretreatment method.

[0060] This example includes the influence of different spectral pretreatment methods on the performance of the woodiness chicken breast grade identification model constructed under the full waveband of 1000-2000 nm, as shown in Table 2. In the PLS-DA model, it can be seen that when using Autoscale, SNV, 1-st and OSC pretreatment methods, the accuracy of the modeling set and the prediction set is higher than that without pretreatment. Among them, the model constructed by Autoscale pretreatment has an accuracy of more than 90.0%, and the discrimination accuracy of the four woodiness grades is more than 80.0%. In the constructed SVM model, the Autoscale-SVM model has the best effect, with the modeling set and prediction set accuracy of 96.1% and 95.0%, respectively, and the discrimination accuracy of the four different woodiness grades in the modeling set is more than 95.0%, and the prediction set is more than 85.0%.​

[0061] The present embodiment includes selecting the results of the discriminant model established at the characteristic wavelength based on different algorithms, as shown in Table 3. The three algorithms of SPA, CARS and UVE are used to screen the characteristic wavelength, and the UVE algorithm retains 43 characteristic wavelengths in the 400-1000 nm wavelength range. The correct rates of the modeling set and the prediction set using the discriminant model are 87.8% and 85.0%, respectively, which is slightly lower than the full wavelength range, but the running speed is greatly improved. In the 1000-2000 nm wavelength range, the SPA algorithm retains 11 characteristic wavelengths, and the correct rate of the modeling set is increased by 0.6% compared with the full wavelength. At this time, the correct rates of the modeling set and the prediction set are 96.7% and 95.0%, respectively.

[0062] The present embodiment also includes the discriminant model of the woodiness chicken breast grade based on the visible-near infrared hyperspectral and short-wave infrared hyperspectral multi-source data fusion, and the results are shown in Table 4. The spectral data fusion model is better than the single visible-near infrared hyperspectral or single short-wave infrared hyperspectral information model, and the correct rates of the modeling set and the prediction set are 98.9% and 96.7%, respectively. The modeling set and the prediction set of the multi-source data fusion model only have 2 misjudgments, and the data fusion model has good discriminant effect on the four grades. The discriminant correct rates of the normal chicken breast and the severely woodiness chicken breast have reached 100%.

[0063] The above describes the embodiments of the present disclosure, and the above description is exemplary and is not limited to the disclosed embodiments. Various exemplary embodiments, features and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings.

[0064] The above only describes the preferred embodiments of the present application, and is not intended to limit the present application. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for detecting the degree of woodiness in chicken breast meat based on hyperspectral imaging technology, characterized by: The method comprises the following steps: S1, preparing samples: selecting chicken breast samples by visual inspection and palpation, sampling and removing surface muscle membranes, and storing them at 4°C; S2, hyperspectral image acquisition and correction: using a visible-near-infrared hyperspectral imaging system and a short-wave infrared hyperspectral imaging system to collect hyperspectral images of samples in the 400-1000 nm visible-near-infrared and 1000-2000 nm short-wave infrared wavelength ranges, performing black and white correction on the collected hyperspectral images to eliminate redundant information, using a white polytetrafluoroethylene plate with a reflectivity of 99.99% for white correction, and performing black correction by covering the lens with a machine cover, and finally obtaining the hyperspectral corrected image through the following formula: Wherein: R is the hyperspectral corrected image, R0 is the original hyperspectral image, B is the standard black correction image, and W is the standard white correction image; S3, spectral information extraction: using image segmentation to extract typical spectral data of chicken breast samples, selecting the head position of the chicken breast as the region of interest, the size of each region being 900 pixels, using Matlab software to identify the ROI and extract spectral data within the ROI, averaging all the spectra of each pixel point in the ROI, and taking the average value as the spectrum of each chicken breast sample; S4, spectral pretreatment method and feature variable selection: the pretreatment methods used are Autoscale, SNV, OSC, Smoothing and 1-st, after processing the spectral data using different pretreatment methods, three methods of successive projection algorithm SPA, CARS and UVE are used to select characteristic wavelengths, PLS-DA and SVM discriminant models based on full wavelength and characteristic wavelength are established, the stability and accuracy of different models are compared, and effective characteristic wavelengths are screened. S5, spectral data fusion strategy: for the above-mentioned using UVE in the visible-near infrared wave band of 43 characteristic wavelengths: 412.07, 416.11, 433.69, 437.77, 451.41, 492.78, 505.3, 509.49, 510.89, 516.48, 519.28, 526.29, 527.7, 529.1, 530.51, 533.32, 534.73, 581.47, 582.89, 584.32, 585.75, 587.17, 588.6, 590.03, 591.46, 605.76, 641.7, 708.39, 739.02, 740.48, 741.94, 743.4, 747.79, 750.71, 760.94, 765.33, 837.12, 850.31, 860.57, 886.93, 897.17, 951.18, 990.40 nm and using SPA in the short wave infrared wave band to obtain 11 characteristic wavelengths: 994.80, 1013.63, 1032.69, 1104.47, 1158.33, 1427.91, 1670.41, 1830.30, 1866.00, 1894.34, 1984.74 nm, the characteristic wavelength spectral data of the two wave bands are fused; when the data fusion is carried out, the spectral data after the pretreatment and the characteristic wavelength screening are spliced into a matrix, and the SVM fusion discrimination model for classifying the degree of woodification of chicken breast is established from the feature fusion level; S6, establishing a discrimination model: the discrimination model establishment methods based on full spectrum, characteristic wave band and spectral data fusion are consistent, all chicken breast samples are divided according to the proportion, the model is established, and the performance of the model is evaluated through the discrimination accuracy of the modeling set and the prediction set.

2. The method for detecting the degree of woodiness in chicken breast meat based on hyperspectral imaging technology according to claim 1, characterized in that: The chicken breast in S1 is divided into four categories of normal chicken breast, slightly woodified chicken breast, moderately woodified chicken breast and severely woodified chicken breast according to the difference in the hardness of the chicken breast when touched.

3. The method for detecting the degree of woodiness in chicken breast meat based on hyperspectral imaging technology according to claim 1, characterized in that: The normal chicken breast, slightly woodified chicken breast, moderately woodified chicken breast and severely woodified chicken breast samples in S1 are 60 each.

4. The method for detecting the degree of woodiness in chicken breast meat based on hyperspectral imaging technology according to claim 1, characterized in that: In S2, 30 minutes before the sample hyperspectral image acquisition, the hyperspectral imaging system is preheated, the scanning test is carried out after the light source is stabilized, the water on the surface of the chicken breast is wiped clean, and then the image acquisition is carried out on the moving platform.

5. The method for detecting the degree of woodiness in chicken breast meat based on hyperspectral imaging technology according to claim 1, characterized in that: The S2 visible-near-infrared hyperspectral imaging system comprises a computer, a camera, an imaging spectrometer, halogen light sources, and a moving platform. The camera is a high-resolution CCD camera with 804*534 single-wavelength image pixels. The imaging spectrometer has a spectral resolution of 2.8 nm and is used to obtain image information and spectral information of chicken breast samples. The halogen light sources are two, which are fixed on both sides of the camera at a projection angle of 45° at a distance of 30 cm from the sample on the moving platform. The distance between the camera lens and the sample on the moving platform is 26 cm. The halogen light source intensity is set to 45 W, the exposure time is set to 3 ms, and the moving speed is 7 mm / s.

6. The method for detecting the degree of woodiness in chicken breast meat based on hyperspectral imaging technology according to claim 1, characterized in that: The S2 short-wave near-infrared hyperspectral imaging system comprises a computer, a camera, an imaging spectrometer, halogen light sources, and a moving platform. The camera is a CL camera with 320*472 single-wavelength image pixels. The imaging spectrometer has a spectral resolution of 6.2 nm. The halogen light sources are fixed at an angle of 45° above the sample on the moving platform at a distance of 31 cm. The distance between the camera lens and the sample on the moving platform is 26 cm. The halogen light source intensity is set to 255 W, the exposure time is set to 3.5 ms, and each sample is scanned at a moving speed of 17 mm / s.

7. The method for detecting the degree of woodiness in chicken breast meat based on hyperspectral imaging technology according to claim 1, characterized in that: In S3, the head end of the most initial lignification is selected as the region of interest, which has good sample representativeness and can fully reflect the characteristics of chicken breast lignification.

8. The method for detecting the degree of woodiness in chicken breast meat based on hyperspectral imaging technology according to claim 1, characterized in that: In S5, the data is standardized before modeling to eliminate dimensions. The spectral data is limited between 0 and 1 using the normalization preprocessing method, and the formula is as follows: wherein: X' is the normalized data, X is the original data, X max is the maximum value of the data, X min is the minimum value of the data.

9. The method for detecting the lignification grade of chicken breast based on hyperspectral imaging technology according to claim 1, characterized in that In S6, all chicken breast samples are divided into modeling set: prediction set = 3:1 according to the proportion, and PLS-DA and SVM algorithms are used to establish a chicken breast lignification grade discrimination model.

Images