A method for detecting heat-killed brucella suis content in chilled pork loin based on structured light hyperspectral imaging technology

Abstract

The application discloses a kind of detection methods of hot killed brucella content in chilled pork loin based on structured light hyperspectral imaging technology.The prediction method includes: with the chilled pork loin inoculated with hot killed brucella as the research object, the structured light hyperspectral image of chilled pork loin is collected using structured light hyperspectral imaging system, and the determination of hot killed brucella content is carried out;The corrected structured light hyperspectral image is demodulated, AC image and DC image are obtained, the full-band spectral information of image is extracted and the characteristic wavelength of chilled pork loin is screened, and a plurality of modeling methods are used to establish hot killed brucella content prediction model based on full-band and characteristic wavelength, and the hot killed brucella content of chilled pork loin can be predicted according to the established model.The application can realize the nondestructive testing of hot killed brucella content in chilled pork loin, with high detection precision and fast speed, thereby providing technical reference for nondestructive testing of chilled meat microbial indicators.

Description

Technical Field

[0001] This invention relates to the field of hyperspectral nondestructive testing, specifically to a method for detecting the content of *Heterobacter pyridostigma* in chilled fresh pork chops based on structured light hyperspectral imaging technology. Background Technology

[0002] Chilled meat, also known as fresh meat, refers to fresh meat whose carcasses are rapidly cooled to 0-4°C after slaughter and maintained at this temperature during subsequent storage and transportation. It boasts advantages such as easy digestibility, excellent flavor, and rich nutrition, making it undoubtedly the meat product with the greatest development potential, especially considering the dietary habits of Chinese consumers. However, its short shelf life has always been a major obstacle to the further development of chilled pork, and external microbial contamination is the primary cause of this shortened shelf life. *Heat-killing bacteria* is a Gram-positive bacterium that can grow rapidly under low-temperature, anaerobic conditions (0-4°C) and is the dominant spoilage bacterium causing the spoilage of chilled pork. Therefore, detecting *Heat-killing bacteria* in chilled pork is of great significance in evaluating its freshness.

[0003] The most common method for detecting microbial indicators in chilled pork is the plate count method specified in national standards, which is accurate and effective in observing colony morphology. In recent years, newer methods for detecting microbial counts include colony test strips and quantitative PCR. While these methods can effectively detect total bacterial counts, they require extensive preparation and specialized operation, are less efficient, and can even damage the meat being tested. Therefore, they are not suitable for large-scale online testing and cannot meet the testing needs of automated factory production and import / export trade. Consequently, the meat industry needs a rapid and non-destructive method for detecting microbial indicators in chilled pork.

[0004] Hyperspectral imaging technology, as an emerging platform technology, combines the advantages of both spectral and imaging technologies. It can not only detect the physicochemical properties of samples but also acquire their spatial distribution information, thus obtaining combined detection results. Existing research indicates that conventional hyperspectral imaging technology is widely used for the non-destructive detection of microbial indicators in chilled pork. However, conventional hyperspectral imaging technology has low spectroscopic efficiency, making it difficult to fully utilize the energy of all incident light. Furthermore, using high-resolution sensors cannot effectively improve resolution, thus affecting prediction accuracy to some extent.

[0005] Structured light is a novel optical technology, a lighting pattern characterized by alternating light and dark illumination with a sinusoidal light intensity distribution. Demodulating the acquired structured light image yields DC and AC images. The DC image is equivalent to the image acquired under uniform illumination, while the AC image enhances spatial resolution and contrast. Structured light hyperspectral imaging systems, using structured light as the imaging source, combine the resolution enhancement of structured light technology with the 'image-spectrum integration' advantage of hyperspectral technology, showing great potential for non-destructive detection of microbial indicators in chilled meat products. However, there is currently no research on the detection of microbial indicators in chilled meat products based on structured light hyperspectral imaging technology.

[0006] Application No. 2018102251286 discloses an online rapid detection method for *Heterospora heat-killing* content in chicken meat. The method involves acquiring hyperspectral images of chicken samples from a calibration set, preprocessing the acquired spectra, identifying the target region, and extracting the average spectral data. The extracted spectral data is then substituted into a formula to obtain the final result. This invention extracts 23 optimal wavelengths from 486 wavelengths across the entire spectrum to eliminate a large amount of redundant information and extract useful information, thereby reducing the computational load of data analysis and improving the accuracy of the partial least squares model to meet the needs of large-scale online production in meat processing enterprises. Compared with existing technologies, this invention has the following advantages: It does not require preprocessing of the sample, only performs non-contact spectral scanning, and is non-destructive; it does not use any chemical reagents, making it both environmentally friendly and cost-effective; and it is easy to operate, saves time, and enables large-scale online detection of *Heterospora heat-killing* content in chicken meat.

[0007] This scheme is a non-destructive testing case based on hyperspectral imaging. However, the uniform light used in ordinary hyperspectral imaging technology can only obtain one-dimensional spatial information of the sample, and scattering occurs when it shines on the sample, resulting in degraded image contrast and significant light defocusing, making it unsuitable for food quality testing. In contrast, structured light hyperspectral imaging technology uses structured light that can penetrate deep into the sample tissue from the surface to obtain its two-dimensional spatial information, enabling in-depth analysis of spectral image information and signal enhancement, thereby improving detection accuracy. Summary of the Invention

[0008] To address the aforementioned issues, this invention provides a method for detecting the content of *Heterosporium spp.* in chilled pork chops based on structured light hyperspectral imaging technology. This method can rapidly, accurately, and non-destructively detect the *Heterosporium spp.* content in chilled pork chops, thus meeting the non-destructive testing requirements for microbial indicators in chilled meat products.

[0009] To achieve the above objectives, the present invention provides the following solution:

[0010] This invention proposes a method for detecting the content of *Heat-killing Osmotherium* in chilled fresh pork chops based on structured light hyperspectral imaging technology, comprising the following steps:

[0011] Step 1: Select chilled pork chops that are similar in size and uniform in thickness;

[0012] Step 2: Soak the pre-sterilized chilled pork chops in a suspension of heat-killing bacteria.

[0013] Step 3: Acquire three phases of structured light hyperspectral images of chilled pork chops from 0 to 9 days; calibrate the acquired structured light hyperspectral images to obtain standard images;

[0014] Step four: After step three is completed, the heat-killed Solanum bacteria content of the chilled pork chops is determined.

[0015] Step 5: Demodulate the corrected structured light hyperspectral image to obtain AC and DC images;

[0016] Step 6: Extract full-band spectral information from the structured light hyperspectral image and establish a prediction model for the content of *Heterospora thermophila* based on the full-band spectral information.

[0017] Step 7: Based on Step 6, further screen characteristic wavelengths, optimize the model, and establish a predictive model for the content of *Heterospora thermophila* based on characteristic wavelengths;

[0018] Step 8: Use the established prediction model to detect the content of *Synthia spp.* in chilled pork chops.

[0019] The diameter of chilled pork chops is approximately 10cm and the height is approximately 1cm.

[0020] The pre-sterilization conditions for chilled pork chops were: wiping with 75% alcohol and irradiation with ultraviolet light for 15 minutes; the concentration of the heat-killing *Synthia spp.* bacterial suspension was 10. 3 -10 4 CFU / mL.

[0021] The structured light hyperspectral imaging system consists of a visible-near-infrared hyperspectral imaging system and a DLi CEL5500 DMD digital projector. The DLi CEL5500 digital projector is used to generate the structured light projection mode, and the hyperspectral imaging system is used to acquire hyperspectral images under the structured light mode. The system's light source is set to 250W, the projector's angle of incidence relative to the vertical is 15°, the distance between the light source and the camera lens and the sample is 12.0cm and 24.0cm, respectively, and the camera's exposure time is set to 29ms. Because the optimal wavelength range of the DLiCEL5500 digital projector is 420-700nm, the wavelength range extracted from the structured light hyperspectral images is 420-700nm.

[0022] The culture medium used was STAA agar medium.

[0023] The spatial frequency domain used is F40:40m. -1 The images of the three phases in the spatial frequency domain are demodulated into AC and DC images according to different formulas.

[0024] ENVI 5.1 was used to extract the spectral reflectance of chilled pork chops and perform image segmentation. The average spectral reflectance of the target region was used as the spectral information of the chilled pork chops. The modeling methods used included partial least squares (PLS) and support vector machine (SVM). The preprocessing methods used included standard normal variate (SNV), multiplicative scatter correction (MSC), orthogonal signal correction (OSC), smoothing, and first derivative (1st).

[0025] Feature wavelength selection methods include the Successive projections algorithm (SPA) and the Competitive adaptive reweighted sampling (CARS).

[0026] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0027] This invention applies structured light hyperspectral imaging technology to the detection of *Heterostilbene* content in chilled pork chops, enabling rapid, accurate, and non-destructive detection of *Heterostilbene* content, thus providing assistance for the non-destructive detection of microbial indicators in chilled meat products. Attached Figure Description

[0028] Figure 1 This is a flowchart of a method for detecting the content of *Heat-killing bacteria* in chilled pork chops based on structured light hyperspectral imaging technology.

[0029] Figure 2 DC and AC images of chilled pork chop samples.

[0030] Figure 3 The DC reflectance spectrum (a) and AC reflectance spectrum (b) of chilled pork chops are shown.

[0031] Figure 4The optimal modeling result for chilled pork chops across the entire spectrum.

[0032] Figure 5 SPA screening results for spectral information in the 420-700nm band of DC images.

[0033] Figure 6 SPA screening results for spectral information in the 420-700nm band of AC images.

[0034] Figure 7 The optimal modeling results for the characteristic wavelength selection of chilled pork ribs for SPA are shown in the figure.

[0035] Figure 8 CARS filtering results for spectral information in the 420-700nm band of DC images.

[0036] Figure 9 CARS filtering results for spectral information in the 420-700nm band of AC images.

[0037] Figure 10 The optimal modeling results for CARS screening of characteristic wavelengths for chilled pork chops are shown in the figure. Detailed Implementation

[0038] The technical solution of the present invention will be further described in detail with reference to the following specific examples.

[0039] The purpose of this invention is to provide a method for detecting the content of *Heterosporium spp.* in chilled pork chops based on structured light hyperspectral imaging technology. This method can quickly, accurately, and non-destructively detect the content of *Heterosporium spp.* in chilled pork chops, thereby meeting the non-destructive testing requirements for microbial indicators in chilled meat products.

[0040] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0041] like Figure 1 As shown, this invention provides a method for detecting the content of *Heterobacter pyridostigma* in chilled fresh pork chops based on structured light hyperspectral imaging technology, comprising the following steps:

[0042] Step 1: Select chilled pork chops that are similar in size and uniform in thickness;

[0043] Step 2: Soak the pre-sterilized chilled pork chops in a suspension of heat-killing bacteria.

[0044] Step 3: Acquire three phases of structured light hyperspectral images of chilled pork chops from 0 to 9 days; calibrate the acquired structured light hyperspectral images to obtain standard images;

[0045] Step four: After step three is completed, the heat-killed Solanum bacteria content of the chilled pork chops is determined.

[0046] Step 5: Demodulate the corrected structured light hyperspectral image to obtain AC and DC images;

[0047] Step 6: Extract full-band spectral information from the structured light hyperspectral image and establish a prediction model for the content of *Heterospora thermophila* based on the full-band spectral information.

[0048] Step 7: Based on Step 6, further screen characteristic wavelengths, optimize the model, and establish a predictive model for the content of *Heterospora thermophila* based on characteristic wavelengths;

[0049] Step 8: Use the established prediction model to detect the content of *Synthia spp.* in chilled pork chops.

[0050] Step one specifically includes:

[0051] In a sterile laminar flow hood, chilled pork chops are evenly cut into round pieces with a diameter of about 10cm and a height of about 1cm.

[0052] Step two specifically includes:

[0053] (1) The fully activated *Heterospora thermophila* was diluted with 0.85% sterile physiological saline to prepare a *Heterospora thermophila* suspension, and the concentration of the suspension was adjusted to 10. 3 -10 4 CFU / mL.

[0054] (2) Wipe with 75% alcohol and sterilize under ultraviolet light for 15 min. Soak the pre-sterilized chilled pork chop sample in heat-killing bacteria suspension for 15 s and store at 4℃.

[0055] Step three specifically includes:

[0056] (1) Acquisition of structured light hyperspectral images: The structured light hyperspectral imaging system is composed of a visible-near-infrared hyperspectral imaging system and a DLi CEL5500 DMD digital projector. The DLi CEL5500 digital projector is used to generate the structured light projection mode, and the hyperspectral imaging system is used to acquire hyperspectral images under the structured light mode. The light source of the system is set to 250W, the projector is at a vertical incident angle of 15°, the light source and the camera lens are 12.0cm and 24.0cm away from the sample, respectively, and the camera exposure time is set to 29ms. Because the optimal wavelength range of the DLi CEL5500 digital projector is 420-700nm, the wavelength range extracted from the structured light hyperspectral images is 420-700nm. The three phases are spatial frequency domain F40:40m.-1 Below 0、

[0057] (2) Hyperspectral Image Correction: Before acquiring hyperspectral images, dark correction and white correction are required to reduce the thickness differences between samples, weaken the camera's dark current effect, and reduce the influence of ambient light on the hyperspectral images. Dark correction uses a dark cap with very low reflectivity to the camera lens, while white correction uses a white polytetrafluoroethylene (PTFE) plate. The corrected reflectance image is then calculated using the following formula:

[0058]

[0059] Where R is the corrected reflectance image, I0 is the initial reflectance image, and B... r It is a dark reference image, I r It is a white reference image, which, after image correction, can be used for subsequent extraction and modeling of spectral values.

[0060] Step four specifically includes:

[0061] Following the procedure outlined in GB / T 4789.2-2016 "National Food Safety Standard - Microbiological Examination of Food: Determination of Total Colony Count", 10g of chilled pork ribs were aseptically cut at appropriate refrigeration intervals. The ribs were then minced using sterilized scissors and placed in a sterile homogenizing bag containing 90mL of 0.85% physiological saline. The bag was then placed in a shaking incubator and shaken for 10 minutes. Multiple dilutions were sequentially applied at a 1:9 ratio. Three suitable dilutions were homogenized, and 0.1mL of each was transferred to cooled, solidified STAA agar medium. The bacterial suspension was then evenly spread onto the medium using a spreader, one spreader per dilution. After spreading, the medium was incubated at 30±1℃ for 48 hours before bacterial counting.

[0062] Step five specifically includes:

[0063] The corrected structured light hyperspectral image was demodulated using the following two formulas to obtain the following results: Figure 2 The AC and DC images shown are as follows:

[0064]

[0065]

[0066] Step six specifically includes:

[0067] (1) Extraction of full-band spectral information from structured light hyperspectral images: The spectral reflectance of chilled pork chop samples was extracted and the images were segmented using the image processing software ENVI 5.1. First, the corrected hyperspectral image of the target sample was opened. Then, the region of interest was selected using ENVI 5.1 software, and the spectral reflectance of the target region was extracted. Finally, the average spectral reflectance of the target region was used as the spectral information of the chilled pork chop sample. The extracted spectral information is shown below. Figure 3 As shown.

[0068] (2) Prediction model for *Heat-killing bacteria* content in chilled pork chops: Before modeling, all samples were divided into a calibration set and a prediction set. One sample out of every four samples was selected for the prediction set, resulting in a final ratio of 3:1 between the calibration and prediction sets. It was also important to ensure that the variation range of *Heat-killing bacteria* content in the calibration set covered the variation range of the prediction set. The modeling methods used included partial least squares (PLS) and support vector machine (SVM). The preprocessing methods used included standard normal variate (SNV), multiplicative scatter correction (MSC), orthogonal signal correction (OSC), smoothing, and first derivative (1st). The results showed that in the 420-700 nm wavelength range, the DC component SNV-SVM model and the AC component 1st-PLS model had better performance, and their respective prediction coefficients (R0) were significantly higher. p 2 The R² values ​​are 0.922 and 0.928, respectively, and the root mean square error of prediction (RMSEP) are 0.455 log CFU / g and 0.433 log CFU / g, respectively. Comparing the R² values ​​of the two models... p 2 The RMSEP and other metrics show that the modeling results for the AC component are slightly better than those for the DC component. Furthermore, the relative percent deviation (RPD) of the model built using the AC component reaches 3.450, which is greater than 3.0, indicating excellent predictive ability. The RPD of the model built using the DC component is 3.292, which is lower than that of the AC component. The optimal modeling result is shown in the figure below. Figure 4As shown, the model established using spectral information in the 420-700nm band of the AC component has good predictive performance and can be used for non-destructive detection of the heat-killing bacteria content in chilled pork chops.

[0069] Step seven specifically includes:

[0070] (1) Extraction of characteristic wavelengths based on SPA algorithm: The SPA algorithm was used to screen the characteristic wavelengths in the 420-700nm band of the DC component of chilled pork ribs samples. Figure 5 This study examines the changes in root mean square error (RMSE) with the increase of the number of feature variables and the distribution of these feature variables during the SPA screening process. A smaller RMSE indicates higher model accuracy. Figure 5 (a) It can be seen that when the number of characteristic wavelengths is greater than 12, the root mean square error gradually tends to stabilize. Therefore, 12 characteristic wavelengths are selected as the output result. The distribution of specific characteristic variables is as follows: Figure 5 As shown in (b), the 12 characteristic wavelengths finally selected are 422, 423, 461, 465, 480, 498, 507, 543, 557, 576, 632, and 692 nm. Figure 6 This study investigates the changes in root mean square error (RMSE) and the distribution of characteristic variables as the number of characteristic variables increases during the screening of AC components in the 420-700 nm band for chilled pork rib samples using SPA. Figure 6 (a) It can be seen that when the number of characteristic wavelengths is greater than 10, the root mean square error gradually tends to stabilize. Therefore, 10 characteristic wavelengths are also selected as the output result. The distribution of specific characteristic variables is as follows: Figure 6 As shown in (b), the final 10 selected characteristic wavelengths are 423, 426, 428, 431, 434, 439, 447, 577, 642, and 668 nm. The optimal modeling result is as follows: Figure 7 As shown.

[0071] (2) Extraction of characteristic wavelengths based on CARS algorithm: CARS was used to screen the characteristic wavelengths in the 420-700nm band of the DC component of chilled pork ribs samples. Figure 8 In the diagram, (a), (b), and (c) represent the changes in sampling variance, root mean square error of cross-validation, and regression coefficient path as the number of sampling runs increases, respectively. For example... Figure 8 As shown in (a), the number of sampling variations shows a decreasing trend as the number of sampling runs increases. Figure 8 (b) shows that the root mean square error of cross-validation exhibits a trend of first decreasing slowly and then increasing rapidly with the increase of the number of sampling runs. This indicates that during the entire sampling process, after removing all irrelevant variables, the removal of effective variables begins. The optimal point is when the root mean square error of cross-validation reaches its minimum value, signifying that all irrelevant variables have been removed. Figure 8 (c) It can be seen that when the number of sampling runs reaches 25, the 21 variables retained are the characteristic wavelengths to be screened, which are 428, 435, 450, 454, 456, 460, 473, 490, 500, 511, 512, 521, 532, 539, 542, 543, 559, 560, 606, 634, and 692 nm respectively. Figure 9 In the figures (a), (b), and (c), the changes in sampling variance, root mean square error of cross-validation, and regression coefficient path as the number of sampling runs increased during the screening of characteristic wavelengths in the 420-700nm band of the AC component for chilled pork chops using CARS are shown. Similarly, Figure 9 (a) indicates that as the number of sampling runs increases, the sampling variance shows a decreasing trend, while... Figure 9 (b) Determine the optimal point at which all irrelevant variables are removed based on the minimum root mean square error of cross-validation. Figure 9 (c) It can be seen that when the number of sampling runs reaches 29, the 14 retained variables are the selected characteristic wavelengths, which are 420, 431, 449, 450, 454, 457, 483, 523, 525, 576, 577, 581, 627, and 675 nm respectively. The optimal modeling result is as follows: Figure 10 As shown.

[0072] (3) SPA was used to screen characteristic wavelengths for the two components, resulting in 12 and 10 characteristic wavelengths respectively. The SPA-SNV-SVM model for the DC component and the SPA-SNV-PLS model for the AC component showed better predictive performance for *Heterospora thermophila* content. Their respective R... p 2 The values ​​were 0.910 and 0.855, respectively; RMSEP was 0.435 log CFU / g and 0.610 log CFU / g, respectively; and RPD was 3.443 and 2.449, respectively. Characteristic wavelengths for the two components were screened using CARS, yielding 21 and 14 characteristic wavelengths respectively. The CARS-MSC-SVM model for the DC component and the CARS-OSC-SVM model for the AC component showed the best prediction results for *Heterospora thermophila* content, with their respective RMSEP values ​​being 0.435 log CFU / g and 0.610 log CFU / g, respectively. p 2 The values ​​were 0.919 and 0.883, respectively; the RMSEP was 0.413 log CFU / g and 0.508 log CFU / g, respectively; and the RPD was 3.626 and 2.941, respectively. The results indicate that the model established using the characteristic wavelengths of the DC component in the 420-700 nm band has better predictive performance and is more suitable for the non-destructive detection of *Synthia spp.* content in chilled pork chops.

[0073] This invention discloses a method for detecting the content of *O. heat-killing bacterium* in chilled pork chops based on structured light hyperspectral imaging technology. Traditional methods for detecting microbial indicators require extensive preparation and specialized operation, are inefficient, and can even damage the meat being tested, failing to meet the needs of automated factory production and import / export trade. While conventional hyperspectral imaging technology is widely used for non-destructive testing of microbial indicators in chilled pork, its low spectroscopic efficiency makes it difficult to fully utilize the energy of all incident light, and its low resolution improvement efficiency affects prediction accuracy. In contrast, structured light hyperspectral imaging systems combine the resolution enhancement of structured light technology with the 'image-spectrum integration' advantage of hyperspectral technology, showing great potential for non-destructive testing of microbial indicators in chilled meat. However, there is currently no research on detecting microbial indicators in chilled meat based on structured light hyperspectral imaging technology. Therefore, this invention applies structured light hyperspectral imaging technology to the detection of *O. heat-killing bacterium* content in chilled pork chops, thus providing a technical reference for the non-destructive testing of microbial indicators in chilled meat.

[0074] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions claimed by the present invention.

Claims

1. A method for detecting the content of *Heat-killing Osmotherium* in chilled fresh pork chops based on structured light hyperspectral imaging technology, characterized in that, Includes the following steps: Step 1: Select chilled pork chops that are similar in size and uniform in thickness; Step 2: Soak the pre-sterilized chilled pork chops in a suspension of heat-killing bacteria. Step 3: Acquire three phases of structured light hyperspectral images of chilled pork chops from 0 to 9 days; calibrate the acquired structured light hyperspectral images to obtain standard images; Step four: After step three is completed, the heat-killed Solanum bacteria content of the chilled pork chops is determined. Step 5: Demodulate the corrected structured light hyperspectral image to obtain AC and DC images; Step six involves extracting spectral information from the AC or DC images of the structured light hyperspectral image in the 420-700 nm band range to establish a PLS (Polygonella tinctoria) content prediction model. In step six, ENVI 5.1 is used to extract the spectral reflectance of chilled pork chops and perform image segmentation. The average spectral reflectance of the target region is ultimately used as the spectral information of the chilled pork chops. Based on the spectral information from the AC image in the structured light hyperspectral image in the 420-700 nm band range, the first derivative method (1st algorithm) is used for preprocessing to establish a PLS PLS content prediction model. Step 7: Use the established prediction model to detect the content of heat-killing bacteria in chilled pork chops.

2. The method according to claim 1, characterized in that: In step one, the diameter of the chilled pork chop is approximately 10 cm and the height is approximately 1 cm.

3. The method according to claim 1, characterized in that: In step two, the pre-sterilization conditions for the chilled pork chops are: wiping with 75% alcohol and irradiation with ultraviolet light for 15 minutes; the concentration of the heat-killing *Synthia spp.* bacterial suspension is 10. 3 -10 4 CFU / mL.

4. The method according to claim 1, characterized in that: The structured light hyperspectral imaging system used in step three consists of a visible-near-infrared hyperspectral imaging system and a DLi CEL5500 DMD digital projector. The DLi CEL5500 digital projector is used to generate the structured light projection mode, and the hyperspectral imaging system is used to acquire hyperspectral images under the structured light mode. The light source of the system is set to 250 W, the projector's angle of incidence relative to the vertical is 15°, the distance between the light source and the camera lens and the sample is 12.0 cm and 24.0 cm, respectively, and the camera's exposure time is set to 29 ms. Because the optimal wavelength range of the DLi CEL5500 digital projector is 420-700 nm, the wavelength range extracted from the structured light hyperspectral image is 420-700 nm.

5. The method according to claim 1, characterized in that: The selective culture medium used for determining the content of *Heat-killing Fever* in step four is STAA agar medium.

6. The method according to claim 1, characterized in that: The spatial frequency domain used in step five is F40:40m. -1 The images of the three phases in this spatial frequency domain are demodulated into AC and DC images according to different formulas, where: In the formula, , , These represent the three phase-shifted reflection images acquired.

7. The method according to claim 1, characterized in that: In step six, characteristic wavelengths in the 420-700 nm band are screened, and a predictive model for the content of *Heat-killing Fever* is established based on the spectral information of the characteristic wavelengths. The characteristic wavelength selection methods include the continuous projection algorithm SPA and the competitive adaptive reweighting algorithm CARS.

8. The method according to claim 7, characterized in that: In step six, the SPA algorithm is used to extract the characteristic wavelengths in the 420-700 nm band of the DC image in the structured light hyperspectral image, and the SNV algorithm is used for preprocessing to establish the SVM thermocrystal content prediction model.

Images