Meat freshness quality detection system and method based on spectral image detection chip

Abstract

The invention discloses a meat freshness quality detection system and method based on a spectral image detection chip, and relates to the technical field of spectral detection. The method is used for solving the problems of high equipment price, complex operation, low precision and difficulty in wide application in the existing meat freshness detection technology. The invention provides a meat freshness quality detection system. The freshness detection module is used for respectively carrying out preprocessing operations such as averaging, dark background noise reduction, lens shadow correction and light field correction on a meat target spectrum image shot by the spectrum camera, judging meat to which the meat target belongs, and detecting the freshness of the meat target according to different types of freshness requirements; and high-efficiency and low-cost meat freshness detection is realized. The method is suitable for nondestructive meat freshness detection.

Description

Technical Field

[0001] The present invention relates to the technical field of spectral detection, and in particular to a meat freshness quality detection system and method based on a spectral image detection chip. Background Art

[0002] With the rapid development of the meat products industry and consumers paying more and more attention to food quality and safety, they are also increasingly concerned about the freshness of the meat they consume daily.

[0003] Freshness is a comprehensive evaluation of the sensory characteristics, physical and chemical indicators, colony count and other characteristics of meat. It is a key indicator used to characterize the nutritional health, safety and edibility of meat. The conventional measurement method of meat freshness still uses conventional destructive testing methods. In recent years, the demand for non-destructive freshness testing has become increasingly strong. Through the unremitting research of various research institutes, universities and related industries, non-destructive testing methods for meat freshness have also made good progress. At present, they are mainly divided into the following categories: First, a professional spectrometer detection device is used to perform single-point detection of the spectral information of meat. For example, the patent document CN103163083A published on June 19, 2013, discloses a dual-band spectral fusion method and system for multi-parameter detection of fresh meat. This method uses a spectrometer to collect the spectral signal of the meat sample to be tested, and then completes the output of quality information such as freshness by processing and modeling the spectral signal. This method uses the spectral information of meat to complete the freshness detection with high accuracy, but the spectrometer is expensive and the operation process is complicated. It is suitable for scientific research institutes and professional institutions, but it is difficult to meet the needs of daily people. In addition, it can only perform single-point spectral detection on meat, and cannot fully reflect the quality information such as the freshness of meat.

[0004] Secondly, the spectral signal of meat is collected by spectral detection equipment, and the signal is processed to realize the output of quality information such as freshness. For example, the patent document published on March 29, 2019: CN109540805A discloses a meat freshness detection method and device. This method uses a spectral detector to collect spectral data of the meat sample to be tested under a halogen lamp light source, and transfers the data into the constructed freshness algorithm model. Finally, the freshness information of the meat to be tested is output. Compared with the traditional spectrometer detection method, this method has achieved a significant reduction in volume and cost, but the sample to be tested needs to be placed on the spectral detector during the test, and the freshness detection is completed using a halogen lamp light source. There are still problems such as complex operation process, low degree of intelligence, and difficulty for ordinary users to use.

[0005] In addition, there are also portable imaging devices such as mobile phone cameras that collect images of meat samples and extract feature data for analysis to achieve freshness detection. For example, the patent document CN106053468A published on October 26, 2016, discloses a method for detecting the freshness of meat based on a smartphone, which uses a smartphone camera to collect color pictures of meat samples, and uses the RGB information in the image to convert it into L*, a*, and b* values ​​in the LAB color space to assess the freshness quality of the meat. This method uses a mobile phone camera to obtain image data for analysis, and the acquisition device is compact and the portability is greatly improved. However, the information obtained by the mobile phone camera on the meat sample is less close to the effect of human eye observation, and is greatly affected by environmental factors and has a high false detection rate, so it is impossible to achieve accurate meat freshness result output. Summary of the invention

[0006] The present invention is used to solve the problems of expensive equipment, complex operation, low precision and difficulty in achieving widespread application in existing meat freshness detection technology. The present invention uses a spectral camera to achieve spectral imaging with a single shot, and a matching algorithm can achieve efficient, low-cost and accurate non-destructive meat freshness detection.

[0007] To achieve the above object, the present invention provides the following technical solutions: The present invention provides a meat freshness quality detection system based on a spectral image detection chip, the system comprising a power supply module, a spectral image acquisition module, a meat storage bin and a freshness detection module; The spectral image acquisition module includes a spectral camera, a wide-spectrum white light source and a homogenizer; The freshness detection module includes a spectral image preprocessing component, a meat recognition component, and a freshness detection component; The wide-spectrum white light source is used to provide light for the meat targets in the meat storage bin, and the light is evenly distributed under the action of the light homogenizer; The spectral camera is embedded with a spectral image detection chip, which is used to capture the spectral image of the meat target in a uniform light environment and send it to the spectral image preprocessing component; The spectral image preprocessing component is used to preprocess the spectral image and then send it to the meat recognition component; The meat identification component is used to identify the category of the meat target and then send it to the freshness detection component; The freshness detection component is used to detect the freshness of meat targets according to different categories of freshness requirements; The power module is used to provide working power for the spectral image acquisition module and the freshness detection module.

[0008] Furthermore, in a preferred embodiment, the meat recognition component is embedded with a meat target detection algorithm; The specific meat target detection algorithm is: Calculate the boundary in the spectral image by calculating the gradient information in the spectral image; Constructing the position coordinates containing the meat target sample according to the sample center and boundary of the meat target; Segmenting the spectral image of the meat target according to the position coordinates of the meat target sample; The spectral image of the meat target is input into the deep neural network model for training until the category of the meat target is identified.

[0009] Furthermore, in another preferred embodiment, the above gradient information formula is:

[0010] The sample center formula for the meat target above is:

[0011]

[0012] in, G x is the horizontal gradient, G y is the vertical gradient, x i and y i are the horizontal coordinates and vertical coordinates of the boundary between the target sample and the image background, respectively. n represents the total number of boundary points, i Indicates i The index of the boundary points.

[0013] Furthermore, in a preferred embodiment, the above-mentioned freshness detection component is embedded with a freshness prediction method; The freshness prediction method is: The spectral data processing model and freshness prediction model of meat target samples were constructed by machine vision methods such as partial least squares. The spectral data processing model is used to pre-process the meat spectral data; The preprocessing includes spectrum normalization, first-order differential transformation and spectrum differential correction; A freshness prediction model was constructed using TVB-N content as the criterion for dividing meat freshness. The freshness prediction model is used to detect the freshness of pretreated meat spectral data.

[0014] Furthermore, in a preferred embodiment, the above spectrum normalization formula is:

[0015] The above first-order differential transformation formula is:

[0016] The above spectral differential correction formula is: C i = ( S i - b i ) / k i in, I norm is the normalized spectral data, I is the original spectral data, I max and I min represent the maximum and minimum values ​​of the spectral intensity, respectively. I i and I i+1 The wavelengths are λ i and λ i+1 The spectral intensity value at C i is the corrected spectral information, S i For the i The first-order differential spectral information of meat samples, k i It is i The slope coefficient obtained by univariate linear regression of the first-order differential spectral information of the meat samples is: b i For the i The intercept for the meat samples, represents the continuous wavelength points in the spectral data, Represents the wavelength growth amount, which is a constant.

[0017] Furthermore, there is a preferred embodiment, the spectral band of the above-mentioned spectral camera is 350~950nm; The spatial resolution of the above spectral camera is 1200×1600; The field angle H of the spectral camera lens is 150°.

[0018] Furthermore, in a preferred embodiment, the wavelength range of the above-mentioned wide-spectrum white light source is 350-950nm.

[0019] Furthermore, in a preferred embodiment, the color rendering index of the above-mentioned broad-spectrum white light source is greater than 95.

[0020] Furthermore, in a preferred embodiment, the light diffuser plate is made of white light diffuser material.

[0021] The present invention also provides a detection method implemented by the meat freshness quality detection system based on the spectral image detection chip according to any one of the above, the method comprising the following steps: Step 1: Preprocess the meat spectral image, including averaging multiple images, subtracting dark background noise, lens shading correction and light field correction; Step 2: Input the preprocessed spectral image into the meat target detection algorithm, and the meat target detection algorithm determines the category of the meat target sample; Step 3: Input the identified meat target sample into the freshness prediction method to obtain the freshness information result.

[0022] The existing meat freshness detection application solution adopts a single spectral data fitting technology and a single deep neural network recognition solution, which can only realize the freshness detection of a single meat. Compared with the existing technology, the advantages of the present invention are: 1. The meat freshness quality detection system proposed in the present invention uses a freshness detection module to perform pre-processing operations such as averaging multiple spectral images of meat targets taken by a spectral camera, subtracting dark background noise, lens shadow correction, and light field correction, and distinguish the meat to which it belongs. It also detects the freshness of meat targets according to the freshness requirements of different categories, thereby achieving efficient and low-cost meat freshness detection.

[0023] Furthermore, compared with the prior art, the present invention performs multiple preprocessing operations on the meat target spectral image, thereby eliminating the need to place the meat target sample to be tested at a designated position or on a spectral detector, and also eliminating the need to process the meat sample to be tested, such as segmentation, thereby achieving real-time, non-destructive meat freshness detection.

[0024] Furthermore, the present invention is combined with a snapshot imaging spectral camera, which can quickly and effectively obtain the spectral information and image information of the target meat, ensuring the real-time and reliability of the detection task.

[0025] Furthermore, the present invention uses spectral data and spectral image data as data basis, combines artificial neural network and artificial intelligence technology, realizes the type recognition of common meat such as pork, cattle, sheep, etc., and realizes the corresponding meat freshness detection.

[0026] The invention is suitable for non-destructive meat freshness detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0028] Figure 1 It is a schematic diagram of the meat freshness quality detection system based on the spectral image detection chip of the present invention; Figure 2 It is a connection diagram of the meat freshness quality detection system based on the spectral image detection chip described in the present invention.

[0029] Among them, 1 represents the spectral camera, 2 represents the lens, 3 represents the heating and demisting wire equipment, 4 represents the light homogenizing plate, 5 represents the meat freshness detection module, 6 represents the meat storage bin, 7 represents the white light source, 8 represents the power module, 9 represents the meat storage bin door, and 10 represents the meat freshness detection terminal. DETAILED DESCRIPTION

[0030] In the following description, specific details such as specific system structures and technologies are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0031] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings. The following implementation will help those skilled in the art to further understand the present invention, but will not limit the present invention in any form.

[0032] Implementation method 1, see Figure 1 This embodiment is described. In order to realize efficient, low-cost and accurate non-destructive meat freshness detection, a meat freshness quality detection system based on a spectral image detection chip is proposed. The system includes a power module, a spectral image acquisition module, a meat storage bin and a freshness detection module. The spectral image acquisition module includes a spectral camera, a wide-spectrum white light source and a light homogenizer; The freshness detection module includes a spectral image preprocessing component, a meat recognition component and a freshness detection component; The broad-spectrum white light source is used to provide light for the meat target in the meat storage bin, and the light is evenly distributed under the action of the light homogenizing plate; The spectral camera is embedded with a spectral image detection chip, which is used to capture the spectral image of the meat target in a uniform light environment and send it to the spectral image preprocessing component; The spectral image preprocessing component is used to preprocess the spectral image and then send it to the meat recognition component; The meat identification component is used to identify the category of the meat target and then send it to the freshness detection component; The freshness detection component is used to detect the freshness of the meat target according to the freshness requirements of different categories; The power supply module is used to provide working power for the spectral image acquisition module and the freshness detection module.

[0033] In practical application, if Figure 1 As shown, the power supply module is a regulated power supply, which is used to provide a stable working power environment for the spectral image acquisition module and the freshness detection module.

[0034] The spectral image acquisition module includes a spectral camera, a wide-spectrum white light source, and a light homogenizer; The spectral camera is used to collect spectral images of target meat samples in the meat storage bin. The spectral band of the spectral camera is 350~950nm, which can cover the ultraviolet to near-infrared band, and can effectively detect the image information and spectral information of meat at different freshness without processing the samples to be tested. The spatial resolution of the spectral camera is 1200×1600, which can effectively ensure the imaging quality of the meat target sample image. The field of view angle H direction of the spectral camera lens is 150°, which can ensure that the camera field of view covers the bottom of the meat storage bin, and ensure that the meat samples placed in the meat storage bin can collect complete spectral image data. A wide-spectrum white light source is connected to the bottom of the spectral camera, and the spectral camera and the wide-spectrum white light source are powered by the same working motherboard.

[0035] Furthermore, a spectral image detection chip is embedded in the spectral camera, and the spectral image detection chip mainly has the following functions: 1. Spectral data acquisition: The spectral image detection chip is responsible for collecting the spectral information of meat targets at different wavelengths. It can capture spectral data and spatial image information at the same time to generate spectral images with high spectral resolution. 2. High spectral resolution: Spectral image detection chips usually have high spectral resolution and can cover a wide range of wavelengths (such as visible light, near infrared, etc.). This enables the system to obtain the spectral characteristics of meat targets at different wavelengths, providing rich data support for subsequent freshness detection. 3. Real-time data processing: The spectral image detection chip can process the collected spectral data in real time and generate spectral images. These spectral images are then sent to the spectral image preprocessing component for further processing and analysis. 4. High sensitivity and low noise: Spectral image detection chips usually have the characteristics of high sensitivity and low noise, and can accurately capture weak spectral signals, which is particularly important for detecting the freshness of meat, because the spectral characteristics related to freshness may be very subtle.

[0036] The wide-spectrum white light source is used to provide a stable and controllable light source environment for the spectral camera. The light source has a wavelength range of 350-950nm, is a wide-spectrum white light source, and has a color rendering index of more than 95, which can effectively meet the spectral detection requirements of freshness detection.

[0037] The light homogenizing plate is used to ensure the uniformity of the light source of the freshness detection system. It is made of white light homogenizing material, with a relatively stable overall transmittance from ultraviolet to near-infrared bands, with little change in the line shape of the light source, and can effectively play the role of uniform light, so that the light distribution in each area inside the meat storage warehouse is uniform.

[0038] The meat storage bin is a specific application scenario of the freshness detection system, including but not limited to meat food storage bins in refrigerators and freezers.

[0039] The freshness detection module includes a spectral image preprocessing component, a meat recognition component and a freshness detection component.

[0040] The spectral image preprocessing component is used to perform preprocessing operations on the spectral images taken by the spectral camera. The spectral image preprocessing operations include multiple averaging, dark background noise reduction, lens shadow correction, light field correction and other processes. The purpose is to effectively reduce the random noise in the spectral image and improve the signal-to-noise ratio of the spectral image.

[0041] The meat recognition component is used to identify the category of the target meat sample, specifically including common meats such as pork, beef, and lamb, and its identification method is based on a meat target detection algorithm.

[0042] The freshness detection component is used to judge the freshness of the meat sample identified by the meat identification system, and its judgment method is based on the meat freshness detection algorithm.

[0043] The meat freshness quality detection system proposed in this embodiment uses a freshness detection module to perform preprocessing operations such as averaging multiple spectral images of meat targets taken by a spectral camera, subtracting dark background noise, lens shading correction, and light field correction, and identify the meat to which it belongs. It also detects the freshness of meat targets according to freshness requirements of different categories, thereby achieving efficient and low-cost meat freshness detection.

[0044] Furthermore, this embodiment performs multiple preprocessing operations on the meat target spectral image, thereby eliminating the need to place the target meat sample to be tested at a designated location or on a spectral detector, and also eliminating the need to process the meat sample to be tested, such as segmentation, thereby achieving real-time, non-destructive meat freshness detection.

[0045] Embodiment 2: This embodiment is a specific description of the meat identification component described in the above embodiment 1; The meat identification component is embedded with a meat target detection algorithm for determining the category to which the target meat sample belongs.

[0046] Among them, the meat target detection algorithm is specifically as follows: The meat target detection algorithm calculates the boundary in the image by calculating the gradient information in the spectral image, and calculates the boundary according to the sample center of the meat target. And the calculated boundary construction contains the location coordinates of the meat target sample ,in represents the coordinates of the upper left corner of the bounding box, Indicates the coordinates of the lower right corner of the bounding box. The spectral image of the meat target is segmented using the position coordinate information of the meat, including: The gradient calculation method of the meat target: ,in, G x is the horizontal gradient, G y is the vertical gradient. The center of the target sample is calculated as: , .in, x i and y i are the horizontal and vertical coordinates of the boundary between the target sample and the image background, respectively. n represents the total number of boundary points, i Indicates i The index of the boundary points.

[0047] This embodiment also loads the spectral image of the meat target obtained in the previous process into the deep neural network model for training, realizes the identification of common meat types such as pork, cattle, and sheep, and inputs the meat spectral image data that has completed the meat type identification into the freshness detection component. During the meat type identification process, in order to enable the neural network to fully learn the characteristic differences of different meat types, the spectral image data set of 350nm-950nm is used as the model input. The specific steps include: constructing a spectral image input of 5 spectral bands, including but not limited to spectral band images with spectral bands of 450nm, 500nm, 650nm, 850nm and 900nm as the central wavelengths. The specific wavelength range can be defined according to the characteristic differences of different meat band images.

[0048] Implementation method 3: This implementation method is to specifically explain the freshness detection component described in the above implementation method 1; The freshness detection component is embedded with a freshness prediction method for realizing freshness detection of meat targets.

[0049] The specific freshness prediction method is: The spectral data processing model and freshness prediction model of meat target samples were constructed by machine vision methods such as partial least squares. The spectral data processing model is used to pre-process the meat spectral data; The preprocessing includes spectrum normalization, first-order differential transformation and spectrum differential correction; A freshness prediction model was constructed using TVB-N content as the criterion for dividing meat freshness. The freshness prediction model is used to detect the freshness of pretreated meat spectral data.

[0050] This embodiment realizes the freshness detection of different meats by constructing a meat freshness prediction model, and uses TVB-N content as the classification standard for meat freshness. According to the regulations, TVB-N>15mg / 100g is not fresh; TVB-N content between 10-15mg / 100g is sub-fresh meat; TVB-N<10mg / 100g is fresh meat. The spectral data of the target meat sample and the TVB-N freshness prediction model are constructed through machine vision solutions such as partial least squares. Specifically, the TVB-N value and the freshness status of the meat are predicted based on the meat spectral data.

[0051] Further, according to the meat freshness prediction method, the meat spectral data is preprocessed, including but not limited to spectral normalization, first-order differential transformation and spectral differential correction; The spectrum normalization is calculated as: .

[0052] in,I norm is the normalized spectral data, I is the original spectral data, I max and I min represent the maximum and minimum values ​​of the spectral intensity, respectively.

[0053] The first-order differential transformation calculation method is: .

[0054] in, I i and I i+1 The wavelengths are λ i and λ i+1 The spectral intensity value at .

[0055] The spectrum differential correction calculation method is: C i = ( S i - b i ) / k i .

[0056] in, C i is the corrected spectral information, S i For the i The first-order differential spectral information of meat samples, k i It is i The slope coefficient obtained by univariate linear regression of the first-order differential spectral information of the meat samples is: b i For the i The intercept for the meat samples, represents the continuous wavelength points in the spectral data, Represents the wavelength growth amount, which is a constant.

[0057] Implementation method 4: See Figure 2 This embodiment describes a detection method implemented by a meat freshness quality detection system based on a spectral image detection chip, and the method comprises the following steps: Step 1: Preprocess the meat spectral image, including averaging multiple images, subtracting dark background noise, lens shading correction and light field correction; Step 2: Input the preprocessed spectral image into the meat target detection algorithm, and the meat target detection algorithm determines the category of the meat target sample; Step 3: Input the identified meat target sample into the freshness prediction method to obtain the freshness information result.

[0058] In practical application, this embodiment is Figure 2 The connection is made as shown, the meat storage bin door 9 of the meat storage bin 6 is opened, the heating demisting wire device 3 is used to thaw the frozen meat, and the meat sample placed in the meat storage bin 6 is irradiated with a wide-spectrum white light source 7. Under the action of the light homogenizing plate 4, the lens 2 of the spectral camera 1 collects the spectral image data of the meat sample, and after the spectral image data is preprocessed by the built-in computing unit of the spectral camera, it is sent to the meat freshness detection terminal 10 for freshness detection through the meat freshness detection module 5. At the same time, the power supply module 8 is used as the working power supply to power the above-mentioned electrical equipment.

[0059] The light source has a wavelength range of 350-950nm, is a wide-spectrum white light source, and has a color rendering index of more than 95. It can meet the spectrum detection requirements of freshness detection, and at the same time, it can be used as an illumination light source to cooperate with the spectral camera to output color images for users to observe the freshness of the target meat.

[0060] The light homogenizing plate 4 is made of white light homogenizing material, and has a relatively stable overall transmittance from ultraviolet to near-infrared bands, has little effect on the linear shape of the light source, and can play a role in uniform light.

[0061] The multi-spectral camera is a CM series spectral camera, and other small-volume staring spectral cameras may also be used. The spectral coverage band is 350~950nm, covering the ultraviolet to near-infrared band, and can output color images of meat samples inside the drawer for users to observe.

[0062] The specific workflow is as follows: Light up the broad-spectrum white light source: the spectral camera is powered and turned on, the broad-spectrum white light source is connected to the spectral camera, and after the camera is turned on, the light source automatically lights up and illuminates the meat sample at the bottom of the drawer.

[0063] Spectral image acquisition: The spectral camera collects spectral images of meat samples by adjusting the exposure time through the automatic AE module.

[0064] Spectral image preprocessing: Perform preprocessing operations such as averaging multiple images, removing background noise, and multiplying the lens shading coefficient on the collected spectral images to obtain spectral images of meat samples with a high signal-to-noise ratio.

[0065] The meat target detection algorithm is used to identify the category of the meat target. First, the coordinate position of the target meat sample in the spectral image is calculated to identify the category of the target meat sample, including common meat such as pork, beef, and lamb. Target meat coordinate calculation. This embodiment calculates the gradient of the spectral data collected by the meat freshness detection system, including: ,in, G x is the horizontal gradient, G y is the vertical gradient.

[0066] Eliminate interference edges through non-maximum suppression, calculate all edges of the target meat sample, and retain the target meat sample and background edges. Obtain the pixel center of the target meat sample based on the target meat edge ,include: , , and are the horizontal and vertical coordinates of the boundary between the target sample and the image background, , They are the horizontal and vertical coordinates of the meat samples respectively.

[0067] Combine the target meat and background edges to construct the target class in the spectral image data coordinates .

[0068] Identify the target meat type, obtain the spectral image of the target meat according to the target meat coordinates, and construct a spectral band image. The specific spectral band image needs to be selected reasonably, and the spectral bands with large differences in the characteristics of different meats in the corresponding spectral bands are screened. The present embodiment constructs band images including but not limited to 450nm, 500nm, 650nm, 850nm, and 900nm as central wavelengths. Specifically, the band range can be clarified according to the characteristic differences of different meat samples in different band ranges. For example, if the target meat has a significant difference in characteristics from other meats at 700nm, the present embodiment constructs a band image with 700nm as the central wavelength as the model input. This embodiment uses 5-band spectral images as the data basis and uses a deep neural network model to identify the category of meat target samples, including common meats such as pork, cattle, and sheep.

[0069] The freshness prediction method is used to detect the freshness of meat targets, including: meat freshness calibration, volatile basic nitrogen-TVBN data regression, and common meat freshness prediction.

[0070] Among them, the calibration of meat freshness: Volatile basic nitrogen-TVBN can regularly indicate the changes in meat freshness in the evaluation of meat freshness, and is an important physical and chemical indicator in the detection of meat food freshness. The spectral image data and spectral data in the database are used to establish a freshness detection algorithm model. The meat freshness data is completed in the freshness detection module. The spectral image acquisition is completed. After the test is completed, the volatile basic nitrogen-TVBN value test is performed, and the meat samples are divided into three grades of fresh, sub-fresh, and corrupt according to the TVBN grading standard to label the meat spectral image.

[0071] TVBN data regression: This implementation method uses the spectral data of the target meat sample as the data basis, pre-processes the spectral data through normalization, first-order differential transformation and other methods, and constructs the spectral characteristic data of the meat to be tested. Based on the meat TVBN value as the reference value of the spectral characteristic data, a multivariate regression model is established by minimizing the covariance between the spectral data and the TVBN value. The predicted value of TVBN is obtained based on the prediction results of the multivariate regression. Referring to the predicted value and true value of TVBN, the freshness of common meats such as pork, beef, and lamb is judged.

[0072] According to the execution mode of the above embodiment, the freshness detection and quality analysis can be performed on different meats, the freshness level of different pixel points can be judged and finally statistics and analysis can be performed to output the freshness detection result of the sample. The meat spectral image acquisition and freshness and quality information result output can be completed, and eating suggestions can be provided to users.

[0073] It should be noted that the above implementation is the whole process of meat freshness quality detection. The execution process in this implementation is not fixed. According to the characteristics of different meat spectral data, different spectral spectral data preprocessing methods can be used to make corresponding changes to the execution process of meat freshness.

[0074] The above description is only an embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may be modified and varied in various ways.

Claims

1. The meat freshness quality detection system based on spectral image detection chip is characterized by: The system includes a power module, a spectral image acquisition module, a meat storage bin and a freshness detection module; The spectral image acquisition module includes a spectral camera, a wide-spectrum white light source and a homogenizer; The freshness detection module includes a spectral image preprocessing component, a meat recognition component, and a freshness detection component; The wide-spectrum white light source is used to provide light for the meat targets in the meat storage bin, and the light is evenly distributed under the action of the light homogenizer; The spectral camera is embedded with a spectral image detection chip, which is used to capture the spectral image of the meat target in a uniform light environment and send it to the spectral image preprocessing component; The spectral image preprocessing component is used to preprocess the spectral image and then send it to the meat recognition component; The meat identification component is used to identify the category of the meat target and then send it to the freshness detection component; The freshness detection component is used to detect the freshness of meat targets according to different categories of freshness requirements; The power module is used to provide working power for the spectral image acquisition module and the freshness detection module.

2. The meat freshness quality detection system based on spectral image detection chip according to claim 1 is characterized in that: The meat recognition component has a meat object detection algorithm embedded in it; The specific meat target detection algorithm is: Calculate the boundary in the spectral image by calculating the gradient information in the spectral image; Constructing the position coordinates containing the meat target sample according to the sample center and boundary of the meat target; Segmenting the spectral image of the meat target according to the position coordinates of the meat target sample; The spectral image of the meat target is input into the deep neural network model for training until the category of the meat target is identified.

3. The meat freshness quality detection system based on spectral image detection chip according to claim 2 is characterized in that: The gradient information formula is: The sample center formula for the meat target is: in, G x is the horizontal gradient, G y is the vertical gradient, and are the horizontal coordinates and vertical coordinates of the boundary between the target sample and the image background, respectively. n represents the total number of boundary points, i Indicates i The index of the boundary points.

4. The meat freshness quality detection system based on spectral image detection chip according to claim 1 is characterized in that: The freshness detection component has a built-in freshness prediction method; The freshness prediction method is: The spectral data processing model and freshness prediction model of meat target samples were constructed by machine vision methods such as partial least squares. The spectral data processing model is used to pre-process the meat spectral data; The preprocessing includes spectrum normalization, first-order differential transformation and spectrum differential correction; A freshness prediction model was constructed using TVB-N content as the criterion for dividing meat freshness. The freshness prediction model is used to detect the freshness of pretreated meat spectral data.

5. The meat freshness quality detection system based on spectral image detection chip according to claim 4 is characterized in that: The spectral normalization formula is: The first-order differential transformation formula is: The spectral differential correction formula is: C i = ( S i - b i ) / k i in, is the normalized spectral data, I is the original spectral data, and represent the maximum and minimum values ​​of the spectral intensity, respectively. and The wavelengths are and The spectral intensity value at C i is the corrected spectral information, S i For the i The first-order differential spectral information of meat samples, k i It is i The slope coefficient obtained by univariate linear regression of the first-order differential spectral information of the meat samples is: b i For the i The intercept for the meat samples, represents the continuous wavelength points in the spectral data, Represents the wavelength growth amount, which is a constant.

6. The meat freshness quality detection system based on spectral image detection chip according to claim 1 is characterized in that: The spectral band of the spectral camera is 350~950nm; The spatial resolution of the spectral camera is 1200 × 1600; The field of view angle H of the spectral camera lens is 150°.

7. The meat freshness quality detection system based on spectral image detection chip according to claim 1 is characterized in that: The wavelength range of the broadband white light source is 350~950nm.

8. The meat freshness quality detection system based on spectral image detection chip according to claim 7 is characterized in that: The color rendering index of the broad spectrum white light source is greater than 95.

9. The meat freshness quality detection system based on spectral image detection chip according to claim 1 is characterized in that: The light diffuser is made of white light diffuser material.

10. The detection method implemented by the meat freshness quality detection system based on spectral image detection chip according to any one of claims 1 to 9, characterized in that: The method is: Step 1: Preprocess the meat spectral image, including averaging multiple images, subtracting dark background noise, lens shading correction and light field correction; Step 2: Input the preprocessed spectral image into the meat target detection algorithm, and the meat target detection algorithm determines the category of the meat target sample; Step 3: Input the identified meat target sample into the freshness prediction method to obtain the freshness information result.

Images