Food production line visual quality inspection method and device based on spectrum technology

Abstract

The invention discloses a food production line visual quality inspection method and device based on a spectrum technology, and relates to the technical field of food production line visual quality inspection. Visual imaging equipment is used for obtaining a food visual picture of an outlet section of a finished product conveying belt of a food production line; the visual analysis module adopts a defect detection algorithm to perform visual analysis on the food visual picture at the inlet section of the finished product conveying belt; the spectral imaging equipment disclosed by the invention realizes plane movement in a certain range in the middle of the finished product conveying belt to realize automatic tracking of the contour center point of a defective part, and obtains spectral data of the defective part of suspected defective food in a relatively static state, so that the accuracy of the spectral data can be greatly improved; the trained PLSR model is used for carrying out defect analysis on the spectral data of the defect part, and whether defects exist or not and a defect classification result can be output more quickly and accurately; automatic quality inspection is carried out on food on a food production line, the false detection and missed detection probability is reduced, and the method can be suitable for different types of food production and processing scenes.

Description

Technical Field

[0001] The present invention relates to the technical field of visual quality inspection of food production lines, and in particular to a method and device for visual quality inspection of food production lines based on spectral technology. Background Art

[0002] The current visual quality inspection technology of food production lines is generally based on machine vision detection technology, which uses cameras, light sources, image processing algorithms, etc. to detect the appearance, shape, color and other characteristics of food. Or through deep learning algorithms such as convolutional neural networks (CNN), a large amount of image data is trained to improve the detection accuracy.

[0003] At present, the above-mentioned visual quality inspection technology of food production lines has poor adaptability to complex scenes and lighting changes, which easily leads to false detection and missed detection. The training cycle of deep learning algorithms is long, requires a large amount of labeled data, and it is difficult to explain the detection results. Summary of the invention

[0004] In order to solve the problem that the above-mentioned visual quality inspection technology of food production line has poor adaptability to complex scenes and lighting changes, which easily leads to false detection and missed detection. The deep learning algorithm has a long training cycle, requires a large amount of labeled data, and is difficult to explain the technical problems of the detection results, the present invention provides a food production line visual quality inspection method and device based on spectral technology. The following technical solutions are adopted: A food production line visual quality inspection method based on spectral technology, comprising the following steps: Step 1, using a visual imaging device to obtain a visual image of food at the exit section of a finished product conveyor belt of a food production line; Step 2: The visual analysis module uses a defect detection algorithm to perform visual analysis on the food visual image at the entrance section of the finished product conveyor belt, mark the contour of the defective part, and calculate the center point of the contour of the defective part; Step 3, control the spectral imaging device installed above the finished product conveyor belt of the food production line to move to the center point of the contour of the defective part, control the speed to be consistent with the finished product conveyor belt speed of the food production line, and track the center point of the contour of the defective part; Step 4: The spectral imaging device performs high-resolution spectral imaging on the defect position to obtain spectral data of the defect position; Step 5: Use the trained PLSR model to perform defect analysis on the spectral data of the defective part, and output whether there is a defect and the defect classification result; Step 6: If it is determined that there is a defect, the sorting robot above the exit section of the finished product conveyor belt is controlled to remove the food at the position corresponding to the defect.

[0005] By adopting the above technical solution, it is different from the single defect detection solution of spectral detection or visual detection used in traditional food production lines. First, a visual imaging device is used to obtain the visual picture of the food at the outlet section of the finished product conveyor belt of the food production line. The visual detection here is only to find out the suspected defective food. The suspected defective food can be a food with different color or shape, or a foreign body that is not a produced food, such as condiments. The defect detection algorithm used can be a color-based or shape-based defect detection algorithm. The corresponding position of the suspected defective food at the outlet section of the finished product conveyor belt of the food production line can be quickly screened out, the defect position can be contoured, and the center point can be found. The center point can be a geometric center point or a mass point. For example, the find Contours function and moments function used in the OpenCV environment can quickly find the center point of the defect part contour. The position of the center point of the defect part contour can be marked with coordinates on the electronic map constructed based on the finished product conveyor belt area. The movement of the spectral imaging device installed above the finished product conveyor belt of the food production line can be in various forms, such as using a mechanical arm, an electric cross slide, etc., which can realize plane movement within a certain range in the middle of the finished product conveyor belt to realize automatic tracking of the center point of the contour of the defective part; Therefore, the spectral imaging device and the suspected defective food on the finished product conveyor belt remain relatively still, and the spectral data of the defective part of the suspected defective food is obtained in a relatively still state, which can greatly improve the accuracy of the spectral data. The defect analysis of the spectral data of the defective part using the trained PLSR model can output the existence of defects and defect classification results more quickly and accurately. Finally, with the help of sorting robots, it can realize the automatic sorting and removal of defective food, condiments and foreign objects; It can efficiently and accurately automate the quality inspection of food on the food production line, reduce the probability of false detection and missed detection, and can be applied to different types of food production and processing scenarios.

[0006] Optionally, the finished product conveyor belt is divided into a finished product conveyor belt entrance section, a spectral tracking detection section and a finished product conveyor belt exit section. Assume that the total length of the finished product conveyor belt is S, the finished product conveyor belt entrance section is one-third of the entrance end of the finished product conveyor belt, the spectral tracking detection section is one-third of the middle part of the finished product conveyor belt, and the finished product conveyor belt exit section is one-third of the exit end of the finished product conveyor belt.

[0007] By adopting the above technical solution, by dividing the finished product conveyor belt into a finished product conveyor belt entrance section, a spectral tracking detection section and a finished product conveyor belt exit section, modular rapid installation and deployment of equipment with different functions can be achieved.

[0008] Optionally, in step 2, the defect detection algorithm is: extracting visual features from the image, inputting the visual features into a defect classifier based on machine vision, and when the defect classifier outputs the visual features as defect features, extracting the contour of the area corresponding to the defect features, and marking them as the defect part contour.

[0009] Optionally, the Canny algorithm is used to extract the edge contour of the area corresponding to the defect feature as the defect part contour, and the center point of each detected defect part contour is calculated as the center point of the defect part contour. The optical flow method or Kalman filter tracking algorithm is used in continuous frames to track the center point of the defect part contour.

[0010] By adopting the above technical solutions, in the early stage of visual analysis, it is also necessary to grayscale, filter, detect edges and other operations on the image in order to better extract contours. Use OpenCV's findContours function to detect contours in the image. For each detected contour, calculate its center point (center of mass). Tracking these center points in consecutive frames, optical flow, Kalman filtering or other tracking algorithms can be used to quickly determine suspected defects on visual analysis chips with lower computing power.

[0011] Optionally, in step 3, the spectral imaging device is installed on the planar moving part of the planar moving mechanism, and is driven by the planar moving part to move in a plane above the spectral tracking and detection section of the finished product conveyor belt. The controller of the planar moving mechanism interacts with the visual analysis module to track the coordinates of the center point of the contour of the defective part, and exchanges the conveying speed data with the finished product conveyor belt controller. The planar moving part first drives the spectral imaging device to move to the top of the center point of the contour of the defective part, and then matches the conveying speed, maintaining it for 1-3 seconds, and controls the spectral imaging device to perform high-resolution spectral imaging of the area below to obtain spectral data of the defective part.

[0012] By adopting the above technical solution, the planar moving mechanism can be an electric cross slide and other equipment, which can drive the spectral imaging device to move in a plane within a certain area, realize automatic tracking of the suspected defect position, and move forward at the same speed as the suspected defect, thereby acquiring spectral data in a relatively static state. The spectral data acquisition is more targeted and can greatly improve the accuracy of subsequent defect judgment based on the spectral data.

[0013] Optionally, the specific steps of using the trained PLSR model in step 5 to perform defect analysis on the spectral data of the defect part are: Step 51, performing normalization preprocessing on the spectral data; Step 52, the goal of the trained PLSR model is to find a set of projection vectors w and c that maximize the covariance of the spectral data X and the attribute Y respectively; Step 53, use the trained PLSR model to perform defect prediction analysis on the defect part spectral data: ; is the predicted attribute value, which includes different pre-classified defect types, X is the spectral data of the defect part, are the regression coefficients of the PLSR model.

[0014] By adopting the above technical solution, the spectral data needs to be preprocessed before analysis, including noise removal, baseline correction, normalization, etc.

[0015] Use PLSR models to relate preprocessed spectral data to known properties such as concentration.

[0016] Select the optimal number of components through cross-validation and train the model. Split the dataset into training and test sets. Apply PLSR to the training set and use cross-validation to select the optimal number of components.

[0017] The trained PLSR model is used to perform defect prediction analysis on the spectral data of defective parts, which can quickly and accurately determine the defect type.

[0018] A food production line visual quality inspection device based on spectral technology, used to implement a food production line visual quality inspection method based on spectral technology, the food production line visual quality inspection device includes a visual imaging device, a visual analysis module, a spectral imaging device, a planar moving mechanism, a chip-based spectral imaging tracking controller and a defect analysis computer; The visual imaging device is installed above the entrance section of the finished product conveyor belt through a bracket, and the visual analysis module is communicatively connected with the visual imaging device; The plane moving mechanism is installed above the spectrum tracking detection section of the finished product conveyor belt through a bracket, the spectrum imaging device is installed on the plane moving part of the plane moving mechanism, and the spectrum imaging tracking controller is connected to the visual analysis module for communication and controls the execution action of the plane moving mechanism; The defect analysis computer is connected to the spectral imaging device for communication, exchanges spectral data of defective parts, and uses the trained PLSR model to perform defect analysis on the spectral data of defective parts, and outputs whether there is a defect and a defect classification result.

[0019] Optionally, a sorting robot is also included, which communicates with the defect analysis computer to exchange visual feature position data of defects, and the sorting robot's sorting arm removes defective food corresponding to the defective visual feature position.

[0020] Optionally, the visual imaging device is a high-definition visual camera, and the spectral imaging device is a spectral camera; The visual analysis module includes a memory and a visual analysis chip. The memory is communicatively connected to the visual imaging device, and the visual analysis chip is communicatively connected to the memory. The visual images of the visual imaging device are collected, and a defect detection algorithm is used to perform a visual analysis on the visual images of food at the entrance section of the finished product conveyor belt. The visual analysis results are stored in the memory, and the spectral imaging tracking controller communicates with the memory to exchange the visual analysis results.

[0021] Optionally, the planar moving mechanism is an electric cross slide.

[0022] In summary, the present invention includes at least one of the following beneficial technical effects: first, a visual imaging device is used to obtain a visual image of food at the outlet section of the finished product conveyor belt of the food production line, and visual inspection only uses a defect detection algorithm to find out suspected defective food, which can be a color- or shape-based defect detection algorithm, and can quickly screen out the corresponding position of the suspected defective food at the outlet section of the finished product conveyor belt of the food production line, contour the defect position, and find the center point, which can be a geometric center point or a mass point. The position of the center point of the defect part contour can be marked with coordinates on an electronic map constructed based on the finished product conveyor belt area; The spectral imaging device installed above the finished product conveyor belt of the food production line can realize plane movement within a certain range in the middle of the finished product conveyor belt to realize automatic tracking of the center point of the defective part contour, and obtain the spectral data of the defective part of the suspected defective food in a relatively static state, which can greatly improve the accuracy of the spectral data. The defect analysis of the spectral data of the defective part can be performed using the trained PLSR model, which can output the existence of defects and defect classification results more quickly and accurately. In conjunction with the sorting robot, it can realize the automatic sorting and removal of defective food; It can efficiently and accurately perform automated quality inspection on food condiments and foreign matter on the food production line, reduce the probability of false detection and missed detection, and can be applied to different types of food production and processing scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a flow chart of a food production line visual quality inspection method based on spectral technology of the present invention; Figure 2 This is a schematic diagram of the connection principle of electrical components of a food production line visual quality inspection device based on spectral technology of the present invention; Figure 3 It is a schematic diagram of the structural principle of a food production line visual quality inspection device based on spectral technology of the present invention.

[0024] Explanation of the accompanying drawings: 1. visual imaging device; 2. visual analysis module; 21. memory; 22. visual analysis chip; 3. spectral imaging device; 4. planar moving mechanism; 5. spectral imaging tracking controller; 6. defect analysis computer; 7. sorting robot; 100. finished product conveyor belt. DETAILED DESCRIPTION

[0025] The present invention is further described in detail below in conjunction with the accompanying drawings.

[0026] The embodiment of the present invention discloses a food production line visual quality inspection method and device based on spectral technology.

[0027] Reference Figure 1-Figure 3 , Embodiment 1, a food production line visual quality inspection method based on spectral technology, comprising the following steps: Step 1, using a visual imaging device 1 to obtain a visual image of food at the exit section of a finished product conveyor belt 100 of a food production line; Step 2, the visual analysis module 2 uses a defect detection algorithm to perform a visual analysis on the food visual image at the entrance section of the finished product conveyor belt 100, marks the contour of the defective part, and calculates the center point of the contour of the defective part; Step 3, control the spectral imaging device 3 installed above the finished product conveyor belt 100 of the food production line to move to the center point of the defective part contour, control the speed to be consistent with the finished product conveyor belt 100 of the food production line, and track the center point of the defective part contour; Step 4, the spectral imaging device 3 performs high-resolution spectral imaging on the defect position to obtain spectral data of the defect position; Step 5: Use the trained PLSR model to perform defect analysis on the spectral data of the defective part, and output whether there is a defect and the defect classification result; Step 6: If it is determined that there is a defect, the sorting robot 7 above the exit section of the finished product conveyor belt 100 is controlled to remove the food at the position corresponding to the defect.

[0028] Different from the single defect detection scheme of spectral detection or visual detection used in traditional food production lines, a visual imaging device 1 is first used to obtain a visual picture of food at the exit section of the finished product conveyor belt 100 of the food production line. The visual detection here is only to find out suspected defective food. Suspected defective food can be food with different colors or shapes, or foreign matter that is not produced food, such as condiments, etc. A defect detection algorithm is used, which can be a color- or shape-based defect detection algorithm. The corresponding position of the suspected defective food at the exit section of the finished product conveyor belt 100 of the food production line can be quickly screened out, the defect position is contoured, and the center point is found. The center point can be a geometric center point or a mass point. For example, the find Contours function and moments function used in the OpenCV environment can quickly find the center point of the defect part contour. The position of the center point of the defect part contour can be marked with coordinates on the electronic map constructed based on the finished product conveyor belt 100 area; The movement of the spectral imaging device 3 installed above the finished product conveyor belt 100 of the food production line can be in various forms, such as using a mechanical arm, an electric cross slide, etc., which can realize plane movement within a certain range in the middle of the finished product conveyor belt 100 to realize automatic tracking of the center point of the contour of the defective part; Thus, the spectral imaging device 3 and the suspected defective food on the finished product conveyor belt 100 remain relatively still, and the spectral data of the defective part of the suspected defective food is obtained in a relatively still state, which can greatly improve the accuracy of the spectral data. The defect analysis of the spectral data of the defective part is performed using the trained PLSR model, and the presence of defects and defect classification results can be output more quickly and accurately. Finally, the sorting robot 7 can be used to automatically sort and remove defective food, condiments, and foreign objects; It can efficiently and accurately automate the quality inspection of food on the food production line, reduce the probability of false detection and missed detection, and can be applied to different types of food production and processing scenarios.

[0029] In Example 2, the finished product conveyor belt 100 is divided into an entrance section of the finished product conveyor belt 100, a spectral tracking detection section and an exit section of the finished product conveyor belt 100. Assuming that the total length of the finished product conveyor belt 100 is S, the entrance section of the finished product conveyor belt 100 is one-third of the entrance end of the finished product conveyor belt 100, the spectral tracking detection section is one-third of the middle part of the finished product conveyor belt 100, and the exit section of the finished product conveyor belt 100 is one-third of the exit end of the finished product conveyor belt 100.

[0030] By dividing the finished product conveyor belt 100 into an entrance section of the finished product conveyor belt 100, a spectrum tracking detection section and an exit section of the finished product conveyor belt 100, modular rapid installation and deployment of equipment with different functions can be achieved.

[0031] In Example 3, in step 2, the defect detection algorithm is: extracting visual features from the image, inputting the visual features into a defect classifier based on machine vision, and when the defect classifier outputs the visual features as defect features, extracting the contour of the area corresponding to the defect features and marking them as the defect part contour.

[0032] In Example 4, the Canny algorithm is used to extract the edge contour of the area corresponding to the defect feature as the defect part contour, and the center point of each detected defect part contour is calculated as the center point of the defect part contour. The optical flow method or Kalman filter tracking algorithm is used in continuous frames to track the center point of the defect part contour.

[0033] In the early stages of visual analysis, the image needs to be grayed, filtered, and edge detected to better extract contours. Use OpenCV's findContours function to detect contours in the image. For each detected contour, calculate its center point (center of mass). Tracking these center points in consecutive frames, optical flow, Kalman filtering, or other tracking algorithms can be used to quickly determine suspected defects on visual analysis chips with lower computing power.

[0034] Example 5, in step 3, the spectral imaging device 3 is installed on the planar moving part of the planar moving mechanism 4, and is driven by the planar moving part to move in a plane above the spectral tracking and detection section of the finished product conveyor belt 100. The controller of the planar moving mechanism 4 interacts with the visual analysis module 2 to track the coordinates of the center point of the contour of the defective part, and interacts with the controller of the finished product conveyor belt 100 to exchange conveying speed data. The planar moving part first drives the spectral imaging device 3 to move to the center point of the contour of the defective part, and then matches the conveying speed, maintaining it for 1-3 seconds, and controls the spectral imaging device 3 to perform high-resolution spectral imaging of the area below to obtain spectral data of the defective part.

[0035] The planar moving mechanism 4 can be an electric cross slide or other equipment, which can drive the spectral imaging device 3 to move in a planar manner within a certain area, realize automatic tracking of the suspected defect position, and move forward at the same speed as the suspected defect, thereby acquiring spectral data in a relatively static state. The spectral data acquisition is more targeted and can greatly improve the accuracy of subsequent defect judgment based on the spectral data.

[0036] In Example 6, the specific steps of using the trained PLSR model in step 5 to perform defect analysis on the defect part spectral data are: Step 51, performing normalization preprocessing on the spectral data; Step 52, the goal of the trained PLSR model is to find a set of projection vectors w and c that maximize the covariance of the spectral data X and the attribute Y respectively; Step 53, use the trained PLSR model to perform defect prediction analysis on the defect part spectral data: ; is the predicted attribute value, which includes different pre-classified defect types, X is the spectral data of the defect part, are the regression coefficients of the PLSR model.

[0037] Before analysis, the spectral data needs to be preprocessed, including noise removal, baseline correction, normalization, etc.

[0038] Use PLSR models to relate preprocessed spectral data to known properties such as concentration.

[0039] Select the optimal number of components through cross-validation and train the model. Split the dataset into training and test sets. Apply PLSR to the training set and use cross-validation to select the optimal number of components.

[0040] The trained PLSR model is used to perform defect prediction analysis on the spectral data of defective parts, which can quickly and accurately determine the defect type.

[0041] Embodiment 7, a food production line visual quality inspection device based on spectral technology, used to implement a food production line visual quality inspection method based on spectral technology, the food production line visual quality inspection device includes a visual imaging device 1, a visual analysis module 2, a spectral imaging device 3, a plane moving mechanism 4, a chip-based spectral imaging tracking controller 5 and a defect analysis computer 6; The visual imaging device 1 is installed above the entrance section of the finished product conveyor belt 100 through a bracket, and the visual analysis module 2 is connected to the visual imaging device 1 for communication; The plane moving mechanism 4 is installed above the spectrum tracking detection section of the finished product conveyor belt 100 through a bracket, the spectrum imaging device 3 is installed on the plane moving part of the plane moving mechanism 4, and the spectrum imaging tracking controller 5 is connected to the visual analysis module 2 for communication and controls the execution action of the plane moving mechanism 4; The defect analysis computer 6 is connected to the spectral imaging device 3 for communication, exchanges spectral data of defective parts, and uses the trained PLSR model to perform defect analysis on the spectral data of defective parts, and outputs whether there is a defect and the defect classification result.

[0042] Embodiment 8 also includes a sorting robot 7, which is connected in communication with the defect analysis computer 6 to exchange visual feature position data of defects, and the sorting robot arm of the sorting robot 7 removes defective foods corresponding to the defective visual feature positions.

[0043] Embodiment 9, the visual imaging device 1 is a high-definition visual camera, and the spectral imaging device 3 is a spectral camera; The visual analysis module 2 includes a memory 21 and a visual analysis chip 22. The memory 21 is communicatively connected to the visual imaging device 1, and the visual analysis chip 22 is communicatively connected to the memory 21. The visual image of the visual imaging device 1 is collected, and a defect detection algorithm is used to perform a visual analysis on the visual image of the food at the entrance section of the finished product conveyor belt 100. The visual analysis results are stored in the memory 21, and the spectral imaging tracking controller 5 communicates with the memory 21 to exchange the visual analysis results.

[0044] In the tenth embodiment, the planar moving mechanism 4 is an electric cross slide.

[0045] The above are all preferred embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A food production line visual quality inspection method based on spectral technology, characterized by: The following steps are involved: Step 1, using a visual imaging device (1) to obtain a visual image of food at an exit section of a finished product conveyor belt (100) of a food production line; Step 2, the visual analysis module (2) uses a defect detection algorithm to perform a visual analysis on the food visual image at the entrance section of the finished product conveyor belt (100), marks the contour of the defective part, and calculates the center point of the contour of the defective part; Step 3, controlling the spectral imaging device (3) installed above the finished product conveyor belt (100) of the food production line to move to the position directly above the center point of the contour of the defective part, controlling the speed to be consistent with that of the finished product conveyor belt (100) of the food production line, and tracking the center point of the contour of the defective part; Step 4, the spectral imaging device (3) performs high-resolution spectral imaging on the defect position to obtain spectral data of the defect position; Step 5: Use the trained PLSR model to perform defect analysis on the spectral data of the defective part, and output whether there is a defect and the defect classification result; Step 6: If it is determined that there is a defect, the sorting robot (7) above the exit section of the finished product conveyor belt (100) is controlled to remove the food at the position corresponding to the defect.

2. The method for visual quality inspection of a food production line based on spectral technology according to claim 1, characterized in that: The finished product conveyor belt (100) is divided into a finished product conveyor belt (100) entrance section, a spectral tracking detection section, and a finished product conveyor belt (100) exit section. Assuming that the total length of the finished product conveyor belt (100) is S, the finished product conveyor belt (100) entrance section is one-third of the entrance end of the finished product conveyor belt (100), the spectral tracking detection section is one-third of the middle part of the finished product conveyor belt (100), and the finished product conveyor belt (100) exit section is one-third of the exit end of the finished product conveyor belt (100).

3. The method for visual quality inspection of a food production line based on spectral technology according to claim 2, characterized in that: In step 2, the defect detection algorithm is: extract visual features from the image, input the visual features into a defect classifier based on machine vision, and when the defect classifier outputs the visual features as defect features, extract the contour of the area corresponding to the defect features and mark it as the defect part contour.

4. The method for visual quality inspection of a food production line based on spectral technology according to claim 3 is characterized by: The Canny algorithm is used to extract the edge contour of the area corresponding to the defect feature as the defect part contour. For each detected defect part contour, its center point is calculated as the center point of the defect part contour. The optical flow method or Kalman filter tracking algorithm is used in continuous frames to track the center point of the defect part contour.

5. The method for visual quality inspection of a food production line based on spectral technology according to claim 4, characterized in that: In step 3, the spectral imaging device (3) is installed on the planar moving part of the planar moving mechanism (4), and is driven by the planar moving part to move in a planar manner above the spectral tracking detection section of the finished product conveyor belt (100). The controller of the planar moving mechanism (4) and the visual analysis module (2) interactively track the coordinates of the center point of the contour of the defective part, and interact with the controller of the finished product conveyor belt (100) to exchange the conveying speed data. The planar moving part first drives the spectral imaging device (3) to move to the center point of the contour of the defective part, and then matches the conveying speed, and maintains it for 1-3 seconds, and controls the spectral imaging device (3) to perform high-resolution spectral imaging of the area below to obtain spectral data of the defective part.

6. The method for visual quality inspection of a food production line based on spectral technology according to claim 5, characterized in that: The specific steps in step 5 to use the trained PLSR model to perform defect analysis on the spectral data of the defect part are: Step 51, performing normalization preprocessing on the spectral data; Step 52, the goal of the trained PLSR model is to find a set of projection vectors w and c that maximize the covariance of the spectral data X and the attribute Y respectively; Step 53, use the trained PLSR model to perform defect prediction analysis on the defect part spectral data: ; is the predicted attribute value, which includes different pre-classified defect types, X is the spectral data of the defect part, are the regression coefficients of the PLSR model.

7. A visual quality inspection device for food production lines based on spectral technology, characterized by: A method for visual quality inspection of a food production line based on spectral technology according to claim 6, wherein the visual quality inspection device for a food production line comprises a visual imaging device (1), a visual analysis module (2), a spectral imaging device (3), a planar moving mechanism (4), a chip-based spectral imaging tracking controller (5) and a defect analysis computer (6); The visual imaging device (1) is installed above the entrance section of the finished product conveyor belt (100) via a bracket, and the visual analysis module (2) is communicatively connected to the visual imaging device (1); The planar moving mechanism (4) is installed above the spectral tracking detection section of the finished product conveyor belt (100) via a bracket, the spectral imaging device (3) is installed on the planar moving part of the planar moving mechanism (4), and the spectral imaging tracking controller (5) is connected to the visual analysis module (2) for communication and controls the execution action of the planar moving mechanism (4); The defect analysis computer (6) is connected to the spectral imaging device (3) for communication, exchanges spectral data of defective parts, and uses the trained PLSR model to perform defect analysis on the spectral data of defective parts, and outputs whether there is a defect and a defect classification result.

8. The food production line visual quality inspection device based on spectral technology according to claim 7 is characterized by: It also includes a sorting robot (7), which is connected to the defect analysis computer (6) for communicating and exchanging visual feature position data of defects, and the sorting robot (7)'s sorting mechanical arm removes defective food corresponding to the defective visual feature position.

9. The food production line visual quality inspection device based on spectral technology according to claim 8, characterized in that: The visual imaging device (1) is a high-definition visual camera, and the spectral imaging device (3) is a spectral camera; The visual analysis module (2) comprises a memory (21) and a visual analysis chip (22), wherein the memory (21) is communicatively connected to the visual imaging device (1), and the visual analysis chip (22) is communicatively connected to the memory (21), collects visual images of the visual imaging device (1), and uses a defect detection algorithm to perform visual analysis on the visual images of food at the entrance section of the finished product conveyor belt (100), and stores the visual analysis results in the memory (21), and the spectral imaging tracking controller (5) communicates and exchanges the visual analysis results with the memory (21).

10. The food production line visual quality inspection device based on spectral technology according to claim 9, characterized in that: The planar moving mechanism (4) is an electric cross slide.

Images