Clean vegetable processing quality management and control method and system based on machine vision

Abstract

The invention discloses a clean vegetable processing quality control method and system based on machine vision, and relates to the technical field of vegetable processing, and the method comprises the steps: collecting a vegetable contour image in real time through an industrial camera installed on a vegetable sorting line, and carrying out the image processing and feature extraction to obtain feature data; performing dimensionless processing on the characteristic data to obtain a pixel perturbation data set and a contour steady-state data set, performing vegetable browning risk assessment by calculating an oxidation driving characteristic index BDI, triggering an interference analysis instruction when the browning risk exists, calculating a visual interference elimination index VDI to perform abnormal interference assessment, and determining the browning risk of the vegetable. And when visual interference does not exist, a comprehensive structure analysis instruction is executed, a structure disintegration consistency index SCI is calculated for comprehensive abnormal risk assessment, and recognition and final quality judgment of the state of the irreversible oxidation channel are achieved. The method has dynamic responsiveness and abnormity traceability, and the accuracy of vegetable browning identification and the intelligent level of processing and screening are improved.

Description

Technical Field

[0001] This invention relates to the field of vegetable processing technology, specifically to a method and system for quality control of pre-processed vegetables based on machine vision. Background Technology

[0002] With the rapid development of the pre-processed vegetable industry, traditional quality assessment methods relying on manual inspection are no longer sufficient to meet the demands of large-scale, high-efficiency production. Machine vision technology, due to its non-contact, high-precision, and automated characteristics, has become a key means of achieving intelligent identification and dynamic monitoring in pre-processed vegetables. In practical applications, vegetables pass through sorting lines at high speeds. Images captured by industrial cameras not only reflect their shape and texture but also uncover deeper quality information through multi-channel image processing. Especially in quality control, by modeling the features of potential physical damage, corrosion, water film residue, and texture breakage in the images, a comprehensive perception of the freshness, integrity, and processing degree of vegetables can be achieved. Browning, as a typical indicator of vegetable quality decline, often indicates that vegetables have entered the oxidative deterioration stage and needs to be identified and judged as a key aspect of the quality control process.

[0003] The analysis of vegetable browning mainly relies on color detection and brightness change judgment. However, these methods often suffer from problems such as strong subjectivity, high sensitivity to interference, and unstable accuracy. On the one hand, color changes are easily affected by external lighting conditions, water film reflection, image contrast, and other interference factors, causing non-browning areas to be misjudged as abnormal areas. On the other hand, many physical damages may appear similar to browning areas in images, easily leading to false alarms triggered by the system. In addition, existing methods generally lack in-depth assessment of the structural response of vegetables, making it difficult to capture the microstructural anomalies before browning occurs, causing the judgment results to often lag behind actual quality changes. These limitations can easily lead to the coexistence of unidentified high-risk vegetables and normal vegetables being rejected on automated production lines, seriously affecting the efficiency of vegetable grading and product quality stability. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method and system for quality control of pre-processed vegetables based on machine vision, which solves the problems mentioned in the background.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for quality control of pre-processed vegetables based on machine vision, comprising the following steps:

[0006] S1. The outline image of the vegetables is captured in real time by the installed industrial camera, and the outline image is transmitted to the central processing unit in real time via USB interface for image processing and feature extraction to obtain feature data. Then, the feature data is processed to obtain pixel perturbation data group and outline steady state data group.

[0007] S2. Calculate the oxidation-driven feature index (BDI) based on the pixel perturbation data set, and perform a browning risk assessment of vegetables by comparing it with the preset browning-driven activation threshold (BAT). If the assessment indicates that there is a browning risk, execute the interference analysis command.

[0008] S3. Execute the interference analysis command, calculate the visual interference removal index VDI through the contour steady-state data group, and evaluate the abnormal interference with the preset interference anomaly removal threshold VIR. Execute the comprehensive structural analysis command when the evaluation shows that there is no visual interference.

[0009] S4. Execute the comprehensive structural analysis command, calculate the structural disintegration consistency index (SCI) based on the oxidation-driven characteristic index (BDI) and the visual interference removal index (VDI), and conduct a comprehensive anomaly risk assessment with the preset structural disintegration critical threshold (SCR).

[0010] Preferably, S1 includes S11 and S12;

[0011] S11. Install an industrial camera in the detection section before the vegetable sorting line, and trigger the visual trigger signal built into the industrial camera when the vegetables pass under the industrial camera. At this time, the industrial camera starts the shooting command and captures the outline image of the vegetables in real time.

[0012] The industrial camera is an industrial camera with no less than two million pixels and a frame rate of no less than 30fps.

[0013] S12. The industrial camera establishes a connection with the central processing unit through the USB interface and establishes a transmission channel through the USB protocol to transmit the acquired contour image to the data processing center in real time for image processing and to obtain the image feature set.

[0014] The image processing includes image channel expansion, multi-color space expansion, and texture structure response mapping;

[0015] The image extension channel is used to detect local water film regions formed by reflection in the contour image based on the image edge blur difference algorithm, and to construct a pseudo α layer for the local reflective water film regions based on the image edge gradient field. Then, Gaussian blur difference and Sobel edge fusion are used to generate a weight map to construct an RGBA image.

[0016] The multi-color space expansion is used to correct the brightness of the contour image using a bi-segmented normalized brightness redistribution function, and then the corrected contour image is converted into an HSV image through a color wheel space angle mapping algorithm.

[0017] The texture structure response mapping is used to perform two-dimensional convolution on the contour image at 0°, 45°, 90° and 135° according to the principal axis distribution law of the directional filter bank based on the texture direction, and extract Gabor response maps in different directions.

[0018] Preferably, S1 also includes S13;

[0019] S13. The image feature set is extracted through the data processing center to obtain feature data, and then the feature data is processed to be dimensionless to obtain pixel perturbation data set and contour steady state data set.

[0020] The pixel perturbation data set includes the α channel color spot irregular diffusion radius pa, frequency ripple jump point density pb, and local brightness tomography pc;

[0021] The profile steady-state data set includes the α-channel irregular diffusion area pd and the profile bilateral corner point concentration pe;

[0022] The α channel layer in the RGBA image is extracted using channel separation technology. Combined with the transparency threshold customized by the user according to actual needs, the α channel is thresholded to extract the contours of abnormal distribution areas with transparency greater than the transparency threshold. The area enclosed by the contours of the abnormal distribution areas is calculated to obtain the irregular diffusion area pd of the α channel. Based on the extracted abnormal region contours, the principal component direction fitting method is used to restore the principal axis direction of the patch, and the radial distance sequence from the principal axis centerline to each contour point is calculated. The maximum difference in this sequence is defined as the irregular diffusion radius pa of the α channel color patch.

[0023] The Gabor response map is traversed in 3x3 sliding windows, and a texture response threshold is set. This threshold is determined by the user based on actual needs. Within each sliding window, the number N of pixels with a texture response value G higher than the texture response threshold is counted and compared with the current response region area A to obtain the frequency ripple jump point density pb. Specifically: In the formula, M represents the total window above the texture response threshold, and N... j and A j These represent the number of pixels in the sliding window j that are above the texture response threshold and the area of ​​the response region, respectively.

[0024] The V channel is extracted from an HSV image using matrix channel separation technology. Edge detection is then performed on the extracted V channel image using the Laplacian operator to extract regions of abrupt brightness changes. The mean intensity of these abrupt brightness changes is calculated to obtain the local brightness tomography rate (pc). Specifically: In the formula, N represents the total number of pixels in the V-section. This represents the Laplacian response value of the V channel at pixel i;

[0025] After weighted average grayscale conversion of the RGBA image, the Canny algorithm is used to extract the boundary region with gradient intensity in the range of 50 to 150, and the Shi-Tomasi algorithm is used to extract corner points to obtain the effective number of corner points nc. The contour fitting function is used to extract the boundary closed region and the total area of ​​the closed region ab. The contour bilateral corner point concentration pe is obtained by calculating the ratio of the effective number of corner points nc to the total area of ​​the region ab.

[0026] Preferably, S2 includes S21;

[0027] S21. Based on the pixel perturbation data group, the oxidation-driven characteristic index (BDI) is obtained through summarization and calculation. It is used to measure the structural response intensity in the oxidation initiation stage and reflects the initial reaction process of vegetables before entering the visible browning pathway. The specific formula is as follows.

[0028]

[0029] In the formula, ln represents the logarithmic function and exp represents the exponential decay function.

[0030] Preferably, S2 also includes S22;

[0031] S22. Sort all oxidation-driven characteristic indices (BDI) obtained in the past three months from smallest to largest, and use the quantile method to set the 90th quantile of the sorted index as the preset browning-driven activation threshold (BAT). Then, use the real-time obtained oxidation-driven characteristic index (BDI) and the preset browning-driven activation threshold (BAT) to conduct a vegetable browning risk assessment. The specific assessment scheme is as follows.

[0032] When the oxidation-driven characteristic index (BDI) is less than the browning-driven activation threshold (BAT), it indicates that the vegetables are healthy and will automatically be released into the packaging process.

[0033] When the oxidation-driven characteristic index BDI is greater than or equal to the browning-driven activation threshold BAT, it indicates that the vegetable is at risk of browning. At this time, the vegetable is marked as an abnormal vegetable and interference analysis is performed.

[0034] Preferably, S3 includes S31;

[0035] S31. When the risk assessment of browning in vegetables indicates the presence of browning risk, execute the interference analysis instruction.

[0036] The interference analysis command is used to perform summary calculations based on the contour steady-state data set to obtain the Visual Interference Removal Index (VDI), which is used to identify non-browning pseudo-abnormalities and remove pseudo-browning identification caused by visual condition distortion. The specific formula is as follows.

[0037]

[0038] In the formula, sin represents the sine function, ln represents the logarithmic function, and π represents pi, which is taken to two decimal places.

[0039] Preferably, S3 also includes S32;

[0040] S32. Sort all visual interference removal indices (VDI) obtained in the past three months from smallest to largest, and use the quantile method to set the 85th quantile of the sorted index as the preset interference anomaly removal threshold (VIR). Then, evaluate the abnormal interference by comparing the real-time visual interference removal index (VDI) with the preset interference anomaly removal threshold (VIR). The specific evaluation scheme is as follows.

[0041] When the Visual Interference Removal Index (VDI) is less than the Interference Anomaly Removal Threshold (VIR), it indicates that there is no visual interference, and the comprehensive structural analysis command is executed at this time.

[0042] When the visual interference removal index VDI is greater than or equal to the interference anomaly removal threshold VIR, it indicates the presence of visual interference and misjudgment of browning. In this case, a second iteration analysis is performed through S1.

[0043] Preferably, S4 includes S41;

[0044] S41. When the abnormal interference assessment indicates that there is no visual interference, execute the comprehensive structural analysis command;

[0045] The integrated structural analysis command is used to comprehensively calculate the obtained oxidation-driven characteristic index (BDI) and visual interference rejection index (VDI) to obtain the structural disintegration consistency index (SCI), which is used to determine whether the sample has entered the irreversible oxidation channel stage. The specific formula is as follows:

[0046]

[0047] In the formula, ln represents the logarithmic function, and tanh represents the hyperbolic tangent function.

[0048] Preferably, S4 also includes S42;

[0049] S42. Sort all structural disintegration consistency indices (SCIs) obtained in the past three months from smallest to largest, and use the quantile method to set the 95th quantile of the sorted indices as the preset structural disintegration critical threshold (SCR). Then, conduct a comprehensive anomaly risk assessment by combining the real-time obtained structural disintegration consistency indices (SCIs) with the preset structural disintegration critical threshold (SCR). The specific assessment scheme is as follows.

[0050] When the Structural Disintegration Consistency Index (SCI) is less than or equal to the Structural Disintegration Critical Threshold (SCR), it indicates that the vegetable is in an abnormal critical state. At this time, the vegetable is marked as a critical transition vegetable and transferred to the manual review channel for manual review and included in the short-term sales channel.

[0051] When the Structural Disintegration Consistency Index (SCI) is greater than the Structural Disintegration Critical Threshold (SCR), it indicates that the vegetable is in a browning and damaged state. At this time, the vegetable is marked as browning and damaged and is directly judged as unqualified and removed and recycled.

[0052] A machine vision-based quality control system for processed vegetables includes a data extraction module, a browning analysis module, an interference analysis module, and a comprehensive risk assessment module.

[0053] The data extraction module is used to capture the outline image of vegetables in real time using the installed industrial camera, and transmit the outline image to the central processing unit in real time via USB interface for image processing and feature extraction to obtain feature data. Then, the feature data is processed to obtain pixel perturbation data group and outline steady state data group.

[0054] The browning analysis module is used to calculate the oxidation-driven feature index (BDI) based on the pixel perturbation data set, and to perform a browning risk assessment of vegetables with the preset browning-driven activation threshold (BAT). When the assessment indicates that there is a browning risk, the module executes the perturbation analysis command.

[0055] The interference analysis module is used to execute interference analysis instructions, calculate the visual interference rejection index (VDI) through the contour steady-state data set, evaluate abnormal interference with the preset interference anomaly rejection threshold (VIR), and execute the comprehensive structural analysis instructions when the evaluation shows that there is no visual interference.

[0056] The comprehensive risk assessment module is used to execute comprehensive structural analysis instructions, calculate the structural disintegration consistency index (SCI) based on the oxidation-driven characteristic index (BDI) and the visual interference removal index (VDI), and perform a comprehensive anomaly risk assessment with the preset structural disintegration critical threshold (SCR).

[0057] This invention provides a method and system for quality control of pre-processed vegetables based on machine vision. It has the following beneficial effects:

[0058] (1) This method achieves a low-latency and high-precision image acquisition mechanism by setting up a high-resolution industrial camera on the vegetable sorting line to capture contour images in real time and transmitting the images to the central processing unit in real time via a USB interface. In the image processing stage, an image extension channel, multi-color space mapping and texture structure response processing strategy are introduced. This series of composite image processing strategies provides high-dimensional and robust input data for subsequent anomaly recognition. Feature data is obtained by feature extraction, and then the feature data is processed to obtain pixel perturbation data group and contour steady state data group, which effectively makes up for the problem of single dimension and insufficient sensitivity in the judgment of epidermal structure state in the existing technology.

[0059] (2) This method constructs an oxidation-driven feature index (BDI) based on pixel perturbation data sets to assess the potential trend of vegetables entering the browning transformation path. Subsequently, the frequency ripple jump point density (pb) is introduced as a texture fluctuation factor. The BDI is constructed using an exponential suppression function and a logarithmic function. The browning activation threshold (BAT) is set using the 90th percentile method, and the real-time acquired BDI is used to assess the browning risk of vegetables, improving the model's adaptability to different seasons, batches, and vegetable varieties, and achieving dynamic determination of oxidation risk. If the browning risk assessment indicates the presence of browning risk, a visual interference removal index (VDI) is constructed based on the contour steady-state data set. An inverse interference model is constructed using a combination of trigonometric and logarithmic functions. An abnormal interference assessment is then performed using the 85th percentile method to set the interference removal threshold (VIR) and the real-time acquired VDI. The entire assessment chain forms a dual protection mechanism of primary discrimination and secondary correction, effectively solving the problems of high false detection rate and poor interpretability in traditional image discrimination.

[0060] (3) When the abnormal interference assessment indicates the absence of visual interference, this method executes a comprehensive structural analysis command, nonlinearly fusing the oxidation-driven characteristic index (BDI) and the visual interference removal index (VDI) to construct a structural disintegration consistency index (SCI). The SCI uses logarithmic and hyperbolic sine functions to handle extreme value suppression and interference stability, and finally restores the physical scoring scale through a comprehensive square root, forming a quantifiable and stratified structural disintegration degree score. A structural disintegration critical threshold (SCR) based on the 95th percentile of historical data is introduced, and a comprehensive abnormal risk assessment is performed with the real-time acquired SCI, enabling the distinction between abnormal critical states and complete disintegration states. This method completes a closed-loop process of image acquisition, parameter extraction, risk assessment, and classification, which not only significantly improves the assessment accuracy and operational efficiency, but also constructs a replicable and scalable intelligent screening framework for clean vegetable quality through intelligent threshold management and structural analysis mechanisms. This solves the core problems of subjective judgment, delayed response, and inconsistent processing in traditional methods, significantly improving the quality control level and graded processing capabilities in vegetable processing. Attached Figure Description

[0061] Figure 1 This is a schematic diagram of the steps in the machine vision-based vegetable processing quality control method of the present invention.

[0062] Figure 2 This is a schematic diagram of the process of the vegetable processing quality control system based on machine vision according to the present invention.

[0063] Figure 3 This is a schematic diagram of the progressive interference suppression curve of the present invention;

[0064] Figure 4 This is a diagram showing the relationship between the indicators of the quality control method for pre-processed vegetables based on machine vision, as described in this invention. Detailed Implementation

[0065] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0066] Example 1

[0067] Please see Figure 1 This invention provides a machine vision-based method for quality control in the processing of pre-processed vegetables. To achieve the above objectives, this invention employs the following technical solution, comprising the following steps:

[0068] S1. The outline image of the vegetables is captured in real time by the installed industrial camera, and the outline image is transmitted to the central processing unit in real time via USB interface for image processing and feature extraction to obtain feature data. Then, the feature data is processed to obtain pixel perturbation data group and outline steady state data group.

[0069] S2. Calculate the oxidation-driven feature index (BDI) based on the pixel perturbation data set, and perform a browning risk assessment of vegetables by comparing it with the preset browning-driven activation threshold (BAT). If the assessment indicates that there is a browning risk, execute the interference analysis command.

[0070] S3. Execute the interference analysis command, calculate the visual interference removal index VDI through the contour steady-state data group, and evaluate the abnormal interference with the preset interference anomaly removal threshold VIR. Execute the comprehensive structural analysis command when the evaluation shows that there is no visual interference.

[0071] S4. Execute the comprehensive structural analysis command, calculate the structural disintegration consistency index (SCI) based on the oxidation-driven characteristic index (BDI) and the visual interference removal index (VDI), and conduct a comprehensive anomaly risk assessment with the preset structural disintegration critical threshold (SCR).

[0072] In this embodiment, step S1 involves capturing real-time outline images of vegetables using an industrial camera and transmitting these images to a central processing unit via a USB interface for image processing and feature extraction. Combined with multi-channel image enhancement technology and dimensionless processing, this accurately acquires pixel perturbation data sets and outline steady-state data sets, significantly improving the system's sensitivity to subtle changes in the vegetable's appearance and establishing a quantifiable and traceable quality evaluation foundation. Step S2 constructs an oxidation-driven feature index (BDI) based on the pixel perturbation data sets. This index, compared with a set browning-driven activation threshold (BAT), assesses the risk of browning in vegetables, enabling proactive identification of early oxidation reactions. When the browning risk assessment indicates a risk of browning, the process immediately proceeds to step S3. In step S3, a visual interference rejection index (VDI) is constructed using the outline steady-state data sets, and compared with a preset interference anomaly rejection threshold (VIR) to assess and reject false anomalies such as reflections, water spots, and physical indentations. This structure, through dual-index linkage and a dynamic threshold mechanism, avoids the high misjudgment rate of single-index indicators and poor ability to identify non-target interference in existing technologies, making anomaly identification more accurate and stable. In S4, the Structure Disintegration Consistency Index (SCI) is calculated by fusing the Oxidation-Driven Characteristic Index (BDI) and the Visual Interference Removal Index (VDI). This SCI is then compared with the preset Structure Disintegration Critical Threshold (SCR) for a comprehensive anomaly risk assessment, completing the final grading of abnormal vegetables and distinguishing them into three levels: healthy, critical, and damaged. Corresponding measures are then taken, such as release, transfer to manual review, or removal, depending on the level of the vegetable. Compared to traditional methods based on human sensory evaluation or single-parameter scoring, this method not only improves the objectivity and repeatability of the assessment but also demonstrates higher versatility and scalability in multi-batch and multi-variety processing. This method effectively solves the core pain points of traditional technologies, such as high subjective misjudgment rate, large detection delay, and poor interference adaptability, achieving fully automated and high-precision quality control of perception, identification, judgment, and screening in the processing of clean vegetables.

[0073] Example 2

[0074] Please refer to Figure 1 and Figure 4 Specifically: S1 includes S11 and S12;

[0075] S11. Install an industrial camera in the detection section before the vegetable sorting line, and trigger the visual trigger signal built into the industrial camera when the vegetables pass under the industrial camera. At this time, the industrial camera starts the shooting command and captures the outline image of the vegetables in real time.

[0076] The industrial camera is an industrial camera with no less than two million pixels and a frame rate of no less than 30fps.

[0077] S12. The industrial camera establishes a connection with the central processing unit through the USB interface and establishes a transmission channel through the USB protocol to transmit the acquired contour image to the data processing center in real time for image processing and to obtain the image feature set.

[0078] The image processing includes image channel expansion, multi-color space expansion, and texture structure response mapping;

[0079] The image extension channel is used to detect local water film regions formed by reflection in the contour image based on the image edge blur difference algorithm, and to construct a pseudo α layer for the local reflective water film regions based on the image edge gradient field to simulate the light transmittance distribution of the water film in the region. Then, Gaussian blur difference and Sobel edge fusion are used to generate a weight map to construct an RGBA image.

[0080] The multi-color space expansion is used to correct the brightness of the contour image using a bi-segmented normalized brightness redistribution function, and then the corrected contour image is converted into an HSV image through a color wheel space angle mapping algorithm. During the mapping process, brightness is kept as the main axis to avoid highlight distortion.

[0081] The texture structure response mapping is used to perform two-dimensional convolution on the contour image at 0°, 45°, 90° and 135° according to the principal axis distribution law of the directional filter bank based on the texture direction, that is, the texture corresponding to the horizontal distribution is 0°. This extracts Gabor response maps in different directions.

[0082] S1 also includes S13;

[0083] S13. The image feature set is extracted through the data processing center to obtain feature data, and then the feature data is processed to be dimensionless to obtain pixel perturbation data set and contour steady state data set.

[0084] The pixel perturbation data set includes the α channel color spot irregular diffusion radius pa, frequency ripple jump point density pb, and local brightness tomography pc;

[0085] The profile steady-state data set includes the α-channel irregular diffusion area pd and the profile bilateral corner point concentration pe;

[0086] The α channel layer in the RGBA image is extracted using channel separation technology. Combined with the transparency threshold customized by the user according to actual needs, the α channel is thresholded to extract the contours of abnormal distribution areas with transparency greater than the transparency threshold. The area enclosed by the contours of the abnormal distribution areas is calculated to obtain the irregular diffusion area pd of the α channel, which represents the asymmetry of the reflectivity of the bright and dark edges of the image and simulates the intensity of mirror interference. Based on the extracted abnormal region contours, the principal component direction fitting method is used to restore the principal axis direction of the patch and calculate the radial distance sequence from the center line of the principal axis to each contour point. The maximum difference in this sequence is defined as the irregular diffusion radius pa of the α channel color spot, which is used to measure the radial expansion amplitude and diffusion unevenness of the abnormal area and reflect the geometric dispersion of wet spots and damaged areas.

[0087] The Gabor response map is traversed in 3x3 sliding windows, and a texture response threshold is set. This threshold is determined by the user based on actual needs. Within each sliding window, the number N of pixels with a texture response value G higher than the texture response threshold is counted and compared with the current response area A to obtain the frequency jump point density pb. This density reflects the texture fracture trend and fracture risk level of the epidermal microstructure. Specifically: In the formula, M represents the total window above the texture response threshold, and N... j and A j These represent the number of pixels in the sliding window j that are above the texture response threshold and the area of ​​the response region, respectively.

[0088] The V channel is extracted from an HSV image using matrix channel separation technology. Edge detection is then performed on the extracted V channel image using the Laplacian operator to extract regions of abrupt brightness changes. The mean intensity of these abrupt brightness changes is calculated to obtain the local brightness tomography rate (pc), which reflects the probability of a sharp decrease in brightness within the region. Specifically: In the formula, N represents the total number of pixels in the V-section. This represents the Laplacian response value of the V channel at pixel i;

[0089] After weighted average grayscale conversion of the RGBA image, the Canny algorithm is used to extract the boundary region with gradient intensity in the range of 50 to 150, and the Shi-Tomasi algorithm is used to extract corner points to obtain the effective number of corner points nc. The contour fitting function is used to extract the boundary closed region and the total area of ​​the closed region ab. By calculating the ratio of the effective number of corner points nc to the total area of ​​the region ab, the contour bilateral corner point concentration pe is obtained, which represents non-oxidative indentation and physical extrusion, reflecting potential physical fatigue and processing damage.

[0090] In this embodiment, industrial cameras with at least two megapixels and a frame rate of at least 30fps are deployed at the front end of the vegetable sorting line. A visual triggering mechanism is used to acquire vegetable outline images in real time, which are then transmitted at high speed via USB to a data processing center for image extension channel construction, water film transmittance modeling, color space redistribution, and directional texture response mapping, forming a high-dimensional image feature set. The data processing center extracts feature data from this set and then performs dimensionless processing to construct pixel perturbation data groups and outline steady-state data groups. This enables accurate perception and dimensionless expression of multi-dimensional structural anomalies such as micro-texture, brightness abrupt changes, reflection perturbations, and indentations on the vegetable surface. This comprehensive implementation not only significantly enhances the system's ability to identify real quality problems under complex image interference backgrounds but also provides accurate and traceable image physical evidence for subsequent anomaly mechanism construction. It solves problems such as false positives due to reflection, failure to recognize weak textures, and strong subjectivity of quality parameters in traditional image processing modes. It achieves a complete process upgrade from image acquisition, structural modeling, and parameter extraction, significantly improving the stability, accuracy, and adaptability of the clean vegetable quality identification system.

[0091] Example 3

[0092] Please refer to Figure 1 and Figure 4 Specifically: S2 includes S21;

[0093] S21. Based on the pixel perturbation data group, the oxidation-driven characteristic index (BDI) is obtained through summarization and calculation. It is used to measure the structural response intensity in the oxidation initiation stage and reflects the initial reaction process of vegetables before entering the visible browning pathway. The specific formula is as follows.

[0094]

[0095] In the formula, ln represents the logarithmic function, exp represents the exponential decay function, |pa| takes the absolute value to avoid non-physical shifts caused by minimum and negative values, exp(-pc) represents the exponential suppression factor. If the local brightness tomography rate pc approaches 0, then the suppression factor term approaches 1. The square root is used to construct the structural response balance. The logarithmic function ln is used to compress the entire exponent, so that the distribution is stable in the intermediate value range.

[0096] The following is a table of examples of the oxidation-driven characteristic index (BDI);

[0097]

[0098] S2 also includes S22;

[0099] S22. Sort all oxidation-driven characteristic indices (BDI) obtained in the past three months from smallest to largest, and use the quantile method to set the 90th quantile of the sorted index as the preset browning-driven activation threshold (BAT). Then, use the real-time obtained oxidation-driven characteristic index (BDI) and the preset browning-driven activation threshold (BAT) to conduct a vegetable browning risk assessment. The specific assessment scheme is as follows.

[0100] When the oxidation-driven characteristic index (BDI) is less than the browning-driven activation threshold (BAT), it indicates that the vegetables are healthy and will automatically be released into the packaging process.

[0101] When the oxidation-driven characteristic index BDI is greater than or equal to the browning-driven activation threshold BAT, it indicates that the vegetable is at risk of browning. At this time, the vegetable is marked as an abnormal vegetable and interference analysis is performed.

[0102] In this embodiment, by comprehensively calculating the pixel perturbation data group in the vegetable image feature data, the oxidation-driven feature index (BDI) is constructed. The composite formula is constructed using the logarithmic function, the exponential inhibition factor, and the structural balance radical formula, which effectively measures the initial oxidation reaction trend of vegetables before entering the visible browning path.

[0103] The physical significance of the formula lies in accurately depicting the oxidation response trend of vegetables before browning occurs, and achieving early warning identification through the fusion analysis of different types of perturbations. (1+|pa|) 2 The radial diffusion amplitude represents the dominant term for structural disturbance, and pa represents the irregular diffusion radius of the α channel spot, reflecting the radial expansion of the wet spot and the early browning region. The square of the addition of one strengthens its driving force in the formula, and the absolute value is used to suppress the disturbance shift caused by directional fluctuations. The perturbation intensity of the frequency ripple jump point density pb on the overall structural stability is represented by the exponential suppression function of the local brightness tomography rate pc in the denominator. The sum of the squares of the above two terms and the square root represent the synthetic response intensity of different microstructural perturbations, which improves the physical integrity and structural coupling of the overall model. The logarithmic compression ln is used to convert the exponentially growing response intensity into a stable intermediate interval distribution, which enhances the model's tolerance to extreme perturbation data and improves the stability and reliability of discrimination. The α-channel irregular diffusion radius pa simulates geometric instability. When the diffusion range of vegetable skin or spots suddenly increases, the α-channel irregular diffusion radius pa will rapidly increase, causing instability to the overall structure. Therefore, its square is used as the structural driving force. The frequency jump point density pb reflects the texture imbalance. The higher the frequency jump point density pb, the more high-frequency jumps exist in the local skin, indicating that the microstructure damage tends to be more serious. However, its effect is amplified when the brightness is stable. If the brightness changes drastically, the effect is visually shielded. Therefore, an inhibition factor needs to be introduced. The local brightness discontinuity rate pc provides context adjustment capability. The brightness discontinuity rate is not only a direct risk parameter but also a modulatory factor of the frequency jump point density pb, reflecting the dynamic coupling structure between the three and closely matching the visual interference and physical oxidation state of real images.

[0104] Subsequently, based on historical oxidation-driven characteristic index (BDI) data from the past three months, the 90th quantile was extracted using a quantile method to set the browning-driven activation threshold (BAT), forming a risk identification standard with dynamic adaptive capabilities. When the real-time BDI exceeds the BAT threshold, vegetables are automatically marked as abnormal and enter the subsequent interference identification process; when the BDI falls below the BAT threshold, they are directly released into the packaging process. The implementation of this module effectively achieves early identification and accurate warning of browning trends, avoiding the insensitivity of traditional methods to identifying pre-browning states, improving the system's forward-looking judgment ability regarding potential quality deterioration, providing a scientific basis for subsequent grading and processing decisions, and significantly enhancing the stability and intelligence level of quality control in the processing of clean vegetables.

[0105] Example 4

[0106] Please refer to Figure 1 and Figure 4 Specifically: S3 includes S31;

[0107] S31. When the risk assessment of browning in vegetables indicates the presence of browning risk, execute the interference analysis instruction.

[0108] The interference analysis command is used to perform summary calculations based on the contour steady-state data set to obtain the Visual Interference Removal Index (VDI), which is used to identify non-browning pseudo-abnormalities and remove pseudo-browning identification caused by visual condition distortion. The specific formula is as follows.

[0109]

[0110] In the formula, sin represents the sine function, ln represents the logarithmic function, and π represents pi, which is taken to two decimal places. Used to adjust the phase of trigonometric functions and smooth the periodic disturbance model, 1+ln(1+pe) 2 This is used to compress excessively large wrinkle values ​​and enhance stability against extreme interference. The reciprocal processing is used to construct a relationship where the higher the degree of wrinkle, the lower the score.

[0111] The following is a table of examples of the Visual Interference Removal Index (VDI);

[0112]

[0113] S3 also includes S32;

[0114] S32. Sort all visual interference removal indices (VDI) obtained in the past three months from smallest to largest, and use the quantile method to set the 85th quantile of the sorted index as the preset interference anomaly removal threshold (VIR). Then, evaluate the abnormal interference by comparing the real-time visual interference removal index (VDI) with the preset interference anomaly removal threshold (VIR). The specific evaluation scheme is as follows.

[0115] When the Visual Interference Removal Index (VDI) is less than the Interference Anomaly Removal Threshold (VIR), it indicates that there is no visual interference, and the comprehensive structural analysis command is executed at this time.

[0116] When the visual interference removal index VDI is greater than or equal to the interference anomaly removal threshold VIR, it indicates the presence of visual interference and misjudgment of browning. In this case, a second iteration analysis is performed through S1.

[0117] In this embodiment, when the risk assessment of browning in vegetables indicates the presence of browning risk, an interference analysis command is immediately executed. The Visual Interference Removal Index (VDI) is obtained by summarizing and calculating the profile steady-state data set. This index uses a sine function to adjust the phase and a logarithmic function to compress wrinkle anomalies. It also establishes a judgment mechanism that high interference corresponds to low scores through a reciprocal form, thereby accurately identifying non-browning false anomalies caused by light reflection, water film, or epidermal indentation.

[0118] The physical meaning of the formula is to identify pseudo-abnormal regions that are unrelated to browning but appear in images as browning-like morphologies. To detect fluctuations in the edges of bright and dark areas caused by local optical perturbations, a periodic analysis is introduced to enhance the model's fitting of periodic disturbances. The α-channel irregular diffusion area pd is used as a spatial measure of the disturbance scale. The sin function introduces a periodic response mechanism to the α-channel irregular diffusion area pd, increasing the modeling capability for periodic disturbances. The phase adjustment amount shifts the peak point of the function's output value, avoiding the structural zero response problem when the irregular diffusion area pd of the α channel is an integer, and enhancing the model's sensitivity to interference modes. The squaring operation further nonnegates the function result, making it directly usable for interference energy scoring. To address the risk assessment of interference in complex structures, a nonlinear mapping was established: "the more complex the morphology, the higher the potential for interference, and the lower the Visual Interference Removal Index (VDI)." The contour-side corner point concentration (pe) was introduced as a complexity indicator reflecting information such as edge structure folds, skin indentations, and structural wrinkles. This was achieved through the use of pe... 2 To enhance the sensitivity of the response curve to extreme folding, the logarithmic function ln is used to compress the variation range of the contour's bilateral corner point concentration pe, preventing exponential growth that could lead to runaway when the contour's bilateral corner point concentration pe is extremely high, while maintaining discriminative power at medium to low values. The reciprocal processing constructs the physical logic that "the larger the contour's bilateral corner point concentration pe, the smaller the score for this item," reflecting that the more complex the structure, the higher the risk of interference. The two results are combined using an additive approach to achieve a fusion evaluation of optical and structural interference factors, avoiding the identification bias of single-factor models.

[0119] Subsequently, an interference anomaly removal threshold VIR is set, and the visual interference removal index VDI acquired in real time is compared with the preset interference anomaly removal threshold VIR to evaluate the presence of visual interference, achieving dynamic assessment. If the visual interference removal index VDI is lower than the interference anomaly removal threshold VIR, it indicates that there are no false anomalies in the current image, and the system can continue to execute the next stage of comprehensive structural analysis; otherwise, image re-sampling and parameter re-extraction processes will be triggered. This effectively constructs a detection, removal, and backtracking mechanism for interference sources, significantly improving the accuracy of identifying false browning, reducing the misjudgment rate caused by environmental noise, and enhancing the robustness and adaptability of anomaly identification in complex optical scenes, thereby ensuring the accuracy and reliability of the entire vegetable quality control process.

[0120] Example 5

[0121] Please refer to Figure 1 , Figure 3 and Figure 4 Specifically: S4 includes S41;

[0122] S41. When the abnormal interference assessment indicates that there is no visual interference, execute the comprehensive structural analysis command;

[0123] The integrated structural analysis command is used to comprehensively calculate the obtained oxidation-driven characteristic index (BDI) and visual interference rejection index (VDI) to obtain the structural disintegration consistency index (SCI), which is used to determine whether the sample has entered the irreversible oxidation channel stage. The specific formula is as follows:

[0124]

[0125] In the formula, ln represents the logarithmic function, tanh represents the hyperbolic tangent function, and BDI... 2 The oxidation tendency is used to amplify the driving force of the structure. ln(1+BDI) is used to smooth the upper limit and avoid high value runaway. tanh(VDI) makes the interference removal part approach the saturation value of 1, establishes the gradual interference suppression curve, and the comprehensive square root is used to restore the original physical dimensions and maintain a reasonable scoring range.

[0126] The following is a table of examples of the Structural Disintegration Consistency Index (SCI);

[0127]

[0128] S4 also includes S42;

[0129] S42. Sort all structural disintegration consistency indices (SCIs) obtained in the past three months from smallest to largest, and use the quantile method to set the 95th quantile of the sorted indices as the preset structural disintegration critical threshold (SCR). Then, conduct a comprehensive anomaly risk assessment by combining the real-time obtained structural disintegration consistency indices (SCIs) with the preset structural disintegration critical threshold (SCR). The specific assessment scheme is as follows.

[0130] When the Structural Disintegration Consistency Index (SCI) is less than or equal to the Structural Disintegration Critical Threshold (SCR), it indicates that the vegetable is in an abnormal critical state. At this time, the vegetable is marked as a critical transition vegetable and transferred to the manual review channel for manual review and included in the short-term sales channel.

[0131] When the Structural Disintegration Consistency Index (SCI) is greater than the Structural Disintegration Critical Threshold (SCR), it indicates that the vegetable is in a browning and damaged state. At this time, the vegetable is marked as browning and damaged and is directly judged as unqualified and removed and recycled.

[0132] In this embodiment, the oxidation-driven characteristic index (BDI) and the visual interference removal index (VDI) are fused nonlinearly, and the BDI is used to achieve the desired result. 2 By enhancing the driving force of structural instability, smoothing extreme values ​​with ln(1+BDI), suppressing interference effects with tanh(VDI), and restoring the scoring scale with the square root, the accurate determination of whether the vegetable structure has entered the irreversible oxidation channel is achieved.

[0133] The physical meaning of the formula aims to integrate the oxidation trend index BDI, which reflects the actual structural changes, with the interference suppression index VDI, which eliminates visual misjudgment, to jointly analyze whether vegetables have entered the abnormally critical browning structural disintegration channel. 2 The square term amplifies the dominant role of oxidation tendency in structural changes, reflecting a key factor in the cumulative degree of structural instability; ln(1+BDI) is used to logarithmically compress the oxidation driving value, suppressing the abnormal amplification that may be caused by high values ​​of the oxidation driving characteristic index BDI, ensuring that the distribution remains stable within a reasonable range; the numerator as a whole represents the intensity of oxidation aggravation exhibited by vegetables within the visible range of the image; tanh(VDI) uses the hyperbolic tangent function to nonlinearly saturate-model the interference suppression ability of the visual interference removal index VDI. When the visual interference removal index VDI is very large and approaches 1, 1+tanh(VDI) forms a regulation term greater than 1, so that the greater the visual interference, the smaller the overall structural disintegration consistency index SCI value, thus playing a role in the reverse suppression of interference; the denominator as a whole reflects the negative feedback influence of image visual interference factors on the structural disintegration judgment result; the square root of the entire formula is used to increase the BDI value. 2 The resulting magnitude returns to a physically explainable range, causing the Structural Disintegration Consistency Index (SCI) to fall within an appropriate scoring range.

[0134] The Oxidation-Driven Characteristic Index (BDI) is used to characterize the intrinsic chemical and structural changes in the vegetable's skin and tissue structure caused by oxidation, emphasizing the possibility of genuine pathological changes. The Visual Disturbance Removal Index (VDI) is used to identify non-structural pseudo-anomalies caused by ambient light reflection, water spots, indentations, etc., emphasizing the reliability of visual data. Together, they constitute two types of information sources for judging whether vegetables have undergone genuine deterioration. In engineering logic, the assessment of structural anomaly risk must comprehensively consider both the dimensions of "genuine deterioration" and "data reliability." Relying solely on the Oxidation-Driven Characteristic Index (BDI) can be misled by interfering factors, while the Visual Disturbance Removal Index (VDI) itself cannot reflect the true state of vegetables. The essence of fusion is to establish a comprehensive and balanced index.

[0135] Subsequently, based on the structural collapse critical threshold (SCR) of the 95th percentile value of historical data, the structural state of vegetables was divided into two levels: "abnormal critical" and "irreversible damage," which were then manually reviewed and removed, respectively. This implementation method effectively achieved deep-level quality state recognition driven by multi-source image features, solving the problems of unclear identification of critical states and non-tiered processing strategies in traditional methods. It improved the refinement of screening strategies and the accuracy of decision-making, ultimately achieving the goal of efficient diversion of abnormal vegetables and intelligent closed-loop control for quality assurance.

[0136] Example 6

[0137] Please refer to Figure 2A machine vision-based quality control system for processed vegetables includes a data extraction module, a browning analysis module, an interference analysis module, and a comprehensive risk assessment module.

[0138] The data extraction module is used to capture the outline image of vegetables in real time using the installed industrial camera, and transmit the outline image to the central processing unit in real time via USB interface for image processing and feature extraction to obtain feature data. Then, the feature data is processed to obtain pixel perturbation data group and outline steady state data group.

[0139] The browning analysis module is used to calculate the oxidation-driven feature index (BDI) based on the pixel perturbation data set, and to perform a browning risk assessment of vegetables with the preset browning-driven activation threshold (BAT). When the assessment indicates that there is a browning risk, the module executes the perturbation analysis command.

[0140] The interference analysis module is used to execute interference analysis instructions, calculate the visual interference rejection index (VDI) through the contour steady-state data set, evaluate abnormal interference with the preset interference anomaly rejection threshold (VIR), and execute the comprehensive structural analysis instructions when the evaluation shows that there is no visual interference.

[0141] The comprehensive risk assessment module is used to execute comprehensive structural analysis instructions, calculate the structural disintegration consistency index (SCI) based on the oxidation-driven characteristic index (BDI) and the visual interference removal index (VDI), and perform a comprehensive anomaly risk assessment with the preset structural disintegration critical threshold (SCR).

[0142] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for quality control of pre-processed vegetables based on machine vision, characterized by: Includes the following steps: S1. The outline image of the vegetables is captured in real time by the installed industrial camera, and the outline image is transmitted to the central processing unit in real time via USB interface for image processing and feature extraction to obtain feature data. Then, the feature data is processed to obtain pixel perturbation data group and outline steady state data group. S2. Calculate the oxidation-driven feature index (BDI) based on the pixel perturbation data set, and perform a browning risk assessment of vegetables by comparing it with the preset browning-driven activation threshold (BAT). When the assessment indicates that there is a browning risk, execute the interference analysis command. S3. Execute the interference analysis command, calculate the visual interference removal index VDI through the contour steady-state data set, and evaluate the abnormal interference with the preset interference anomaly removal threshold VIR. Execute the comprehensive structural analysis command when the evaluation shows that there is no visual interference. S4. Execute the comprehensive structural analysis command, calculate the structural disintegration consistency index (SCI) based on the oxidation-driven characteristic index (BDI) and the visual interference removal index (VDI), and conduct a comprehensive anomaly risk assessment with the preset structural disintegration critical threshold (SCR).

2. The method for quality control of pre-processed vegetables based on machine vision according to claim 1, characterized in that: S1 includes S11 and S12; S11. Install an industrial camera in the detection section before the vegetable sorting line, and trigger the visual trigger signal built into the industrial camera when the vegetables pass under the industrial camera. At this time, the industrial camera starts the shooting command and captures the outline image of the vegetables in real time. The industrial camera is an industrial camera with no less than two million pixels and a frame rate of no less than 30fps. S12. The industrial camera establishes a connection with the central processing unit through the USB interface and establishes a transmission channel through the USB protocol to transmit the acquired contour image to the data processing center in real time for image processing and to obtain the image feature set. The image processing includes image channel expansion, multi-color space expansion, and texture structure response mapping; The image extension channel is used to detect local water film regions formed by reflection in the contour image based on the image edge blur difference algorithm, and to construct a pseudo α layer for the local reflective water film regions based on the image edge gradient field. Then, Gaussian blur difference and Sobel edge fusion are used to generate a weight map to construct an RGBA image. The multi-color space expansion is used to correct the brightness of the contour image using a bi-segmented normalized brightness redistribution function, and then the corrected contour image is converted into an HSV image through a color wheel space angle mapping algorithm. The texture structure response mapping is used to perform two-dimensional convolution on the contour image at 0°, 45°, 90° and 135° according to the principal axis distribution law of the directional filter bank based on the texture direction, and extract Gabor response maps in different directions.

3. The method for quality control of pre-processed vegetables based on machine vision according to claim 2, characterized in that: S1 also includes S13; S13. The image feature set is extracted through the data processing center to obtain feature data, and then the feature data is processed to be dimensionless to obtain pixel perturbation data set and contour steady state data set. The pixel perturbation data set includes the α channel color spot irregular diffusion radius pa, frequency ripple jump point density pb, and local brightness tomography pc; The profile steady-state data set includes the α-channel irregular diffusion area pd and the profile bilateral corner point concentration pe; The α channel layer in the RGBA image is extracted using channel separation technology. Combined with the transparency threshold customized by the user according to actual needs, the α channel is thresholded to extract the contours of abnormal distribution areas with transparency greater than the transparency threshold. The area enclosed by the contours of the abnormal distribution areas is calculated to obtain the irregular diffusion area pd of the α channel. Based on the extracted abnormal region contours, the principal component direction fitting method is used to restore the principal axis direction of the patch, and the radial distance sequence from the principal axis centerline to each contour point is calculated. The maximum difference in this sequence is defined as the irregular diffusion radius pa of the α channel color patch. The Gabor response map is traversed in 3x3 sliding windows, and a texture response threshold is set. This threshold is determined by the user based on actual needs. Within each sliding window, the number N of pixels with a texture response value G higher than the texture response threshold is counted and compared with the current response region area A to obtain the frequency ripple jump point density pb. Specifically: In the formula, M represents the total window above the texture response threshold, and N... j and A j These represent the number of pixels in the sliding window j that are above the texture response threshold and the area of ​​the response region, respectively. The V channel is extracted from an HSV image using matrix channel separation technology. Edge detection is then performed on the extracted V channel image using the Laplacian operator to extract regions of abrupt brightness changes. The mean intensity of these abrupt brightness changes is calculated to obtain the local brightness tomography rate (pc). Specifically: In the formula, N represents the total number of pixels in the V-section. This represents the Laplacian response value of the V channel at pixel i; After weighted average grayscale conversion of the RGBA image, the Canny algorithm is used to extract the boundary region with gradient intensity in the range of 50 to 150, and the Shi-Tomasi algorithm is used to extract corner points to obtain the effective number of corner points nc. The contour fitting function is used to extract the boundary closed region and the total area of ​​the closed region ab. The contour bilateral corner point concentration pe is obtained by calculating the ratio of the effective number of corner points nc to the total area of ​​the region ab.

4. The method for quality control of pre-processed vegetables based on machine vision according to claim 3, characterized in that: S2 includes S21; S21. Based on the pixel perturbation data group, the oxidation-driven characteristic index (BDI) is obtained through summarization and calculation. It is used to measure the structural response intensity in the oxidation initiation stage and reflects the initial reaction process of vegetables before entering the visible browning pathway. The specific formula is as follows. In the formula, ln represents the logarithmic function and exp represents the exponential decay function.

5. The method for quality control of pre-processed vegetables based on machine vision according to claim 4, characterized in that: S2 also includes S22; S22. Sort all oxidation-driven characteristic indices (BDI) obtained in the past three months from smallest to largest, and use the quantile method to set the 90th quantile of the sorted index as the preset browning-driven activation threshold (BAT). Then, use the real-time obtained oxidation-driven characteristic index (BDI) and the preset browning-driven activation threshold (BAT) to conduct a vegetable browning risk assessment. The specific assessment scheme is as follows. When the oxidation-driven characteristic index (BDI) is less than the browning-driven activation threshold (BAT), it indicates that the vegetables are healthy and will automatically be released into the packaging process. When the oxidation-driven characteristic index BDI is greater than or equal to the browning-driven activation threshold BAT, it indicates that the vegetable is at risk of browning. At this time, the vegetable is marked as an abnormal vegetable and interference analysis is performed.

6. The method for quality control of pre-processed vegetables based on machine vision according to claim 5, characterized in that: S3 includes S31; S31. When the risk assessment of browning in vegetables indicates the presence of browning risk, execute the interference analysis instruction. The interference analysis command is used to perform summary calculations based on the contour steady-state data set to obtain the Visual Interference Removal Index (VDI), which is used to identify non-browning pseudo-abnormalities and remove pseudo-browning identification caused by visual condition distortion. The specific formula is as follows. In the formula, sin represents the sine function, ln represents the logarithmic function, and π represents pi, which is taken to two decimal places.

7. The method for quality control of pre-processed vegetables based on machine vision according to claim 6, characterized in that: S3 also includes S32; S32. Sort all visual interference removal indices (VDI) obtained in the past three months from smallest to largest, and use the quantile method to set the 85th quantile of the sorted index as the preset interference anomaly removal threshold (VIR). Then, evaluate the abnormal interference by comparing the real-time visual interference removal index (VDI) with the preset interference anomaly removal threshold (VIR). The specific evaluation scheme is as follows. When the Visual Interference Removal Index (VDI) is less than the Interference Anomaly Removal Threshold (VIR), it indicates that there is no visual interference, and the comprehensive structural analysis command is executed at this time. When the visual interference removal index VDI is greater than or equal to the interference anomaly removal threshold VIR, it indicates the presence of visual interference and misjudgment of browning. In this case, a second iteration analysis is performed through S1.

8. The method for quality control of pre-processed vegetables based on machine vision according to claim 7, characterized in that: S4 includes S41; S41. When the abnormal interference assessment indicates that there is no visual interference, execute the comprehensive structural analysis command; The integrated structural analysis command is used to comprehensively calculate the obtained oxidation-driven characteristic index (BDI) and visual interference rejection index (VDI) to obtain the structural disintegration consistency index (SCI), which is used to determine whether the sample has entered the irreversible oxidation channel stage. The specific formula is as follows: In the formula, ln represents the logarithmic function, and tanh represents the hyperbolic tangent function.

9. The method for quality control of pre-processed vegetables based on machine vision according to claim 8, characterized in that: S4 also includes S42; S42. Sort all structural disintegration consistency indices (SCIs) obtained in the past three months from smallest to largest, and use the quantile method to set the 95th quantile of the sorted indices as the preset structural disintegration critical threshold (SCR). Then, conduct a comprehensive anomaly risk assessment by combining the real-time obtained structural disintegration consistency indices (SCIs) with the preset structural disintegration critical threshold (SCR). The specific assessment scheme is as follows. When the Structural Disintegration Consistency Index (SCI) is less than or equal to the Structural Disintegration Critical Threshold (SCR), it indicates that the vegetable is in an abnormal critical state. At this time, the vegetable is marked as a critical transition vegetable and transferred to the manual review channel for manual review, and then included in the short-term sales channel. When the Structural Disintegration Consistency Index (SCI) is greater than the Structural Disintegration Critical Threshold (SCR), it indicates that the vegetable is in a state of browning damage. At this time, the vegetable is marked as browning damaged vegetable and is directly judged as unqualified and removed and recycled.

10. A machine vision-based quality control system for processed vegetables, comprising the machine vision-based quality control method for processed vegetables as described in any one of claims 1-9, characterized in that: It includes a data extraction module, a browning analysis module, an interference analysis module, and a comprehensive risk assessment module; The data extraction module is used to capture the outline image of vegetables in real time using the installed industrial camera, and transmit the outline image to the central processing unit in real time via USB interface for image processing and feature extraction to obtain feature data. Then, the feature data is processed to obtain pixel perturbation data group and outline steady state data group. The browning analysis module is used to calculate the oxidation-driven feature index (BDI) based on the pixel perturbation data set, and to perform a browning risk assessment of vegetables by comparing it with the preset browning-driven activation threshold (BAT). When the assessment indicates that there is a browning risk, the module executes the perturbation analysis command. The interference analysis module is used to execute interference analysis instructions, calculate the visual interference rejection index (VDI) through the contour steady-state data set, evaluate abnormal interference with the preset interference anomaly rejection threshold (VIR), and execute the comprehensive structural analysis instructions when the evaluation shows that there is no visual interference. The comprehensive risk assessment module is used to execute comprehensive structural analysis instructions, calculate the structural disintegration consistency index (SCI) based on the oxidation-driven characteristic index (BDI) and the visual interference removal index (VDI), and perform a comprehensive anomaly risk assessment with the preset structural disintegration critical threshold (SCR).

Images