Label missing detection device and method for label printing and cutting all-in-one machine based on machine vision

Abstract

The invention discloses a label missing detection device and method for a label printing and cutting all-in-one machine based on machine vision, and relates to the technical field of machine vision. The method comprises the following steps: triggering an industrial camera through an encoder to collect a real-time image of a label printing area; carrying out gray scale conversion on the real-time image to obtain a gray scale map; performing paging line dynamic correction on the grey-scale image to obtain a corrected image; and substituting the corrected image into an image recognition model to obtain a label missing result. According to the method, image space offset caused by equipment fluctuation is effectively eliminated through paging line dynamic correction, and the identification result is prevented from being interfered by deviation; and the label missing result is output through the image recognition model, so that the automatic and accurate judgment of the label missing state is realized, the label missing detection accuracy is improved, and the label production quality detection efficiency is also improved.

Description

Technical Field

[0001] This invention belongs to the field of machine vision technology, specifically relating to a label missing detection device and method for a machine vision-based label printing and cutting integrated machine. Background Technology

[0002] In the development of machine vision, this technology is responsible for visually monitoring the production process of label printing and cutting integrated machines to ensure the integrity of the label products. Through a series of machine vision facilities, such as industrial cameras, image acquisition modules, and algorithm processing units, the acquired label images are transformed into analyzable visual data. This data is then used to form a detection system through image comparison, feature recognition, and other technical logic to accurately determine the presence or absence of labels.

[0003] Existing technology (publication number: CN116824261A) discloses a method, apparatus, and equipment for product defect detection based on 3D printing. The method involves preprocessing the product's triangular mesh model and acquiring cross-sectional images and information of the model. Based on the cross-sectional information, the defect types in the cross-sectional images are determined and labeled, then used as training data images. These training data images are input into a model built based on the YOLO network for training, resulting in a defect detection model. Finally, a target cross-sectional image of the product to be detected is acquired and input into the defect detection model for detection, yielding the defect detection result. This method improves the efficiency and accuracy of identifying internal defects in shoe soles and reduces the false detection rate.

[0004] The aforementioned patents improve the efficiency and accuracy of identifying internal defects in target objects and reduce the false detection rate. However, in practical applications, although the model built based on the YOLO network has high accuracy, the model inference latency is difficult to meet the production line synchronization requirements of millisecond-level response. At the same time, environmental interference can easily lead to a surge in the false alarm rate. Summary of the Invention

[0005] The purpose of this invention is to address the problem that, in practical applications, although the model built based on the YOLO network has high accuracy, the model inference latency is difficult to meet the production line synchronization requirements of millisecond-level response; at the same time, there is also the problem that environmental interference can easily lead to a surge in false alarm rate. Therefore, this invention proposes a label missing detection device and method for a label printing and cutting integrated machine based on machine vision.

[0006] In a first aspect of this invention, a method for detecting missing labels in a label printing and cutting integrated machine based on machine vision is first proposed, the method comprising: An encoder triggers an industrial camera to capture real-time images of the label printing area; The real-time image is converted to grayscale to obtain a grayscale image; Dynamic correction of page breaks is applied to the grayscale image to obtain the corrected image; Substituting the corrected image into the image recognition model yields the missing label result.

[0007] Optionally, the step of performing dynamic pagination line correction on the grayscale image to obtain the corrected image includes: Perform a Hough transform on the grayscale image to obtain candidate pagination lines; Candidate lines are obtained by least-squares fitting, and candidate angles are calculated based on the slope of the candidate lines. The absolute value of the candidate angle is compared with a preset angle threshold. When the absolute value of the candidate angle is greater than the preset angle threshold, a radial transformation matrix is ​​constructed. The corrected image is obtained by performing bilinear interpolation on the image using the radiometric transformation matrix.

[0008] Optionally, the working principle of the image recognition model includes: Step 1: Downsample the corrected image to obtain the downsampled image; Step 2: Calculate the current level of the downsampled image based on the size of the downsampled image and the size of the corrected image; perform contour matching on the current level and calculate the confidence score; Step 3: If the confidence level is less than the confidence threshold, then downsample the downsampled image and execute Step 2; if the confidence level is less than or equal to the confidence threshold or the maximum number of downsampling attempts is reached, then exit the iteration and obtain the optimal downsampled image. Step 4: Perform missing label identification on the optimal downsampled image to obtain the missing label results.

[0009] Optionally, the confidence level calculation process includes: Where P represents the confidence level, C real C represents the edge contour feature map of the current sampled image. temp The edge contour feature map represents the corrected image; Area() represents the pixel area of ​​the region enclosed by the edge contour, F real F represents the multidimensional feature vector of the current sampled image. temp This represents the multidimensional feature vector of the correction image.

[0010] Optionally, the step of identifying missing labels in the optimal downsampled image to obtain the missing label result includes: The IOU1 value is obtained by calculating the topological constraints based on the optimal downsampled graph; If the IOU1 value is less than the warning threshold, an emergency stop signal is sent, and the missing label areas in the optimal downsampled image are marked, and the marked missing label areas are stored in the missing label sequence, and the missing label sequence is determined as the missing label result; If the IOU1 value is greater than or equal to the warning threshold, then a normal signal is sent. The specific details of the IOU1 value obtained from the topological constraint calculation are as follows: Among them, F real F represents the multidimensional feature vector of the optimal downsampled image. temp D represents the multidimensional feature vector of the corrected image. zc D represents the perimeter of the edge contour of the optimal downsampled image. tb This represents the perimeter of the edge contour of the correction diagram.

[0011] In a second aspect of the invention, a label defect detection device for a machine vision-based label printing and cutting integrated machine is provided, the device comprising: Image acquisition module: The encoder triggers an industrial camera to acquire real-time images of the label printing area; Grayscale module: Converts real-time images to grayscale to obtain grayscale images; Correction module: Performs dynamic correction of pagination lines on the grayscale image to obtain a corrected image; Missing label result module: Substitute the corrected image into the image recognition model to obtain the missing label result.

[0012] In one implementation, the correction module includes: a Hough transform module, a candidate angle module, and a comparison module. The Hough transform module is used to perform a Hough transform on the grayscale image to obtain candidate paging lines; The candidate angle module is used to obtain candidate lines by least squares fitting of candidate lines, and to calculate candidate angles based on the slope of the candidate lines. The comparison module is used to compare the absolute value of the candidate angle with a preset angle threshold. When the absolute value of the candidate angle is greater than the preset angle threshold, a radial transformation matrix is ​​constructed. The radial transformation matrix is ​​used to perform bilinear interpolation correction on the image to obtain a corrected image.

[0013] In one implementation, the missing label result module is further used for: The working principle of the image recognition model includes: Step 1: Downsample the corrected image to obtain the downsampled image; Step 2: Calculate the current level of the downsampled image based on the size of the downsampled image and the size of the corrected image; perform contour matching on the current level and calculate the confidence score; Step 3: If the confidence level is less than the confidence threshold, then downsample the downsampled image and execute Step 2; if the confidence level is less than or equal to the confidence threshold or the maximum number of downsampling attempts is reached, then exit the iteration and obtain the optimal downsampled image. Step 4: Perform missing label identification on the optimal downsampled image to obtain the missing label results.

[0014] In one implementation, the confidence level calculation process includes: Where P represents the confidence level, C real C represents the edge contour feature map of the current sampled image. temp The edge contour feature map represents the corrected image; Area() represents the pixel area of ​​the region enclosed by the edge contour, F real F represents the multidimensional feature vector of the current sampled image. temp This represents the multidimensional feature vector of the correction image.

[0015] In one implementation, the step of identifying missing labels in the optimal downsampled image to obtain the missing label result includes: The IOU1 value is obtained by calculating the topological constraints based on the optimal downsampled graph; If the IOU1 value is less than the warning threshold, an emergency stop signal is sent, and the missing label areas in the optimal downsampled image are marked, and the marked missing label areas are stored in the missing label sequence, and the missing label sequence is determined as the missing label result; If the IOU1 value is greater than or equal to the warning threshold, then a normal signal is sent. The specific details of the IOU1 value obtained from the topological constraint calculation are as follows: Among them, F real F represents the multidimensional feature vector of the optimal downsampled image. temp D represents the multidimensional feature vector of the corrected image. zc D represents the perimeter of the edge contour of the optimal downsampled image. tb This represents the perimeter of the edge contour of the correction diagram.

[0016] The beneficial effects of this invention are: This invention proposes a label defect detection device and method for a machine vision-based label printing and cutting integrated machine. Through image grayscale conversion, it simplifies data dimensions, reduces computational load, and enhances brightness feature differences to lay a solid foundation for subsequent analysis. Dynamic correction of page breaks precisely eliminates image spatial offset caused by equipment fluctuations, ensuring the standardization of image input position and fundamentally preventing deviations from interfering with recognition results. Finally, relying on the image recognition model to output the label defect result, it not only achieves automated and accurate determination of the label defect status but also effectively improves the efficiency of label production quality inspection while reducing human error and increasing detection accuracy. Attached Figure Description

[0017] The invention will now be further described with reference to the accompanying drawings.

[0018] Figure 1 A flowchart illustrating a machine vision-based label printing and cutting integrated machine for missing label detection, provided in an embodiment of the present invention. Figure 2 This is a structural diagram of a label missing detection device for a machine vision-based label printing and cutting integrated machine, provided in an embodiment of the present invention. Detailed Implementation

[0019] 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. The term "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and B can represent: A alone, A and B simultaneously, and B alone. Furthermore, descriptions involving "first," "second," etc., in this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined with "first" or "second" can explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0020] 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.

[0021] This invention provides a machine vision-based method for detecting missing labels in an integrated label printing and cutting machine. See also... Figure 1 , Figure 1 A flowchart illustrating a machine vision-based label printing and cutting integrated machine for missing label detection, provided in an embodiment of the present invention. The method includes the following steps: An encoder triggers an industrial camera to capture real-time images of the label printing area; The real-time image is converted to grayscale to obtain a grayscale image; Dynamic correction of page breaks is applied to the grayscale image to obtain the corrected image; Substituting the corrected image into the image recognition model yields the missing label result.

[0022] Based on the machine vision-based label printing and cutting integrated machine defect detection method provided in this embodiment of the invention, the grayscale conversion of the image simplifies the data dimensions to reduce the computational load and enhances the brightness feature differences; dynamic correction of the page break line effectively eliminates the image spatial offset caused by equipment fluctuations, ensuring the positional standardization of the image input and avoiding deviations from interfering with the recognition results; finally, the defect result is output through the image recognition model, realizing the automated and accurate determination of the defect status, which not only improves the detection accuracy to reduce false judgments, but also improves the efficiency of label production quality inspection.

[0023] Specifically, the real-time image (RBG image) is converted to grayscale to obtain a grayscale image.

[0024] In one implementation, performing dynamic pagination line correction on a grayscale image to obtain a corrected image includes: Perform a Hough transform on the grayscale image to obtain candidate pagination lines; Candidate lines are obtained by least-squares fitting, and candidate angles are calculated based on the slope of the candidate lines. The absolute value of the candidate angle is compared with a preset angle threshold. When the absolute value of the candidate angle is greater than the preset angle threshold, a radial transformation matrix is ​​constructed. The corrected image is obtained by performing bilinear interpolation on the image using the radiometric transformation matrix.

[0025] In one implementation, the input label grayscale image is filtered (e.g., Gaussian filtering, median filtering) to eliminate noise interference and avoid noise affecting the accuracy of subsequent line detection. The preprocessed grayscale image is input into the Hough transform line detection module. By accumulating parameters in polar coordinate space, lines in the image that meet the characteristics of paging lines (e.g., length, pixel density) are identified. All candidate paging lines are filtered and output as a set of parameters (including the slope, intercept or polar angle, and polar radius of the line). Using the pixel coordinates of the candidate paging lines obtained by the Hough transform as sample data, the least squares method is used to fit the line, reducing the detection error and obtaining the optimized candidate line after fitting. Based on the slope k of the optimized candidate line, the candidate angle θ corresponding to the line is calculated using the formula θ=arctan(k). The absolute value of the candidate angle is compared with a preset angle threshold. When the absolute value of the candidate angle is greater than the preset angle threshold, an eccentric transformation matrix is ​​constructed. Finally, using the constructed radial transformation matrix, coordinate transformation calculations are performed on each pixel in the grayscale image; bilinear interpolation algorithm is used to fill the blank pixels generated after the transformation with grayscale values ​​to avoid jagged edges or blurring in the image, and finally the corrected image with page break line correction is output.

[0026] In one implementation, dynamic correction of page break lines achieves significant optimization: Candidate page break lines are extracted based on Hough transform, accurately identifying lines that meet the characteristics and effectively avoiding interference from label patterns and noise, ensuring accurate page break line positioning; least squares fitting of candidate lines optimizes line parameters and reduces detection errors, while candidate angles are calculated through slope, providing a quantitative basis for offset determination; by comparing candidate angles with preset thresholds, only images with excessive offset are subjected to radiometric transformation correction, effectively reducing the overall system load; combining radiometric transformation matrix and bilinear interpolation algorithm for image correction eliminates image spatial offsets caused by equipment vibration, printing position fluctuations, etc., correcting page break line angles and avoiding jagged or blurry distortions, ensuring the clarity and positional regularity of the output corrected image, ultimately providing high-quality image input for subsequent missing label recognition, significantly improving the accuracy of missing label detection results; the preset threshold is set by staff based on historical experience.

[0027] In one implementation, the working principle of the image recognition model includes: Step 1: Downsample the corrected image to obtain the downsampled image; Step 2: Calculate the current level of the downsampled image based on the size of the downsampled image and the size of the corrected image; perform contour matching on the current level and calculate the confidence score; Step 3: If the confidence level is less than the confidence threshold, then downsample the downsampled image and execute Step 2; if the confidence level is less than or equal to the confidence threshold or the maximum number of downsampling attempts is reached, then exit the iteration and obtain the optimal downsampled image. Step 4: Perform missing label identification on the optimal downsampled image to obtain the missing label results.

[0028] In one implementation, the image recognition model achieves significant technical effects through a collaborative mechanism of multi-round iterative downsampling and contour matching: the corrected image is downsampled layer by layer and the current level is determined by combining size calculation. Then, the confidence score is calculated by contour matching, which can adaptively select the optimal downsampled image with the highest matching degree with the label features, effectively avoiding the problem of computational redundancy caused by excessively high image resolution or feature loss caused by excessively low resolution. At the same time, through the dual constraints of confidence threshold and maximum number of downsampling times, the controllability of the iteration process is ensured, invalid loops are avoided, and the effective feature regions of the label can be accurately located. Finally, the missing label is identified based on the optimal downsampled image, which greatly improves the efficiency and accuracy of missing label detection.

[0029] In one implementation, the confidence level calculation process includes: Where P represents the confidence level, C real C represents the edge contour feature map of the current sampled image.temp The edge contour feature map represents the corrected image; Area() represents the pixel area of ​​the region enclosed by the edge contour, F real F represents the multidimensional feature vector of the current sampled image. temp This represents the multidimensional feature vector of the correction image.

[0030] In one implementation, the missing label identification process for the optimal downsampled image to obtain the missing label result includes: The IOU1 value is obtained by calculating the topological constraints based on the optimal downsampled graph; If the IOU1 value is less than the warning threshold, an emergency stop signal is sent, and the missing label areas in the optimal downsampled image are marked and stored in the missing label sequence. The missing label sequence is then determined as the missing label result. If the IOU1 value is greater than or equal to the warning threshold, a normal signal is sent; the warning threshold is set based on the staff's historical experience. Specifically, emergency stop signals and normal signals are signals sent by the system to the backend for later processing; The specific values ​​of IOU1 obtained from the topological constraint calculation are as follows: Among them, F real F represents the multidimensional feature vector of the optimal downsampled image. temp D represents the multidimensional feature vector of the corrected image. zc D represents the perimeter of the edge contour of the optimal downsampled image. tb This represents the perimeter of the edge contour of the correction diagram.

[0031] Based on the same inventive concept, embodiments of the present invention also provide a machine vision-based label printing and cutting integrated machine for detecting missing labels. See also Figure 2 , Figure 2 A structural diagram of the label missing detection device for a machine vision-based label printing and cutting integrated machine provided in this embodiment of the invention includes: Image acquisition module: The encoder triggers an industrial camera to acquire real-time images of the label printing area; Grayscale module: Converts real-time images to grayscale to obtain grayscale images; Correction module: Performs dynamic correction of pagination lines on the grayscale image to obtain a corrected image; Missing label results module: Substitute the corrected image into the image recognition model to obtain the missing label results.

[0032] Based on the machine vision-based label printing and cutting integrated machine defect detection device provided in this embodiment of the invention, the image is converted to grayscale, which simplifies the data dimensions to reduce the computational load and enhances the brightness feature differences; the dynamic correction of the page break line effectively eliminates the image spatial offset caused by equipment fluctuations, ensures the positional standardization of the image input, and avoids deviations from interfering with the recognition results; finally, the defect result is output through the image recognition model, realizing the automated and accurate determination of the defect status, which not only improves the detection accuracy to reduce false judgments, but also improves the efficiency of label production quality inspection.

[0033] In one implementation, the correction module includes: a Hough transform module, a candidate angle module, and a comparison module. The Hough transform module is used to perform the Hough transform on a grayscale image to obtain candidate paging lines; The candidate angle module is used to obtain candidate lines by least squares fitting, and to calculate candidate angles based on the slope of the candidate lines. The comparison module is used to compare the absolute value of the candidate angle with a preset angle threshold. When the absolute value of the candidate angle is greater than the preset angle threshold, a radial transformation matrix is ​​constructed. The radial transformation matrix is ​​used to perform bilinear interpolation correction on the image to obtain the corrected image.

[0034] In one implementation, the missing label result module is also used for: The working principle of image recognition models includes: Step 1: Downsample the corrected image to obtain the downsampled image; Step 2: Calculate the current level of the downsampled image based on the size of the downsampled image and the size of the corrected image; perform contour matching on the current level and calculate the confidence score; Step 3: If the confidence level is less than the confidence threshold, then downsample the downsampled image and execute Step 2; if the confidence level is less than or equal to the confidence threshold or the maximum number of downsampling attempts is reached, then exit the iteration and obtain the optimal downsampled image. Step 4: Perform missing label identification on the optimal downsampled image to obtain the missing label results.

[0035] In one implementation, the confidence level calculation process includes: Where P represents the confidence level, C real C represents the edge contour feature map of the current sampled image. temp The edge contour feature map represents the corrected image; Area() represents the pixel area of ​​the region enclosed by the edge contour, F real F represents the multidimensional feature vector of the current sampled image. temp This represents the multidimensional feature vector of the correction image.

[0036] In one implementation, the missing label identification process for the optimal downsampled image to obtain the missing label result includes: The IOU1 value is obtained by calculating the topological constraints based on the optimal downsampled graph; If the IOU1 value is less than the warning threshold, an emergency stop signal is sent, and the missing label areas in the optimal downsampled image are marked and stored in the missing label sequence. The missing label sequence is then determined as the missing label result. If the IOU1 value is greater than or equal to the warning threshold, then a normal signal is sent. The specific values ​​of IOU1 obtained from the topological constraint calculation are as follows: Among them, F real F represents the multidimensional feature vector of the optimal downsampled image. temp D represents the multidimensional feature vector of the corrected image. zc D represents the perimeter of the edge contour of the optimal downsampled image. tb This represents the perimeter of the edge contour of the correction diagram.

[0037] The foregoing has described one embodiment of the present invention in detail, but this content is merely a preferred embodiment and should not be considered as limiting the scope of the present invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the scope of the claims of this invention.

Claims

1. A method for detecting missing labels in a machine vision-based label printing and cutting integrated machine, characterized in that, The method includes: An encoder triggers an industrial camera to capture real-time images of the label printing area; The real-time image is converted to grayscale to obtain a grayscale image; Dynamic correction of page breaks is applied to the grayscale image to obtain the corrected image; Substituting the corrected image into the image recognition model yields the missing label result.

2. The method for detecting missing labels in a machine vision-based label printing and cutting integrated machine according to claim 1, characterized in that, The process of performing dynamic pagination line correction on the grayscale image to obtain the corrected image includes: Perform a Hough transform on the grayscale image to obtain candidate pagination lines; Candidate lines are obtained by least-squares fitting, and candidate angles are calculated based on the slope of the candidate lines. The absolute value of the candidate angle is compared with a preset angle threshold. When the absolute value of the candidate angle is greater than the preset angle threshold, a radial transformation matrix is ​​constructed. The corrected image is obtained by performing bilinear interpolation on the image using the radiometric transformation matrix.

3. The method for detecting missing labels in a machine vision-based label printing and cutting integrated machine according to claim 1, characterized in that, The working principle of the image recognition model includes: Step 1: Downsample the corrected image to obtain the downsampled image; Step 2: Calculate the current level of the downsampled image based on the size of the downsampled image and the size of the corrected image; perform contour matching on the current level and calculate the confidence score; Step 3: If the confidence level is less than the confidence threshold, then downsample the downsampled image and execute Step 2; if the confidence level is less than or equal to the confidence threshold or the maximum number of downsampling attempts is reached, then exit the iteration and obtain the optimal downsampled image. Step 4: Perform missing label identification on the optimal downsampled image to obtain the missing label results.

4. The method for detecting missing labels in a machine vision-based label printing and cutting integrated machine according to claim 3, characterized in that, The confidence level calculation process includes: Where P represents the confidence level, C real C represents the edge contour feature map of the current sampled image. temp The edge contour feature map represents the corrected image; Area() represents the pixel area of ​​the region enclosed by the edge contour, F real F represents the multidimensional feature vector of the current sampled image. temp This represents the multidimensional feature vector of the correction image.

5. The method for detecting missing labels in a machine vision-based label printing and cutting integrated machine according to claim 3, characterized in that, The missing label identification process for the optimal downsampled image to obtain the missing label result includes: The IOU1 value is obtained by calculating the topological constraints based on the optimal downsampled graph; If the IOU1 value is less than the warning threshold, an emergency stop signal is sent, and the missing label areas in the optimal downsampled image are marked, and the marked missing label areas are stored in the missing label sequence, and the missing label sequence is determined as the missing label result; If the IOU1 value is greater than or equal to the warning threshold, then a normal signal is sent. The specific details of the IOU1 value obtained from the topological constraint calculation are as follows: Among them, F real F represents the multidimensional feature vector of the optimal downsampled image. temp D represents the multidimensional feature vector of the corrected image. zc D represents the perimeter of the edge contour of the optimal downsampled image. tb This represents the perimeter of the edge contour of the correction diagram.

6. A label defect detection device for a machine vision-based label printing and cutting integrated machine, characterized in that, The device includes: Image acquisition module: The encoder triggers an industrial camera to acquire real-time images of the label printing area; Grayscale module: Converts real-time images to grayscale to obtain grayscale images; Correction module: Performs dynamic correction of pagination lines on the grayscale image to obtain a corrected image; Missing label result module: Substitute the corrected image into the image recognition model to obtain the missing label result.

7. The label defect detection device for a machine vision-based label printing and cutting integrated machine according to claim 6, characterized in that, The correction module includes: a Hough transform module, a candidate angle module, and a comparison module. The Hough transform module is used to perform a Hough transform on the grayscale image to obtain candidate paging lines; The candidate angle module is used to obtain candidate lines by least squares fitting of candidate lines, and to calculate candidate angles based on the slope of the candidate lines. The comparison module is used to compare the absolute value of the candidate angle with a preset angle threshold. When the absolute value of the candidate angle is greater than the preset angle threshold, a radial transformation matrix is ​​constructed. The radial transformation matrix is ​​used to perform bilinear interpolation correction on the image to obtain a corrected image.

8. The label defect detection device for a machine vision-based label printing and cutting integrated machine according to claim 6, characterized in that, The missing label result module is also used for: The working principle of the image recognition model includes: Step 1: Downsample the corrected image to obtain the downsampled image; Step 2: Calculate the current level of the downsampled image based on the size of the downsampled image and the size of the corrected image; perform contour matching on the current level and calculate the confidence score; Step 3: If the confidence level is less than the confidence threshold, then downsample the downsampled image and execute Step 2; if the confidence level is less than or equal to the confidence threshold or the maximum number of downsampling attempts is reached, then exit the iteration and obtain the optimal downsampled image. Step 4: Perform missing label identification on the optimal downsampled image to obtain the missing label results.

9. The label defect detection device for a machine vision-based label printing and cutting integrated machine according to claim 8, characterized in that, The confidence level calculation process includes: Where P represents the confidence level, C real C represents the edge contour feature map of the current sampled image. temp The edge contour feature map represents the corrected image; Area() represents the pixel area of ​​the region enclosed by the edge contour, F real F represents the multidimensional feature vector of the current sampled image. temp This represents the multidimensional feature vector of the correction image.

10. The label defect detection device for a machine vision-based label printing and cutting integrated machine according to claim 6, characterized in that, The missing label identification process for the optimal downsampled image to obtain the missing label result includes: The IOU1 value is obtained by calculating the topological constraints based on the optimal downsampled graph; If the IOU1 value is less than the warning threshold, an emergency stop signal is sent, and the missing label areas in the optimal downsampled image are marked, and the marked missing label areas are stored in the missing label sequence, and the missing label sequence is determined as the missing label result; If the IOU1 value is greater than or equal to the warning threshold, then a normal signal is sent. The specific details of the IOU1 value obtained from the topological constraint calculation are as follows: Among them, F real F represents the multidimensional feature vector of the optimal downsampled image. temp D represents the multidimensional feature vector of the corrected image. zc D represents the perimeter of the edge contour of the optimal downsampled image. tb This represents the perimeter of the edge contour of the correction diagram.

Images