Multi-dimensional barcode detection method and detection device based on deep learning

Abstract

The present application relates to the technical field of bar code detection, and in particular to a multi-dimensional bar code detection method and device based on deep learning. The method comprises the following steps: S1: acquiring a basic image to be recognized, and performing edge detection on the basic image based on an edge algorithm to identify the corner points of the multi-dimensional bar code; S2: when the corner points of the multi-dimensional bar code cannot be identified, identifying and eliminating the invalid edges of the basic image through a pre-trained multi-dimensional bar code edge processing model, further enlarging the image, identifying the corner points of the multi-dimensional bar code, and confirming the position of the multi-dimensional bar code; and S3: identifying the multi-dimensional bar code, and if the multi-dimensional bar code can be identified, the next multi-dimensional bar code position is acquired and identified through steps S1 and S2; and if the multi-dimensional bar code cannot be identified, the device provided by the system can reduce the influence of the environment background, light, and wear of the bar code itself during the identification process.

Description

Technical Field

[0001] This invention relates to the field of barcode detection technology, specifically to a multidimensional barcode detection method and device based on deep learning. Background Technology

[0002] Barcode identification and inspection refers to the process of detecting and evaluating the quality and conformity of barcode symbols through a series of technical means. The purpose of barcode identification and inspection is to ensure that barcode symbols can be accurately and efficiently read by scanning equipment, thereby guaranteeing the accurate transmission of product information. The core principle of barcode identification and inspection is based on optical scanning and digital encoding technology. A barcode consists of a series of alternating black and white stripes. When the barcode is illuminated by light emitted from a scanner, the light is reflected or absorbed according to the arrangement of the stripes and spaces, forming specific light signals. These light signals are captured by the photoelectric converter inside the scanner and converted into electrical signals, which are then analyzed by a decoder, ultimately converting the electrical signals into digital or character information.

[0003] Typically, barcodes are placed on the outer packaging of products or on electronic screens. However, due to changes in environmental background, light source, or wear and tear on the barcode, it may become unreadable or difficult to read. This may require operators to re-code the barcode or render the entire barcode invalid, affecting the entire business process.

[0004] Based on the above problems, there is an urgent need for a barcode detection device that can quickly eliminate the effects of ambient light background and damage to the barcode itself. Summary of the Invention

[0005] Current barcode detection devices and methods often fail to recognize barcodes when exposed to overexposed or underexposed backgrounds or when the barcode is damaged, requiring manual adjustments or even invalidation, causing unnecessary trouble. This application provides a deep learning-based multidimensional barcode detection method and device to address these issues.

[0006] To achieve the above objectives, the present invention provides the following technical solution: This application discloses a multidimensional barcode detection method based on deep learning, including: S1: Acquire the base image to be recognized, and perform edge detection on the base image based on the edge algorithm to identify the corner points of the multidimensional barcode; S2: When the corners of the multidimensional barcode cannot be identified, the invalid edges of the base image are identified and removed by the pre-trained multidimensional barcode edge processing model. The image is then further enlarged to identify the corners of the multidimensional barcode and confirm its location. S3: Recognize the multidimensional barcode. If it can be recognized, continue to execute steps S1 and S2 to obtain and recognize the next multidimensional barcode position. If the multidimensional barcode cannot be recognized, judge the anomaly type of the multidimensional barcode based on the multidimensional barcode anomaly judgment model to determine whether it is an exposure anomaly or a multidimensional barcode wear anomaly. S4: When the multidimensional barcode is exposed abnormally, adjust the brightness of the image and perform recognition. If recognition is not possible within the set brightness adjustment range, identify the abnormal part of the multidimensional barcode and correct the multidimensional barcode through a correction algorithm. When the multidimensional barcode is worn abnormally, identify the abnormal part of the multidimensional barcode and correct the multidimensional barcode through a correction algorithm.

[0007] The above technical solution, by setting up a corresponding multi-dimensional barcode edge processing model, can identify and remove invalid edges of the base image, preventing the barcode from being difficult to identify and locate due to environmental background; and sets up a corresponding correction algorithm to correct exposure abnormalities and barcode wear abnormalities, ensuring effective barcode recognition.

[0008] Preferably, the processing steps of the multidimensional barcode edge processing model are as follows: S201: A large number of multidimensional barcode basic feature recognition training are performed through a multidimensional barcode edge processing model, and parts of the basic image that do not conform to the basic features of multidimensional barcodes are removed. S202: Retain the possible image portions where the basic feature matching degree exceeds the set threshold, and further enlarge the possible images; S203: Perform grayscale conversion and bilateral filtering on the possible images, and input the filtered and grayscale images into a pre-trained edge processing model that can identify image edges to remove the edges, while retaining the multidimensional barcode processing part.

[0009] The above technical solution further refines the processing steps of the multidimensional barcode edge processing model, enabling the removal of parts of the base image that do not conform to the basic features of the multidimensional barcode, effectively preserving the multidimensional barcode processing part, and preventing the influence of the base image environment background on barcode recognition.

[0010] More preferably, the multidimensional barcode edge processing model is trained using a dynamically adjusted learning rate; and the multidimensional barcode image edge extraction model uses the softmax algorithm in reinforcement learning to probabilistically output the results. The softmax algorithm formula is as follows: , In the formula This represents the average value from point 1 to point t. To reinforce the parameters of the fire-changing strategy in reinforcement learning.

[0011] The above technical solution provides a multi-dimensional barcode image edge extraction model algorithm, which can effectively extract and filter invalid edge parts of the base image.

[0012] More preferably, the training steps of the multidimensional barcode edge processing model are as follows: S211: Extract and remove multidimensional barcodes from the base image using a convolutional neural network with initial weights; S212: Input the base image into the convolutional neural network, and calculate the similarity between the template image containing the actual edge contour of the multidimensional barcode and the base image after edge extraction. S213: Determine whether the calculated similarity exceeds a set threshold; if not, adjust and update the weights of the convolutional neural network according to a preset adjustment rule. S214: The convolutional neural network uses the updated weights to extract the edges of the multidimensional barcode in the base image again and performs similarity judgment again; S215: If so, return to the step of inputting the base image into the convolutional neural network to input the next base image into the convolutional neural network for training, until all training images are trained, output the current convolutional neural network, and obtain the multidimensional barcode edge processing model.

[0013] The above technical solution further refines the training steps of the multidimensional barcode edge processing model, which can effectively remove invalid edges from the base image.

[0014] More preferably, the judgment steps of the multidimensional barcode anomaly judgment model are as follows: S301: Identify and extract abnormal regions of unrecognizable multidimensional barcodes using a pre-trained multidimensional barcode anomaly detection model; S302: Adjust the brightness of unrecognizable multidimensional barcode images upwards and downwards to obtain adjusted images; S303: Compare and judge the white and black pixel blocks of unrecognizable multidimensional barcode images and adjusted images, and calculate the pixel dissimilarity rate; The pixel dissimilarity rate is the percentage of the number of dissimilarity color blocks in the multidimensional barcode image and the adjusted image out of the total number of color blocks; S304: When the pixel anisotropy rate exceeds the set judgment threshold, it is judged as an exposure abnormality; when the pixel anisotropy rate is lower than the set judgment threshold, it is judged as a multidimensional barcode wear abnormality.

[0015] The above technical solution can process the brightness of multidimensional barcode images, calculate the pixel dissimilarity rate, and determine whether the multidimensional barcode image is exposed abnormally or worn abnormally.

[0016] More preferably, the correction process of the correction algorithm is as follows: Obtain each abnormal region of the multidimensional barcode. Using the center of each abnormal region as the first center, draw a first circular region outwards, completely covering the abnormal regions. Identify the first radius of the circular region, and draw a second circular region outwards from the first center, with the second radius being twice the first radius. Within the first circular region, draw n concentric circles centered on the first center, and encode each concentric circle. Determine the blocks within the concentric circles that need adjustment. Obtain normal image units overlapping the black and white barcode areas of the second circular region with the outermost concentric circle region. Identify the part with the highest edge influence feature value in the clearly contacted image portion, and perform extension and stitching correction processing with the outermost concentric circle. After completing the extension and stitching correction processing of the outermost concentric circle, perform extension and stitching correction processing on the inner concentric circles sequentially until all concentric circles are corrected.

[0017] The above technical solution provides a correction algorithm that, by processing and piecing together multidimensional barcodes, can effectively correct multidimensional barcode anomalies caused by exposure abnormalities and wear abnormalities, enabling previously unrecognizable abnormal multidimensional barcodes to be effectively identified.

[0018] A detection device, applied to the deep learning-based multidimensional barcode detection method as described in any one of the above claims, comprising: Multidimensional barcode extraction and recognition unit: used to acquire the base image to be recognized, and to perform edge detection on the base image based on edge algorithms to identify the corner points of the multidimensional barcode; Invalid edge removal unit: When the multidimensional barcode extraction and recognition unit cannot identify the corner points of the multidimensional barcode, the invalid edges of the base image are identified and removed by the pre-trained multidimensional barcode edge processing model. Multidimensional barcode anomaly detection unit: When the barcode cannot be recognized, it can determine the anomaly type of the multidimensional barcode based on the multidimensional barcode anomaly detection model, and determine whether it is an exposure anomaly or a multidimensional barcode wear anomaly. Multidimensional barcode correction unit: It can correct multidimensional barcode images with exposure errors and barcode wear errors through correction algorithms.

[0019] Preferably, the invalid edge removal unit includes: Edge removal module: Through a large amount of training on the basic feature recognition of multidimensional barcodes using a multidimensional barcode edge processing model, parts of the basic image that do not conform to the basic features of multidimensional barcodes are removed. The first processing module: preserves the possible image parts whose basic feature matching degree exceeds the set threshold, and further magnifies the possible images; The second processing module performs grayscale conversion and bilateral filtering on the possible images, and inputs the filtered and grayscale images into a pre-trained edge processing module that can identify image edges to remove the edges, while retaining the multidimensional barcode processing part. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart of the deep learning-based multidimensional barcode detection method of this application; Figure 2 This is a flowchart of the first embodiment of step S2 in Figure 1 of this application; Figure 3 This is a flowchart of the second embodiment of step S2 in Figure 1 of this application; Figure 4 This is a detailed flowchart of step S3 in this application.

[0022] Figure 5 This is a block diagram of the connection of the testing device in this application.

[0023] In the picture: 1. Multidimensional barcode extraction and recognition unit; 2. Invalid edge removal unit; 3. Edge removal module; 4. First processing module; 5. Second processing module; 6. Multidimensional barcode anomaly judgment unit; 7. Multidimensional barcode correction unit. Detailed Implementation

[0024] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0025] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0026] Please see Figures 1-5 While existing barcode detection devices and methods can effectively and quickly confirm and identify barcode positions, they are not ideal for certain situations. These include environments with significant background interference, excessively dark or overexposed conditions, or barcode wear. Therefore, this application discloses a deep learning-based multidimensional barcode detection method, including: S1: Acquire the base image to be recognized and perform edge detection on the base image based on edge algorithms to identify the corner points of the multidimensional barcode. Edge algorithms are algorithms that run at the network edge, close to the data source. Relying on an open platform integrating core capabilities such as network, computing, storage, and applications, they can provide front-end services locally. Unlike traditional cloud computing models, applications start directly at the edge, greatly shortening the data transmission distance and making network service response faster, meeting basic requirements for real-time business, application intelligence, security, and privacy protection. Applying edge algorithms to barcode recognition, by identifying the corner points of the multidimensional barcode, can effectively confirm the location of the barcode.

[0027] S2: When the corner points of the multidimensional barcode cannot be identified, the invalid edges of the base image are identified and removed using a pre-trained multidimensional barcode edge processing model. The image is then further magnified to identify the corner points of the multidimensional barcode and confirm its location. It's worth noting that the pre-trained multidimensional barcode edge processing model provided in this solution is trained by loading a large number of similar multidimensional barcodes and collecting a large number of basic barcode features. This effectively identifies feature parts in the base image that do not belong to the multidimensional barcode, removes these non-multidimensional barcode features, and retains the easily identifiable possible multidimensional barcode parts. It should be noted that the above solution is not a single process. After the first removal of invalid edges from the base image and magnification, if the corner points of the barcode still cannot be identified, the pre-trained multidimensional barcode edge processing model can be used again to remove invalid edges from the magnified image, retaining the identifiable multidimensional barcode parts and further confirming the barcode's location.

[0028] S3: Recognize the multidimensional barcode. If it can be recognized, continue with steps S1 and S2 to obtain and recognize the next multidimensional barcode position. If the multidimensional barcode cannot be recognized, determine the type of anomaly based on the multidimensional barcode anomaly judgment model, identifying whether it is an exposure anomaly or a multidimensional barcode wear anomaly. After confirming the barcode position in step S2, recognition is performed. If it cannot be recognized, it may be because the entire image is overexposed or too dark, causing part of the barcode to be visible and thus unrecognizable, or because the barcode itself is partially damaged due to handling or other reasons. This step can identify the reason why the barcode cannot be recognized, determine whether it is an exposure anomaly or a barcode wear anomaly, and provide corresponding solutions for the two anomaly types.

[0029] S4: When a multidimensional barcode exhibits an exposure anomaly, the image brightness is adjusted and recognition is performed. If recognition fails within the set brightness adjustment range, the abnormal portion of the multidimensional barcode is identified and corrected using a correction algorithm. Similarly, when a multidimensional barcode exhibits a wear anomaly, the abnormal portion is identified and corrected using the same algorithm. After confirming the type of barcode anomaly, corresponding solutions can be implemented for different anomaly types. For example, in the case of an exposure anomaly, brightness adjustment allows the unrecognizable portion of the multidimensional barcode to be identified. If brightness adjustment still fails to reveal the anomaly, it can be classified as a wear anomaly and corrected using the same method, employing a correction algorithm to fill in any missing or unrecognizable blurred portions.

[0030] It is worth mentioning that the method provided in this application can detect multidimensional barcodes, including barcodes, QR codes and three-dimensional barcodes. The above scheme can achieve identification and positioning by setting up a corresponding multidimensional barcode edge processing model; and set up a corresponding correction algorithm to perform corresponding correction processing for exposure abnormalities and barcode wear abnormalities, so as to ensure effective barcode recognition.

[0031] Preferably, the processing steps of the multidimensional barcode edge processing model are as follows: S201: A large number of multidimensional barcode basic feature recognition training are performed through a multidimensional barcode edge processing model. Parts in the basic image that do not conform to the basic features of multidimensional barcodes are removed. This scheme loads a large number of multidimensional barcodes and classifies and trains the various features of multidimensional barcodes. When recognizing the basic image, the various features of the barcode are compared with the basic image, and parts that do not belong to the features of multidimensional barcodes are removed.

[0032] S202: Retain the possible image portions where the basic feature matching degree exceeds the set threshold, and further enlarge the possible images; S203: Perform grayscale conversion and bilateral filtering on the possible images, and input the filtered and grayscale images into a pre-trained edge processing model that can identify image edges to remove the edges, while retaining the multidimensional barcode processing part.

[0033] It is worth mentioning that the above scheme further refines the processing steps of the multidimensional barcode edge processing model, which can remove parts of the basic image that do not conform to the basic features of the multidimensional barcode, effectively retain the multidimensional barcode processing part, and prevent the influence of the background environment of the basic image on barcode recognition.

[0034] The multidimensional barcode edge processing model is trained using a dynamically adjusted learning rate; and the multidimensional barcode image edge extraction model uses the softmax algorithm in reinforcement learning to probabilistically output the results. The softmax algorithm formula is as follows: , In the formula This represents the average value from point 1 to point t. To reinforce the parameters of the fire-changing strategy in reinforcement learning.

[0035] It is worth mentioning that the above solution provides a multi-dimensional barcode image edge extraction model algorithm, which can effectively extract and filter invalid edge parts of the base image.

[0036] The training steps for the multidimensional barcode edge processing model are as follows: S211: Extract and remove multidimensional barcodes from the base image using a convolutional neural network with initial weights; S212: Input the base image into the convolutional neural network, and calculate the similarity between the template image containing the actual edge contour of the multidimensional barcode and the base image after edge extraction. S213: Determine whether the calculated similarity exceeds a set threshold; if not, adjust and update the weights of the convolutional neural network according to a preset adjustment rule. S214: The convolutional neural network uses the updated weights to extract the edges of the multidimensional barcode in the base image again and performs similarity judgment again; S215: If so, return to the step of inputting the base image into the convolutional neural network to input the next base image into the convolutional neural network for training, until all training images are trained, output the current convolutional neural network, and obtain the multidimensional barcode edge processing model.

[0037] The above scheme further refines the training steps of the multidimensional barcode edge processing model, which can effectively remove invalid edges from the base image.

[0038] The judgment steps of the multidimensional barcode anomaly judgment model are as follows: S301: Identify and extract abnormal regions of unrecognizable multidimensional barcodes using a pre-trained multidimensional barcode anomaly detection model; S302: Adjust the brightness of unrecognizable multidimensional barcode images upwards and downwards to obtain adjusted images; S303: Compare and judge the white and black pixel blocks of unrecognizable multidimensional barcode images and adjusted images, and calculate the pixel dissimilarity rate; The pixel dissimilarity rate is the percentage of the number of dissimilarity color blocks in the multidimensional barcode image and the adjusted image out of the total number of color blocks; S304: When the pixel anisotropy rate exceeds the set judgment threshold, it is judged as an exposure abnormality; when the pixel anisotropy rate is lower than the set judgment threshold, it is judged as a multidimensional barcode wear abnormality.

[0039] It is worth mentioning that the above solution can process the brightness of multidimensional barcode images, calculate the pixel dissimilarity rate, and determine whether the multidimensional barcode image is exposed abnormally or worn abnormally.

[0040] More preferably, the correction process of the correction algorithm is as follows: Obtain each abnormal region of the multidimensional barcode. Using the center of each abnormal region as the first center, draw a first circular region outwards, completely covering the abnormal regions. Identify the first radius of the circular region, and draw a second circular region outwards from the first center, with the second radius being twice the first radius. Within the first circular region, draw n concentric circles centered on the first center, and encode each concentric circle. Determine the blocks within the concentric circles that need adjustment. Obtain normal image units overlapping the black and white barcode areas of the second circular region with the outermost concentric circle region. Identify the part with the highest edge influence feature value in the clearly contacted image portion, and perform extension and stitching correction processing with the outermost concentric circle. After completing the extension and stitching correction processing of the outermost concentric circle, perform extension and stitching correction processing on the inner concentric circles sequentially until all concentric circles are corrected.

[0041] The formula for calculating the edge influence feature value is: ; ; ; in, Indicates the eigenvalues ​​that influence edge effects. Let be the edge contact coefficient of the r-th concentric circle in the image to be processed. This represents the difference in edge contact coefficients between adjacent concentric circles. Represents a normalized linear mapping function. This is the maximum brightness value. The region containing the concentric circles. These are the parameters for the normalized linear mapping.

[0042] The above solution provides a correction algorithm that, by processing and piecing together multidimensional barcodes, can effectively correct multidimensional barcode anomalies caused by exposure abnormalities and wear abnormalities, enabling previously unrecognizable abnormal multidimensional barcodes to be effectively identified.

[0043] A detection device, applied to the deep learning-based multidimensional barcode detection method as described in any one of the above claims, comprising: Multidimensional barcode extraction and recognition unit 1: used to acquire the base image to be recognized, and to perform edge detection on the base image based on the edge algorithm to identify the corner points of the multidimensional barcode; Invalid edge removal unit 2: When the multidimensional barcode extraction and recognition unit 1 fails to identify the corner points of the multidimensional barcode, the invalid edges of the base image are identified and removed by the pre-trained multidimensional barcode edge processing model. Multidimensional barcode anomaly judgment unit 6: When the barcode cannot be recognized, it can judge the anomaly type of the multidimensional barcode based on the multidimensional barcode anomaly judgment model, and judge whether it is an exposure anomaly or a multidimensional barcode wear anomaly. Multidimensional barcode correction unit 7: It can correct multidimensional barcode images with exposure abnormalities and barcode wear abnormalities through correction algorithms.

[0044] Preferably, the invalid edge removal unit 2 includes: Edge Removal Module 3: Through a large amount of training on the basic feature recognition of multidimensional barcodes using a multidimensional barcode edge processing model, parts of the basic image that do not conform to the basic features of multidimensional barcodes are removed; First processing module 4: Retains possible image portions where the basic feature matching degree exceeds a set threshold, and further magnifies the possible images; The second processing module 5 performs grayscale conversion and bilateral filtering on the possible images, and inputs the filtered and grayscale possible images into a pre-trained edge processing module that can identify image edges to remove the edges and retain the multidimensional barcode processing part.

[0045] Unless otherwise specified, the equipment components involved in the above embodiments are all conventional equipment components, and the connection methods and control methods involved are all conventional connection methods and control methods unless otherwise specified.

[0046] The present invention has been described in detail above with reference to the embodiments. However, those skilled in the art will understand that, without departing from the spirit of the present invention, various specific parameters in the above embodiments can be changed to form multiple specific embodiments, all of which are common variations of the present invention, and will not be described in detail here.

Claims

1. A multidimensional barcode detection method based on deep learning, characterized in that, include: S1: Acquire the base image to be recognized, and perform edge detection on the base image based on the edge algorithm to identify the corner points of the multidimensional barcode; S2: When the corners of the multidimensional barcode cannot be identified, the invalid edges of the base image are identified and removed by the pre-trained multidimensional barcode edge processing model. The image is then further enlarged to identify the corners of the multidimensional barcode and confirm its location. S3: Recognize the multidimensional barcode. If it can be recognized, continue to execute steps S1 and S2 to obtain and recognize the next multidimensional barcode position. If the multidimensional barcode cannot be recognized, judge the anomaly type of the multidimensional barcode based on the multidimensional barcode anomaly judgment model to determine whether it is an exposure anomaly or a multidimensional barcode wear anomaly. S4: When the multidimensional barcode is exposed abnormally, adjust the brightness of the image and perform recognition. If recognition is not possible within the set brightness adjustment range, identify the abnormal part of the multidimensional barcode and correct the multidimensional barcode through a correction algorithm. When the multidimensional barcode is worn abnormally, identify the abnormal part of the multidimensional barcode and correct the multidimensional barcode through a correction algorithm.

2. The deep learning-based multidimensional barcode detection method according to claim 1, characterized in that, The processing steps of the multidimensional barcode edge processing model are as follows: S201: A large number of multidimensional barcode basic feature recognition training are performed through a multidimensional barcode edge processing model, and parts of the basic image that do not conform to the basic features of multidimensional barcodes are removed. S202: Retain the possible image portions where the basic feature matching degree exceeds the set threshold, and further enlarge the possible images; S203: Perform grayscale conversion and bilateral filtering on the possible images, and input the filtered and grayscale images into a pre-trained edge processing model that can identify image edges to remove the edges, while retaining the multidimensional barcode processing part.

3. The deep learning-based multidimensional barcode detection method according to claim 2, characterized in that, The multidimensional barcode edge processing model is trained using a dynamically adjusted learning rate; and the multidimensional barcode image edge extraction model uses the softmax algorithm in reinforcement learning to probabilistically output the results. The softmax algorithm formula is as follows: ， In the formula This represents the average value from point 1 to point t. To reinforce the parameters of the fire-changing strategy in reinforcement learning.

4. The deep learning-based multidimensional barcode detection method according to claim 2, characterized in that, The training steps for the multidimensional barcode edge processing model are as follows: S211: Extract and remove multidimensional barcodes from the base image using a convolutional neural network with initial weights; S212: Input the base image into the convolutional neural network, and calculate the similarity between the template image containing the actual edge contour of the multidimensional barcode and the base image after edge extraction. S213: Determine whether the calculated similarity exceeds the set threshold; No, then the weights of the convolutional neural network are adjusted and updated according to the preset adjustment rules. S214: The convolutional neural network uses the updated weights to extract the edges of the multidimensional barcode in the base image again and performs similarity judgment again; S215: If so, return to the step of inputting the base image into the convolutional neural network to input the next base image into the convolutional neural network for training, until all training images are trained, output the current convolutional neural network, and obtain the multidimensional barcode edge processing model.

5. The deep learning-based multidimensional barcode detection method according to claim 1, characterized in that, The judgment steps of the multidimensional barcode anomaly judgment model are as follows: S301: Identify and extract abnormal regions of unrecognizable multidimensional barcodes using a pre-trained multidimensional barcode anomaly detection model; S302: Adjust the brightness of unrecognizable multidimensional barcode images upwards and downwards to obtain adjusted images; S303: Compare and judge the white and black pixel blocks of unrecognizable multidimensional barcode images and adjusted images, and calculate the pixel dissimilarity rate; The pixel dissimilarity rate is the percentage of the number of dissimilarity color blocks in the multidimensional barcode image and the adjusted image out of the total number of color blocks; S304: When the pixel anisotropy rate exceeds the set judgment threshold, it is judged as an exposure abnormality; when the pixel anisotropy rate is lower than the set judgment threshold, it is judged as a multidimensional barcode wear abnormality.

6. The deep learning-based multidimensional barcode detection method according to claim 1, characterized in that, The correction process of the correction algorithm is as follows: Obtain each abnormal region of the multidimensional barcode. Using the center of each abnormal region as the first center, draw a first circular region outwards, completely covering the abnormal regions. Identify the first radius of the circular region, and draw a second circular region outwards from the first center, with the second radius being twice the first radius. Within the first circular region, draw n concentric circles centered on the first center, and encode each concentric circle. Determine the blocks within the concentric circles that need adjustment. Obtain normal image units overlapping the black and white barcode areas of the second circular region with the outermost concentric circle region. Identify the part with the highest edge influence feature value in the clearly contacted image portion, and perform extension and stitching correction processing with the outermost concentric circle. After completing the extension and stitching correction processing of the outermost concentric circle, perform extension and stitching correction processing on the inner concentric circles sequentially until all concentric circles are corrected.

7. A detection device, applied to the deep learning-based multidimensional barcode detection method as described in any one of claims 1-6, characterized in that, include: Multidimensional barcode extraction and recognition unit: used to acquire the base image to be recognized, and to perform edge detection on the base image based on edge algorithms to identify the corner points of the multidimensional barcode; Invalid edge removal unit: When the multidimensional barcode extraction and recognition unit cannot identify the corner points of the multidimensional barcode, the invalid edges of the base image are identified and removed by the pre-trained multidimensional barcode edge processing model. Multidimensional barcode anomaly detection unit: When the barcode cannot be recognized, it can determine the anomaly type of the multidimensional barcode based on the multidimensional barcode anomaly detection model, and determine whether it is an exposure anomaly or a multidimensional barcode wear anomaly. Multidimensional barcode correction unit: It can correct multidimensional barcode images with exposure errors and barcode wear errors through correction algorithms.

8. The detection device according to claim 7, characterized in that, The invalid edge removal unit includes: Edge removal module: Through a large amount of training on the basic feature recognition of multidimensional barcodes using a multidimensional barcode edge processing model, parts of the basic image that do not conform to the basic features of multidimensional barcodes are removed. The first processing module: preserves the possible image parts whose basic feature matching degree exceeds the set threshold, and further magnifies the possible images; The second processing module performs grayscale conversion and bilateral filtering on the possible images, and inputs the filtered and grayscale images into a pre-trained edge processing module that can identify image edges to remove the edges, while retaining the multidimensional barcode processing part.

Images