Deep learning-based t7code positioning, identifying and reading method

Through deep learning and image morphology processing technology, the problems of low reading efficiency and poor compatibility of T7Code are solved, efficient and accurate positioning and reading are achieved, and user experience and device adaptability are improved.

CN119005224APending Publication Date: 2024-11-22江苏道达智能科技股份有限公司
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202410893968.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-04
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

The existing T7Code read technology is inefficient, poor device compatibility, poor user experience, and requires manual adjustment of the scanning box and complex operation.

Method used

Using the T7Code positioning recognition and reading method based on deep learning, the barcode position is determined through preprocessing and image morphology processing, combined with the two-stage object detection algorithm and image calibration technology, improve the reading accuracy and efficiency, and through data enhancement and model optimization Improve generalization capabilities and real-time performance.

Benefits of technology

It realizes efficient and accurate T7Code positioning and reading, reduces the time and error rate of manual operation, enhances the compatibility and user experience of the device, and is suitable for a variety of scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119005224A_ABST
    Figure CN119005224A_ABST
Patent Text Reader

Abstract

The invention relates to the field of image recognition and machine learning, and discloses a t7code positioning recognition reading method based on deep learning. Comprising the following steps: step 1, preprocessing an input image; 2, data preparation: collecting photographing data of the equipment; step 3, model training; step 4, image calibration and enhancement, wherein the image detected by the deep learning model has a redundant part; 5, evaluating and optimizing the model to ensure that the performance of the model reaches the expectation; step 6, real-time performance and deployment are considered; and 7, continuously optimizing and updating, putting the model into use, and continuously monitoring and optimizing to deal with the problems of environment change and data drift. A deep learning technology is utilized to construct a positioning identification model aiming at T7Code, an image processing technology is utilized to calibrate and enhance the T7Code, decoding efficiency is improved, n bar codes are randomly extracted to calculate inclination angles and then average values are taken, and calibration operation efficiency is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the fields of image recognition and machine learning, and in particular to a T7code positioning, recognition and reading method based on deep learning. Background Art

[0002] T7Code reading technology is a technology used to identify and read T7Code barcodes. T7Code is a special type of barcode with a unique coding structure and characteristics. T7Code reading technology usually involves the use of computer vision and image processing technology, combined with deep learning or traditional image processing algorithms to accurately identify T7Code barcodes from images and convert them into readable data. This technology can be applied to various scenarios, such as industrial automation, logistics tracking, product identification and other fields.

[0003] The existing t7code reading technology is a handheld barcode reading device that manually aligns the label and scans it one by one. The main problems are as follows: efficiency problem: traditional t7code scanning devices usually require users to manually adjust the alignment position of the scanning frame and the QR code to scan one by one, which is time-consuming and labor-intensive. Equipment compatibility problem: special equipment is required and cannot be expanded or adapted to other scenarios. User experience problem: traditional QR code scanning requires users to manually open the scanning application and align the QR code for scanning, which may be inconvenient for users who are not familiar with the operation. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides a T7code positioning, identification and reading method based on deep learning, which solves the problems of low manual scanning efficiency, time-consuming and labor-intensive, easy misoperation of manual scanning, poor compatibility of barcode equipment and the need for special procurement in the existing T7code reading technology.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a t7code positioning, recognition and reading method based on deep learning, comprising the following steps:

[0006] Step 1: First, pre-process the input image, use the deep learning model to extract and identify the features of the candidate area, extract the candidate area of ​​the T7Code, then use the image morphological processing method to accurately determine the position of the T7Code, and rotate and align the tilted barcode to improve the barcode reading accuracy;

[0007] Step 2: Data preparation: collect the device’s photo data and build a labeled data set;

[0008] Step 3: Model training. In terms of model selection, a two-stage target detection algorithm is adopted;

[0009] Step 4: Image calibration and enhancement. The images detected by the deep learning model may contain redundant parts. In addition, the actual operating environment may have poor lighting and barcode image rotation.

[0010] Step 5: Model evaluation and tuning: After the model training is completed, it needs to be evaluated and tuned to ensure that its performance meets expectations.

[0011] Step 6: Real-time performance and deployment considerations. Before deploying the model to actual applications, its performance and efficiency in a real-time environment need to be considered.

[0012] Step seven: Continuous optimization and updating. Once the model is put into use, it needs to be continuously monitored and optimized to cope with environmental changes and data drift.

[0013] Preferably, in step 2, the shooting scene and image size of the factory machine are relatively fixed, only the light will change with different time periods and seasons, randomly extract pictures of different time periods, and randomly enhance or weaken the light of the image, increase data diversity and reduce the model's dependence on light. In addition, the manufacturer's T7code paper will have two colors, light green or light red, and the proportion of the two colors is not balanced. Extract part of the image data for random color transformation, reduce the model's dependence on color and improve the generalization ability of the model. The specific method of color distortion is to convert the image from the BGR channel to the HSV channel, and randomly transform the HUE channel of the image hue = (hue + random_factor_a)% random_factor_b, where random_factor_a is a random value from 30 to 200, and random_factor_b is a random value from 150 to 200. Then divide the data into training set and test set, and perform normalization.

[0014] Preferably, in step three, three effective feature layers are obtained in the Backbone network through convolution and pooling network feature extraction, which are 52×52×256, 26×26×512, and 13×13×1024 respectively. The previous features are fused in the Neck part, and finally the prediction result is output in the Head part. The output vector includes not only the category of the target, the coordinates of the bounding box and the predicted confidence, but finally a locally unique prediction box is obtained through non-maximum suppression.

[0015] Preferably, in step four, the image morphology processing method is used to eliminate image burrs by binarizing the image and performing multiple dilation and corrosion operations, perform contour detection on the square area, and calculate the angle of the horizontal position. The image is cropped and rotated based on the contour and angle, and the calibrated image is then histogram-equalized to improve the light and dark contrast of the image and enhance the clarity of the barcode.

[0016] Preferably, in step 5, cross-validation is used or a portion of the data is reserved as a validation set to evaluate the performance of the model on different data sets. The effect of the model is measured by indicators such as accuracy, recall, and F1 score, and parameter adjustment and model optimization are performed based on the evaluation results. In addition, regularization and dropout can be used to improve the generalization ability of the model for possible overfitting or underfitting problems.

[0017] Preferably, in step six, in view of the usually large inference time and resource consumption of deep learning models, some optimization strategies such as model pruning and quantization can be adopted to reduce the number of model parameters and computational complexity, so as to improve its running speed and efficiency in embedded devices or real-time systems. In addition, according to the needs of specific scenarios, the model can also be deployed to the cloud or edge for flexible selection.

[0018] Preferably, in step seven, the performance data of the model in actual applications is regularly collected and analyzed, and the model is updated and optimized in combination with user feedback and needs, which includes adding new training data, adjusting model parameters, and improving algorithmic measures to continuously improve the accuracy and stability of the model and maintain its effectiveness and reliability in long-term use.

[0019] The present invention provides a t7code positioning, recognition and reading method based on deep learning.

[0020] Beneficial effects:

[0021] 1. Using deep learning technology, a positioning and recognition model for T7Code is constructed. The model can accurately identify the position of T7Code in the image and provide accurate coordinate information for subsequent reading operations; 2. Based on the accurate positioning of T7Code, image processing technology is used to calibrate and enhance T7Code, which improves the decoding efficiency; 3. Since a T7Code contains more than 30 barcode information in parallel, n barcodes are randomly selected to calculate the tilt angle and then take the average value to improve the efficiency of the calibration operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0023] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] Embodiment 1:

[0025] like Figure 1 As shown, an embodiment of the present invention provides a T7code positioning, recognition and reading method based on deep learning, comprising the following steps:

[0026] Step 1: First, pre-process the input image, use the deep learning model to extract and identify the features of the candidate area, extract the candidate area of ​​the T7Code, then use the image morphological processing method to accurately determine the position of the T7Code, and rotate and align the tilted barcode to improve the barcode reading accuracy;

[0027] Step 2: Data preparation: collect the device’s photo data and build a labeled data set;

[0028] Step three, model training, in model selection, a two-stage target detection algorithm is adopted.

[0029] In step 2, the shooting scene and image size of the factory machine are relatively fixed. Only the light will change with different time periods and seasons. Randomly extract pictures of different time periods, and randomly enhance or weaken the light of the images to increase data diversity and reduce the model's dependence on light. In addition, the manufacturer's T7code paper will have two colors, light green or light red, and the proportion of the two colors is not balanced. Extract part of the image data for random color transformation to reduce the model's dependence on color and improve the model's generalization ability. The specific method of color distortion is to convert the image from the BGR channel to the HSV channel, and randomly transform the image's HUE channel hue = (hue + random_factor_a) % random_factor_b, where random_factor_a is a random value from 30 to 200, and random_factor_b is a random value from 150 to 200. Then divide the data into training set and test set, and perform normalization.

[0030] In step three, three effective feature layers are obtained in the Backbone network through convolution and pooling network feature extraction, which are 52×52×256, 26×26×512, and 13×13×1024 respectively. The previous features are fused in the Neck part, and finally the prediction result is output in the Head part. The output vector includes not only the target category, the coordinates of the bounding box and the predicted confidence, but also the local unique prediction box is obtained through non-maximum suppression.

[0031] Embodiment 2:

[0032] like Figure 1 As shown, an embodiment of the present invention provides a T7code positioning, recognition and reading method based on deep learning, comprising the following steps:

[0033] Step 1: First, pre-process the input image, use the deep learning model to extract and identify the features of the candidate area, extract the candidate area of ​​the T7Code, then use the image morphological processing method to accurately determine the position of the T7Code, and rotate and align the tilted barcode to improve the barcode reading accuracy;

[0034] Step 2: Data preparation: collect the device’s photo data and build a labeled data set;

[0035] Step 3: Model training. In terms of model selection, a two-stage target detection algorithm is adopted;

[0036] Step 4: Image calibration and enhancement. The images detected by the deep learning model may contain redundant parts. In addition, the actual operating environment may have poor lighting and barcode image rotation.

[0037] Step 5: Model evaluation and tuning: After the model training is completed, it needs to be evaluated and tuned to ensure that its performance meets expectations.

[0038] Step 6: Real-time performance and deployment considerations. Before deploying the model to actual applications, its performance and efficiency in a real-time environment need to be considered.

[0039] Step seven: Continuous optimization and updating. Once the model is put into use, it needs to be continuously monitored and optimized to cope with environmental changes and data drift.

[0040] In step 2, the shooting scene and image size of the factory machine are relatively fixed. Only the light will change with different time periods and seasons. Randomly extract pictures of different time periods, and randomly enhance or weaken the light of the images to increase data diversity and reduce the model's dependence on light. In addition, the manufacturer's T7code paper will have two colors, light green or light red, and the proportion of the two colors is not balanced. Extract part of the image data for random color transformation to reduce the model's dependence on color and improve the model's generalization ability. The specific method of color distortion is to convert the image from the BGR channel to the HSV channel, and randomly transform the image's HUE channel hue = (hue + random_factor_a) % random_factor_b, where random_factor_a is a random value from 30 to 200, and random_factor_b is a random value from 150 to 200. Then divide the data into training set and test set, and perform normalization.

[0041] In step three, three effective feature layers are obtained in the Backbone network through convolution and pooling network feature extraction, which are 52×52×256, 26×26×512, and 13×13×1024 respectively. The previous features are fused in the Neck part, and finally the prediction result is output in the Head part. The output vector includes not only the target category, the coordinates of the bounding box and the predicted confidence, but also the local unique prediction box is obtained through non-maximum suppression.

[0042] In step 4, the image morphology processing method is used to eliminate image burrs by binarizing the image and performing multiple dilation and corrosion operations. The contour of the square area is detected, and the angle of the horizontal position is calculated. The image is cropped and rotated based on the contour and angle. The calibrated image is then histogram-equalized to improve the contrast between light and dark in the image and enhance the clarity of the barcode.

[0043] In step 5, cross-validation is used or a portion of the data is reserved as a validation set to evaluate the performance of the model on different data sets. The effectiveness of the model is measured by indicators such as accuracy, recall, and F1 score, and parameter adjustments and model optimization are performed based on the evaluation results. In addition, regularization and dropout can be used to improve the generalization ability of the model for possible overfitting or underfitting problems.

[0044] In step six, in view of the usually large inference time and resource consumption of deep learning models, some optimization strategies such as model pruning and quantization can be used to reduce the number of model parameters and computational complexity to improve its running speed and efficiency in embedded devices or real-time systems. In addition, according to the needs of specific scenarios, the model can also be deployed to the cloud or edge for flexible selection.

[0045] In step seven, the performance data of the model in actual applications is regularly collected and analyzed, and the model is updated and optimized based on user feedback and needs. This includes adding new training data, adjusting model parameters, and improving algorithmic measures to continuously improve the accuracy and stability of the model and maintain its effectiveness and reliability in long-term use.

[0046] The development process record table is shown in Table 1.

[0047] Table 1: T7Code barcode recognition system development process record table

[0048]

[0049]

[0050] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A t7code positioning, recognition and reading method based on deep learning. Features: The following steps are involved: Step 1: First, pre-process the input image, use the deep learning model to extract and identify the features of the candidate area, extract the candidate area of ​​the T7Code, then use the image morphological processing method to accurately determine the position of the T7Code, and rotate and align the tilted barcode to improve the barcode reading accuracy; Step 2: Data preparation: collect the device’s photo data and build a labeled data set; Step 3: Model training. In terms of model selection, a two-stage target detection algorithm is adopted; Step 4: Image calibration and enhancement. The images detected by the deep learning model may contain redundant parts. In addition, the actual operating environment may have poor lighting and barcode image rotation. Step 5: Model evaluation and tuning: After the model training is completed, it needs to be evaluated and tuned to ensure that its performance meets expectations. Step 6: Real-time performance and deployment considerations. Before deploying the model to actual applications, its performance and efficiency in a real-time environment need to be considered. Step seven: Continuous optimization and updating. Once the model is put into use, it needs to be continuously monitored and optimized to cope with environmental changes and data drift.

2. According to the t7code positioning, recognition and reading method based on deep learning according to claim 1, Features: In step 2, the shooting scene and image size of the factory machine are relatively fixed, only the light will change with different time periods and seasons. Randomly extract pictures of different time periods, and randomly enhance or weaken the light of the images to increase data diversity and reduce the model's dependence on light. In addition, the manufacturer's T7code paper will have two colors, light green or light red, and the proportion of the two colors is not balanced. Extract part of the image data for random color transformation to reduce the model's dependence on color and improve the model's generalization ability; the specific method of color distortion is to convert the image from the BGR channel to the HSV channel, and randomly transform the image's HUE channel hue = (hue + random_factor_a) % random_factor_b, where random_factor_a is a random value from 30 to 200, and random_factor_b is a random value from 150 to 200. Then divide the data into training set and test set, and perform normalization.

3. According to the t7code positioning, recognition and reading method based on deep learning according to claim 1, Features: In step three, three effective feature layers are obtained in the Backbone network through convolution and pooling network feature extraction, which are 52×52×256, 26×26×512, and 13×13×1024 respectively. The previous features are fused in the Neck part, and finally the prediction results are output in the Head part. The output vector includes not only the target category, the coordinates of the bounding box and the predicted confidence, but also the local unique prediction box is obtained through non-maximum suppression.

4. According to the t7code positioning, recognition and reading method based on deep learning according to claim 1, Features: In step 4, the image morphology processing method is used to eliminate image burrs by binarizing the image and performing multiple dilation and corrosion operations. The contour of the square area is detected, and the angle of the horizontal position is calculated. The image is cropped and rotated based on the contour and angle. The calibrated image is then histogram-equalized to improve the contrast between light and dark in the image and enhance the clarity of the barcode.

5. According to the t7code positioning, recognition and reading method based on deep learning according to claim 1, Features: In step five, cross-validation is used or a portion of the data is reserved as a validation set to evaluate the performance of the model on different data sets. The effectiveness of the model is measured through indicators such as accuracy, recall, and F1 score, and parameter adjustments and model optimization are performed based on the evaluation results. In addition, for possible overfitting or underfitting problems, regularization and Dropout can be used to improve the generalization ability of the model.

6. According to the t7code positioning, recognition and reading method based on deep learning as described in claim 1, Features: In step six, in view of the usually large inference time and resource consumption of deep learning models, some optimization strategies such as model pruning and quantization can be used to reduce the number of model parameters and computational complexity to improve its running speed and efficiency in embedded devices or real-time systems. In addition, according to the needs of specific scenarios, the model can also be deployed to the cloud or edge for flexible selection.

7. According to the t7code positioning, recognition and reading method based on deep learning according to claim 1, Features: In step seven, the performance data of the model in actual applications is regularly collected and analyzed, and the model is updated and optimized based on user feedback and needs. This includes adding new training data, adjusting model parameters, and improving algorithmic measures to continuously improve the accuracy and stability of the model and maintain its effectiveness and reliability in long-term use.

Citation Information

Cited By

  • Wafer T7 code identification method based on double-channel cross verification

    CN122761387A