Label automatic detection method and system

Abstract

The invention discloses an automatic label detection method and system, relates to the technical field of automatic label detection, and solves the technical problems that adaptive sorting and identification of multiple labels are difficult to carry out according to resource conditions, and content restoration of existing incomplete labels is difficult to carry out. According to the invention, the resource calling factor of the current label scanning equipment is determined based on the computing power and bandwidth of the label scanning equipment, and the identification factor of each label is determined based on the resource calling factor and the name and type of the label; performing content identification on each label based on the identification factor to obtain the content of each label, and dividing the label into a complete label and an incomplete label; complete labels similar to the incomplete labels in the scanning area are obtained based on the content of the labels, the complete labels are marked as similar labels, similar factors are obtained, and content filling is conducted on the incomplete labels based on the similar labels corresponding to the similar factors; the problems of low identification efficiency, resource waste and information loss in a multi-label scene are solved.

Description

Technical Field

[0001] This invention pertains to automatic label detection technology, specifically an automatic label detection method and system. Background Technology

[0002] In traditional manufacturing, product label inspection is typically done manually. For example, in industries such as food, pharmaceuticals, and cosmetics, each product needs a label containing product information, production date, and shelf life. Manual inspection is slow and susceptible to fatigue and lack of concentration, leading to low efficiency. Automated inspection methods and systems can achieve rapid and continuous label inspection, significantly increasing inspection speed on the production line and thus improving overall production efficiency.

[0003] Currently, most automatic tag detection methods and systems struggle to adaptively sort and identify multiple tags in real-time based on resource availability in multi-tag scenarios. They simply perform tag detection in a fixed order or simultaneously. When the performance of the detection equipment degrades, they cannot adaptively adjust the detection strategy, reducing the flexibility and efficiency of the detection system. Furthermore, most automatic tag detection methods and systems are unable to restore the content of incomplete tags, requiring manual content search and restoration, which increases labor costs.

[0004] Therefore, this invention discloses an automatic label detection method and system to solve the above-mentioned technical problems. Summary of the Invention

[0005] This invention aims to at least solve one of the technical problems existing in the prior art. To this end, this invention proposes an automatic tag detection method and system to solve the technical problems of difficulty in adaptively sorting and identifying multiple tags in real time according to resource conditions in multi-tag scenarios, and difficulty in restoring the content of incomplete tags. This invention determines the resource call factor of the current tag scanning device based on the computing power and bandwidth of the tag scanning device, and determines the identification factor of each tag based on the resource call factor, the name and type of each tag; performs content recognition on each tag based on the identification factor to obtain the content of each tag, and classifies complete tags and incomplete tags; obtains complete tags similar to each incomplete tag in the scanning area based on the tag content, and marks the complete tags as similar tags; obtains the similarity factor of each similar tag, and fills the content of incomplete tags based on the similar tags corresponding to the similarity factor, thus solving the above problems.

[0006] To achieve the above objectives, a first aspect of the present invention provides an automatic label detection method, comprising: Obtain the name and type of each tag within the scanning area, as well as the computing power and bandwidth of the tag scanning device; The resource allocation factor for the current tag scanning device is determined based on computing power and bandwidth. The identification factor of each tag is determined based on the resource access factor, the name and type of each tag; Content recognition is performed on each tag based on recognition factors to obtain the content of each tag; the content includes text, images, text and image layout, and the position of text and images in the tag; The content-recognized tags are categorized into complete tags and incomplete tags; When incomplete labels exist within the scanned area, complete labels similar to each incomplete label within the scanned area are obtained based on the content of the labels, and the complete labels are marked as similar labels. Obtain the similarity factor for each similar tag, fill in the incomplete tag with content based on the similar tags corresponding to the similar factors, and report to the administrator.

[0007] Preferably, obtaining the name and type of each tag within the scanning area, as well as the computing power and bandwidth of the tag scanning device, includes: The video recognition technology is used to obtain the names of each label within the manually set scanning area, as well as the type of product corresponding to each label. The computing power available to the label scanning device is obtained in real time through a computing power monitoring tool, and the bandwidth available to the label scanning device is obtained in real time through a bandwidth monitoring tool. The product types include emergency medical products, critical and valuable products, and ordinary products.

[0008] Preferably, determining the resource allocation factor of the current tag scanning device based on computing power and bandwidth includes: Extract the computing power FL and bandwidth BR of the current label scanning device. When the computing power FL is not less than the standard computing power BFL and the bandwidth BR is not less than the standard bandwidth BBR, the resource call factor DZ is set to the maximum value of the standard call range. The standard computing power BFL is determined manually based on the total computing power of the current label scanning device, and the standard bandwidth BBR is determined manually based on the maximum bandwidth of the current label scanning device. The standard call range is determined manually based on the performance of the current label scanning device. When the computing power FL is less than the standard computing power BFL or the bandwidth BR is less than the standard bandwidth BBR, the resource allocation factor DZ of the current tag scanning device is determined based on the computing power FL and the bandwidth BR, and the resource allocation factor DZ satisfies formula (1): DZ=TZ×α / exp(β1×(BFL-FL) / BFL+β2×(BBR-BR) / BBR)(1); Wherein, TZ is the median value of the standard call range; α is the manually set amplitude adjustment coefficient of the 1 / exp() function, and the value range of α is [0,2]; β1 and β2 are both manually set proportional adjustment coefficients greater than 0, and β1+β2=1, β1≥β2; When the resource call factor DZ is greater than the maximum value of the standard call range, the resource call factor DZ is updated with the maximum value; when the resource call factor DZ is less than the maximum value of the standard call range, the resource call factor DZ is updated with the minimum value.

[0009] Preferably, the step of determining the identification factor of each tag based on the resource access factor, the name and type of each tag includes: Extract the resource call factor. When the resource call factor is not less than the call threshold, identify each tag simultaneously or identify each tag in the order of arrangement. The call threshold is the percentile of the standard call range. The percentile is set manually and can be a number at the 60th position. When the resource call factor is less than the call threshold, the names of each tag are extracted sequentially. When the name of the tag contains the target keyword, the keyword factor GZ of the tag is set to the standard key factor. When the name does not contain the target keyword, the keyword factor GZ of the tag is set to 0. The target keyword and the standard key factor are manually set. The types of each tag are extracted sequentially, and the type factor LZ corresponding to the type is extracted from the type factor library based on the type. The identification factor SZ of each tag is determined based on the keyword factor GZ and the type factor LZ. The identification factor SZ satisfies formula (2): SZ=γ×ln(δ1×GZ+δ2×LZ)(2); Where γ is the amplitude adjustment coefficient of the ln() function set manually, and the value range of γ is [0,2]; δ1 and δ2 are the proportional adjustment coefficients set manually, and δ1≥δ2.

[0010] Preferably, the step of obtaining the content of each tag by performing content recognition based on the recognition factor includes: The identification factors of each label are extracted and numbered from largest to smallest. The labels are then scanned sequentially using a label scanning device and a camera to identify their content, resulting in the text, images, text and image layouts, and the positions of the text and images within the labels. The text layout includes the positional relationships between text and text, and between text and images. The image layout includes the positional relationships between images, and between images and text.

[0011] Preferably, classifying the content-identified tags into complete tags and incomplete tags includes: After the label undergoes content recognition, the text and image of the label are extracted sequentially. If the text is incomplete in terms of name, specifications, ingredients, producer information, or precautions, or if the image is incomplete, the label is marked as an incomplete label. When the text is incomplete (e.g., name, specifications, ingredients, manufacturer information, precautions), or the image is incomplete, the label will be marked as a complete label.

[0012] Preferably, the tag-based content obtains complete tags similar to each incomplete tag within the scanned area, including: Extract the content of each tag from several scans in the historical reference data, as well as the complete tags that are similar to the incomplete tags in each tag; wherein, the historical reference data includes the content of each tag from several scans, as well as the complete tags that are similar to each incomplete tag obtained by manual analysis based on the content of the tags; The content of each label, as well as the complete labels that are similar to the incomplete labels, are integrated into several sets of training data and test data. The training data is used to train the artificial intelligence model, and the test data is used to test the trained artificial intelligence model. The artificial intelligence model is adjusted according to the test results. Finally, a similarity evaluation model is obtained with the content of each label as input and the complete labels that are similar to the incomplete labels as output. The artificial intelligence model includes a BP neural network model and an RBF neural network model. The content of each label in this scan is input into the similarity evaluation model to obtain the complete labels that are similar to each incomplete label in this scan.

[0013] Preferably, obtaining the similarity factor for each similar tag includes: Extract the linear distance Li between the current incomplete label and the similar labels of the current incomplete label. When the linear distance Li is greater than the linear threshold, mark the similarity factor of the corresponding complete label as 0; when the linear distance Li is not greater than the linear threshold, determine the similarity factor YZi of the similar labels of the current incomplete label based on the linear distance Li; the similarity factor YZi satisfies formula (3): YZi=1 / exp(Li / (∑(Li / n)+Mode(Li)+P(Li)))(3); Where i is the number of the similar tag of the current incomplete tag, and the value of i is [1, n], and n is the maximum value of the similar tag number; ∑ is the summation symbol, and the summation range is [1, n]; Mode(Li) symbol is to take the mode of Li, and P(Li) is to take the percentile of Li, and the percentile is manually selected.

[0014] Preferably, the step of filling in incomplete tags with content based on similar tags corresponding to similar factors and reporting it to the administrator includes: Extract similarity factors from similar tags, mark the similar tag with the largest similarity factor as the substitute tag, fill in the missing content of incomplete tags based on the content of the substitute tag, and report the filled record to the administrator.

[0015] A second aspect of the present invention provides an automatic label detection system, comprising: a label recognition module, and a data acquisition module and a label information restoration module connected to the label recognition module; The data acquisition module is used to acquire the name and type of each label within the scanning area, as well as the computing power and bandwidth of the label scanning device. The tag recognition module is used to: determine the resource access factor of the current tag scanning device based on computing power and bandwidth; determine the recognition factor of each tag based on the resource access factor, the name and type of each tag; and perform content recognition on each tag based on the recognition factor to obtain the content of each tag; wherein, the content includes text, images, text and image layouts, and the positions of text and images in the tag; The tag information restoration module is used to classify the tags after content recognition into complete tags and incomplete tags; when there are incomplete tags in the scanning area, it obtains complete tags similar to each incomplete tag in the scanning area based on the content of the tags, and marks the complete tags as similar tags; it obtains the similarity factor of each similar tag, fills the content of the incomplete tags based on the similar tags corresponding to the similarity factor, and reports to the administrator.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention determines the resource allocation factor of the current tag scanning device based on its computing power and bandwidth. Based on the resource allocation factor, the name and type of each tag, it determines the identification factor of each tag. Based on the identification factor, it performs content recognition on each tag to obtain its content, classifying tags into complete and incomplete tags. Based on the tag content, it obtains complete tags similar to each incomplete tag within the scanning area and marks these complete tags as similar tags. It acquires the similarity factor of each similar tag and fills in the content of incomplete tags based on the similar tags corresponding to the similarity factor. This solves the technical problems of difficulty in adaptively sorting and recognizing multiple tags in real time according to resource conditions and difficulty in restoring the content of existing incomplete tags in multi-tag scenarios. Through a dual mechanism of intelligent resource scheduling and adaptive content repair, this invention can solve the problems of low recognition efficiency, resource waste, and information loss in multi-tag scenarios, and has significant practical value and industry promotion potential.

[0017] 2. This invention constructs a dynamic and intelligent tag recognition factor calculation system by introducing multi-dimensional correlation analysis between resource call factors and tag attributes. Its core advantages are reflected in the following aspects: First, the quantitative evaluation of resource call factors enables real-time perception of system load status. When the resource call factor exceeds the preset call threshold, the system can automatically start a multi-tag parallel recognition mechanism. When the resource call factor is insufficient, the keyword matching mechanism achieves accurate tag screening. This dual-track design retains the controllability of manually setting target keywords and establishes a unified evaluation benchmark through the quantitative definition of standard key factors. The method of directly assigning a zero value when the tag name does not contain the target keyword effectively reduces the false recognition rate and simplifies the calculation process. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the operation steps of the present invention; Figure 2 This is a schematic diagram illustrating the operational steps for obtaining the resource call factor according to the present invention; Figure 3 This is a schematic diagram of the system modules of the present invention. Detailed Implementation

[0020] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please see Figure 1 The first aspect of this invention provides an automatic label detection method, comprising: Obtain the name and type of each tag within the scanning area, as well as the computing power and bandwidth of the tag scanning device; The resource allocation factor for the current tag scanning device is determined based on computing power and bandwidth. The identification factor of each tag is determined based on the resource access factor, the name and type of each tag; Content recognition is performed on each tag based on recognition factors to obtain the content of each tag; the content includes text, images, text and image layout, and the position of text and images in the tag; The content-recognized tags are categorized into complete tags and incomplete tags; When incomplete labels exist within the scanned area, complete labels similar to each incomplete label within the scanned area are obtained based on the content of the labels, and the complete labels are marked as similar labels. Obtain the similarity factor for each similar tag, fill in the incomplete tag with content based on the similar tags corresponding to the similar factors, and report to the administrator.

[0022] It should be noted that the scanning area is the area on the label that the label scanning device scans.

[0023] It should be noted that the text on the label in this invention includes the name, specifications, ingredients, manufacturer information, and precautions.

[0024] This application obtains the names and types of each tag within the scanning area, as well as the computing power and bandwidth of the tag scanning device, including: The video recognition technology is used to obtain the names of each label within the manually set scanning area, as well as the type of product corresponding to each label. The computing power available to the label scanning device is obtained in real time through a computing power monitoring tool, and the bandwidth available to the label scanning device is obtained in real time through a bandwidth monitoring tool. The product types include emergency medical products, critical and valuable products, and ordinary products.

[0025] It should be noted that computing power monitoring tools can be nvidia-smi or Geekbench; bandwidth monitoring tools can be iperf or CrystalDiskMark.

[0026] Please see Figure 2 In this application, the resource allocation factor for the current tag scanning device is determined based on computing power and bandwidth, including: Extract the computing power FL and bandwidth BR of the current label scanning device. When the computing power FL is not less than the standard computing power BFL and the bandwidth BR is not less than the standard bandwidth BBR, the resource call factor DZ is set to the maximum value of the standard call range. The standard computing power BFL is determined manually based on the total computing power of the current label scanning device, and the standard bandwidth BBR is determined manually based on the maximum bandwidth of the current label scanning device. The standard call range is determined manually based on the performance of the current label scanning device. When the computing power FL is less than the standard computing power BFL or the bandwidth BR is less than the standard bandwidth BBR, the resource allocation factor DZ of the current tag scanning device is determined based on the computing power FL and the bandwidth BR, and the resource allocation factor DZ satisfies formula (1): DZ=TZ×α / exp(β1×(BFL-FL) / BFL+β2×(BBR-BR) / BBR)(1); Wherein, TZ is the median value of the standard call range; α is the manually set amplitude adjustment coefficient of the 1 / exp() function, and the value range of α is [0,2]; β1 and β2 are both manually set proportional adjustment coefficients greater than 0, and β1+β2=1, β1≥β2; When the resource call factor DZ is greater than the maximum value of the standard call range, the resource call factor DZ is updated with the maximum value; when the resource call factor DZ is less than the maximum value of the standard call range, the resource call factor DZ is updated with the minimum value.

[0027] It is worth noting that this invention determines the resource access factor of the current tag scanning device by computing power and bandwidth, which enables refined management and dynamic optimization of device resources. By comparing the device's computing power FL and bandwidth BR with preset standard values ​​BFL and BBR and combining them with the resource access factor DZ calculated by the formula, the system operating efficiency can be effectively improved while ensuring the stability and reliability of the device.

[0028] First, by setting standard computing power and standard bandwidth as benchmarks, this invention can accurately assess the actual performance status of the device. When the computing power and bandwidth of the device reach or exceed the standard values, the maximum value of the standard call range is directly used as the resource call factor, which can make full use of the hardware potential of the device and avoid performance bottlenecks caused by resource waste. When the device performance does not meet the standard, the resource call factor can be dynamically adjusted according to the degree of deviation of computing power and bandwidth by calculating the exponential function in formula (1). Among them, α, as an amplitude adjustment coefficient, can control the scaling range of the overall call factor, while β1 and β2, as proportional adjustment coefficients, can balance the impact of computing power and bandwidth on resource call. This design not only ensures the flexibility of resource allocation but also enables precise control based on the actual load of the equipment.

[0029] This invention effectively avoids situations where resource call factors exceed reasonable ranges by introducing upper and lower limit constraints. When DZ exceeds the maximum value, the maximum value is used to replace it to prevent device overload caused by excessive calls. When DZ is below the minimum value, the minimum value is used to replace it to ensure that the device always maintains basic operating capabilities. This improves performance while avoiding the risk of resource waste and system crashes. Furthermore, this invention enables intelligent decision-making for resource allocation by combining manually set parameters with real-time device performance data, reducing reliance on human experience while improving the system's adaptability. For example, in scenarios with significant fluctuations in device computing power or bandwidth, dynamically adjusting resource allocation factors can effectively address sudden load changes, ensuring the continuity and accuracy of label scanning tasks. In addition, this invention exhibits excellent scalability; when device configurations are upgraded or business requirements change, only standard values ​​or parameter coefficients need to be adjusted to quickly adapt to the new operating environment without reconstructing the entire resource management logic. This modular design significantly reduces system maintenance costs and enhances the flexibility of the overall architecture. From a practical application perspective, this invention can significantly improve the operating efficiency and stability of label scanning devices. By rationally calculating resources and network bandwidth, it reduces latency or errors caused by insufficient resources while lowering energy consumption, thereby optimizing operating costs while ensuring service quality. In summary, the resource allocation factor determination method based on computing power and bandwidth, through scientific parameter settings and dynamic adjustment mechanisms, achieves efficient utilization of device resources, improving system performance while ensuring operational reliability and economy, providing strong support for the intelligent management of label scanning devices.

[0030] It should be noted that the standard call range is used to limit the range of the resource call factor DZ.

[0031] It should be noted that α is used to adjust the degree of influence of computing power FL and bandwidth BR on resource allocation factor DZ; when other conditions remain unchanged, the larger α is, the greater the influence on resource allocation factor DZ, and the smaller α is, the smaller the influence on resource allocation factor DZ.

[0032] It should be noted that β1 and β2 were designed manually based on the total computing power of the current label scanning equipment, and β1 is inversely proportional to the total computing power, while β2 is directly proportional to the total computing power. β1≥β2 because: β1 is multiplied by data related to computing power FL, while β2 is multiplied by data related to bandwidth BR; in the resource call factor DZ analysis of the current tag scanning device, the data related to computing power FL can directly reflect the data analysis capability of the current tag scanning device, while the data related to bandwidth BR reflects the data transmission capability. Therefore, the proportional adjustment coefficient set for the data related to computing power FL in this invention is not less than that for the data related to bandwidth BR.

[0033] In this application, the identification factors of each tag are determined based on the resource access factor, the name and type of each tag, including: Extract the resource call factor. When the resource call factor is not less than the call threshold, identify each tag simultaneously or identify each tag in the order of arrangement. The call threshold is the percentile of the standard call range. The percentile is set manually and can be a number at the 60th position. When the resource call factor is less than the call threshold, the names of each tag are extracted in sequence. When the tag name contains the target keyword, the tag's keyword factor GZ is set to the standard key factor; when the name does not contain the target keyword, the tag's keyword factor GZ is set to 0. The target keyword and the standard key factor are manually set. The types of each tag are extracted sequentially, and the type factor LZ corresponding to the type is extracted from the type factor library based on the type. The identification factor SZ of each tag is determined based on the keyword factor GZ and the type factor LZ. The identification factor SZ satisfies formula (2): SZ=γ×ln(δ1×GZ+δ2×LZ)(2); Where γ is the amplitude adjustment coefficient of the ln() function set manually, and the value range of γ is [0,2]; δ1 and δ2 are the proportional adjustment coefficients set manually, and δ1≥δ2.

[0034] It is worth noting that this invention constructs a dynamic and intelligent tag recognition factor calculation system by introducing multi-dimensional correlation analysis between resource call factors and tag attributes. Its core advantages are reflected in the following aspects: First, the quantitative evaluation of resource call factors enables real-time perception of system load status. When the resource call factor exceeds the preset call threshold, the system can automatically start a multi-tag parallel recognition mechanism. When the resource call factor is insufficient, the keyword matching mechanism achieves accurate tag screening. This dual-track design retains the controllability of manually setting target keywords and establishes a unified evaluation benchmark through the quantitative definition of standard key factors. The method of directly assigning a zero value when the tag name does not contain the target keyword effectively reduces the false recognition rate and simplifies the calculation process.

[0035] This invention achieves flexible model adaptation through parameter configuration of amplitude adjustment coefficient γ and proportional adjustment coefficients δ1 and δ2. The range of values ​​of γ ensures the stability of the output results, while the constraint condition of δ1≥δ2 guarantees the dominant position of keyword factors in the combination calculation. The entire calculation process achieves high configurability through manual parameter setting. Users can adjust core parameters such as threshold percentile, keyword set, key factor values, and adjustment coefficients according to specific application scenarios. This flexibility enables the system to adapt to the label recognition needs under different business scenarios. At the same time, a unified calculation framework is constructed through the mathematical expression of formula (2), providing a clear theoretical basis for subsequent algorithm optimization and model iteration.

[0036] Furthermore, this invention has significant engineering value in practical applications. The hierarchical processing structure reduces system complexity, and the logical design of first judging resource call status and then classifying improves the execution efficiency of the algorithm. The independent feature extraction mechanism based on label name and type helps to achieve modular development. This design idea provides solid theoretical support and practical guidance for building an efficient, reliable, and scalable label recognition system. Finally, through the combination of parameterized configuration and mathematical modeling, the optimal utilization of system resources is achieved while ensuring recognition accuracy, providing an innovative solution for label management in intelligent data processing scenarios.

[0037] It should be noted that the order of arrangement is based on the priority recognition direction set manually. For example, if the priority recognition direction is set to the left side of the scanning area, then the labels closer to the left will be recognized first.

[0038] It should be noted that the type factor library includes: the type factor LZ value of emergency medical products is 1.25, the type factor LZ value of critical and valuable products is 1, and the type factor LZ value of ordinary products is 0.8.

[0039] It should be noted that the target keywords are manually set to prioritize identifying tags of partner companies selected by managers or tags of products with significant value.

[0040] It should be noted that the standard key factors are manually set based on the importance of the partner company or the important value product, and the standard key factors are directly proportional to the importance of the partner company or the important value product.

[0041] It should be noted that γ is used to adjust the influence of the keyword factor GZ and the type factor LZ on the identification factor SZ; when other conditions remain unchanged, the larger γ is, the greater the influence on the identification factor SZ, and the smaller α is, the smaller the influence on the identification factor SZ.

[0042] In this application, the content of each tag is obtained by content recognition based on the recognition factor, including: The identification factors of each label are extracted and numbered from largest to smallest. The labels are then scanned sequentially using a label scanning device and a camera to identify the content of each label, resulting in the text, images, text and image layouts, and the positions of text and images within the label. The text layout includes the positional relationships between text and text, and between text and images. The image layout includes the positional relationships between images, and between images and text.

[0043] In this application, the content-recognized tags are classified into complete tags and incomplete tags, including: After the label undergoes content recognition, the text and image of the label are extracted sequentially. If the text is incomplete in terms of name, specifications, ingredients, producer information, or precautions, or if the image is incomplete, the label is marked as an incomplete label. When the text is missing, the name, specifications, ingredients, manufacturer information, and precautions are incomplete, or the image is missing or incomplete, the label will be marked as a complete label.

[0044] It should be noted that if the text contains incomplete information such as name, specifications, ingredients, manufacturer information, or precautions, it can be understood as: there are missing or non-existent words in the name, specifications, ingredients, manufacturer information, or precautions; if the image is incomplete, it can be understood as: some content of the image is missing.

[0045] In this application, based on the content of the tags, complete tags similar to each incomplete tag within the scanned area are obtained, including: Extract the content of each tag from several scans in the historical reference data, as well as the complete tags that are similar to the incomplete tags in each tag; wherein, the historical reference data includes the content of each tag from several scans, as well as the complete tags that are similar to each incomplete tag obtained by manual analysis based on the content of the tags; The content of each label, as well as the complete labels that are similar to the incomplete labels, are integrated into several sets of training data and test data. The training data is used to train the artificial intelligence model, and the test data is used to test the trained artificial intelligence model. The artificial intelligence model is adjusted according to the test results. Finally, a similarity evaluation model is obtained with the content of each label as input and the complete labels that are similar to the incomplete labels as output. The artificial intelligence model includes a BP neural network model and an RBF neural network model. The content of each label in this scan is input into the similarity evaluation model to obtain the complete labels that are similar to each incomplete label in this scan.

[0046] It should be noted that complete tags similar to each incomplete tag, obtained manually based on text, images, text and image layout, and the position of text and images in the tag, as well as an incomplete tag and a complete tag similar to that incomplete tag, all belong to the same scan.

[0047] It should be noted that this invention selects complete tags similar to each incomplete tag based on the text, images, text and image layout, and the positions of the text and images within the tag. The higher the overlap between the text and images of a complete tag and an incomplete tag, the more similar they are. Similarly, the higher the overlap between the images, text layout, and image layout of a complete tag and an incomplete tag, the more similar they are. Finally, the higher the overlap between the positions of the text and images within the tag, the more similar they are.

[0048] Specifically, the steps for testing the trained AI model using validation data and adjusting the AI ​​model based on the validation results are as follows: The content of each label in the test data is input into the trained artificial intelligence model to obtain the complete labels that are similar to the incomplete labels in each label. The corresponding complete labels that are similar to the incomplete labels are compared with the corresponding complete labels that are similar to the incomplete labels in the test data. When the overlap between the two similar complete images is greater than a threshold (the threshold is obtained based on experience), no parameter adjustment is required, and the next set of test data is tested. If the overlap between the two similar complete images is not greater than the threshold, the corresponding parameters are adjusted until the number of tests corresponding to the overlap between the two similar complete images being greater than the threshold accounts for 90% or more of the total test data. At this point, a similarity evaluation model is obtained, with the content of each label as input and the complete labels that are similar to each incomplete label as output.

[0049] This application obtains similarity factors for each similar tag, including: Extract the linear distance Li between the current incomplete label and the similar labels of the current incomplete label. When the linear distance Li is greater than the linear threshold, mark the similarity factor of the corresponding complete label as 0; when the linear distance Li is not greater than the linear threshold, determine the similarity factor YZi of the similar labels of the current incomplete label based on the linear distance Li; the similarity factor YZi satisfies formula (3): YZi=1 / exp(Li / (∑(Li / n)+Mode(Li)+P(Li)))(3); Where i is the number of the similar tag of the current incomplete tag, and the value of i is [1, n], and n is the maximum value of the similar tag number; ∑ is the summation symbol, and the summation range is [1, n]; Mode(Li) symbol is to take the mode of Li, and P(Li) is to take the percentile of Li, and the percentile is manually selected.

[0050] It should be noted that P(Li) represents the percentile of Li. In this step, the percentile is the value at a specific percentile of a number of linear distances Li. For example, if the number of linear distances Li are sorted in ascending order, and the manually set percentile is 60%, then the data at the 60% position of the number of linear distances Li are the percentiles of the number of linear distances Li. If there is no data at the 60% position of the number of linear distances Li, then the data closest to the 60% position is taken as the percentile of the number of linear distances Li.

[0051] This application populates incomplete tags with content based on similarity factors and corresponding similar tags, and then reports this to the administrator, including: Extract similarity factors from similar tags, mark the similar tag with the largest similarity factor as the substitute tag, fill in the missing content of incomplete tags based on the content of the substitute tag, and report the filled record to the administrator.

[0052] Please see Figure 3 A second aspect of the present invention provides an automatic label detection system, comprising: a label recognition module, and a data acquisition module and a label information restoration module connected to the label recognition module; Data acquisition module: used to acquire the name and type of each tag within the scanning area, as well as the computing power and bandwidth of the tag scanning device; Tag recognition module: used to determine the resource access factor of the current tag scanning device based on computing power and bandwidth; determine the recognition factor of each tag based on the resource access factor, the name and type of each tag; and perform content recognition on each tag based on the recognition factor to obtain the content of each tag; wherein, the content includes text, images, text and image layout, and the position of text and images in the tag; Tag information restoration module: used to classify the tags after content recognition into complete tags and incomplete tags; when there are incomplete tags in the scanning area, the module obtains complete tags similar to each incomplete tag in the scanning area based on the content of the tags, and marks the complete tags as similar tags; obtains the similarity factor of each similar tag, fills the content of the incomplete tags based on the similar tags corresponding to the similarity factor, and reports to the administrator.

[0053] Some of the data in the above formula are calculated by removing dimensions and taking their numerical values. The formula is the closest to the real situation obtained by software simulation of a large amount of collected data. The preset parameters and preset thresholds in the formula are set by those skilled in the art according to the actual situation or obtained through simulation of a large amount of data.

[0054] Working principle of the invention: This invention first obtains the name and type of each tag within the scanning area, as well as the computing power and bandwidth of the tag scanning device; then, it determines the resource access factor of the current tag scanning device based on the computing power and bandwidth; it determines the identification factor of each tag based on the resource access factor, the name and type of each tag; it performs content recognition on each tag based on the identification factor to obtain the content of each tag; finally, it classifies the tags after content recognition into complete tags and incomplete tags; when there are incomplete tags in the scanning area, it obtains complete tags similar to each incomplete tag in the scanning area based on the tag content, and marks the complete tags as similar tags; it obtains the similarity factor of each similar tag, fills the content of the incomplete tags based on the similar tags corresponding to the similarity factor, and reports it to the administrator.

[0055] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. An automatic label detection method, characterized in that, include: Obtain the name and type of each tag within the scanning area, as well as the computing power and bandwidth of the tag scanning device; The resource allocation factor for the current tag scanning device is determined based on computing power and bandwidth. The identification factor of each tag is determined based on the resource access factor, the name and type of each tag; Content recognition is performed on each tag based on recognition factors to obtain the content of each tag; the content includes text, images, text and image layout, and the position of text and images in the tag; The content-recognized tags are categorized into complete tags and incomplete tags; When incomplete labels exist within the scanned area, complete labels similar to each incomplete label within the scanned area are obtained based on the content of the labels, and the complete labels are marked as similar labels. Obtain the similarity factor for each similar tag, fill in the incomplete tag with content based on the similar tags corresponding to the similar factors, and report to the administrator.

2. The automatic label detection method according to claim 1, characterized in that, The acquisition of the names and types of each tag within the scanning area, as well as the computing power and bandwidth of the tag scanning device, includes: The video recognition technology is used to obtain the names of each label within the manually set scanning area, as well as the type of product corresponding to each label. The computing power available to the label scanning device is obtained in real time through a computing power monitoring tool, and the bandwidth available to the label scanning device is obtained in real time through a bandwidth monitoring tool. The product types include emergency medical products, critical and valuable products, and ordinary products.

3. The automatic label detection method according to claim 1, characterized in that, The process of determining the resource allocation factor for the current tag scanning device based on computing power and bandwidth includes: Extract the computing power FL and bandwidth BR of the current tag scanning device. When the computing power FL is not less than the standard computing power BFL and the bandwidth BR is not less than the standard bandwidth BBR, the resource call factor DZ is set to the maximum value of the standard call range. When the computing power FL is less than the standard computing power BFL or the bandwidth BR is less than the standard bandwidth BBR, the resource allocation factor DZ of the current tag scanning device is determined based on the computing power FL and the bandwidth BR, and the resource allocation factor DZ satisfies formula (1): DZ=TZ×α / exp(β1×(BFL-FL) / BFL+β2×(BBR-BR) / BBR)(1); Wherein, TZ is the median value of the standard call range; α is the amplitude adjustment coefficient of the 1 / exp() function, and the value range of α is [0,2]; β1 and β2 are both proportional adjustment coefficients greater than 0, and β1+β2=1, β1≥β2; When the resource call factor DZ is greater than the maximum value of the standard call range, the resource call factor DZ is updated with the maximum value; when the resource call factor DZ is less than the maximum value of the standard call range, the resource call factor DZ is updated with the minimum value.

4. The automatic label detection method according to claim 1, characterized in that, The process of determining the identification factors for each tag based on the resource access factor, the name of each tag, and its type includes: Extract the resource call factor. When the resource call factor is not less than the call threshold, identify each label simultaneously or in the order of arrangement. The call threshold is the percentile of the standard call range. When the resource call factor is less than the call threshold, the names of each tag are extracted sequentially. When the name of the tag contains the target keyword, the keyword factor GZ of the tag is set to the standard key factor; when the name does not contain the target keyword, the keyword factor GZ of the tag is set to 0. The types of each tag are extracted sequentially, and the type factor LZ corresponding to the type is extracted from the type factor library based on the type. The identification factor SZ of each tag is determined based on the keyword factor GZ and the type factor LZ. The identification factor SZ satisfies formula (2): SZ=γ×ln(δ1×GZ+δ2×LZ)(2); Where γ is the amplitude adjustment coefficient of the ln() function, and the value range of γ is [0,2]; δ1 and δ2 are the proportional adjustment coefficients, and δ1≥δ2.

5. The automatic label detection method according to claim 1, characterized in that, The process of obtaining the content of each tag by content recognition based on recognition factors includes: The identification factors of each label are extracted and numbered from largest to smallest. The labels are then scanned sequentially using a label scanning device and a camera to identify their content, resulting in the text, images, text and image layouts, and the positions of the text and images within the labels. The text layout includes the positional relationships between text and text, and between text and images. The image layout includes the positional relationships between images, and between images and text.

6. The automatic label detection method according to claim 1, characterized in that, The process of classifying content-identified tags into complete tags and incomplete tags includes: After the label undergoes content recognition, the text and image of the label are extracted sequentially. If the text is incomplete in terms of name, specifications, ingredients, producer information, or precautions, or if the image is incomplete, the label is marked as an incomplete label. When the text is incomplete (e.g., name, specifications, ingredients, manufacturer information, precautions), or the image is incomplete, the label will be marked as a complete label.

7. The automatic label detection method according to claim 1, characterized in that, The tag-based content obtains complete tags within the scanned area that are similar to each incomplete tag, including: Extract the content of each tag from several scans in the historical reference data, as well as the complete tags that are similar to the incomplete tags in each tag; wherein, the historical reference data includes the content of each tag from several scans, as well as the complete tags that are similar to each incomplete tag obtained by manual analysis based on the content of the tags; The content of each label, as well as the complete labels that are similar to the incomplete labels, are integrated into several sets of training data and test data. The training data is used to train the artificial intelligence model, and the test data is used to test the trained artificial intelligence model. The artificial intelligence model is adjusted according to the test results. Finally, a similarity evaluation model is obtained with the content of each label as input and the complete labels that are similar to the incomplete labels as output. The artificial intelligence model includes a BP neural network model and an RBF neural network model. The content of each label in this scan is input into the similarity evaluation model to obtain the complete labels that are similar to each incomplete label in this scan.

8. The automatic label detection method according to claim 1, characterized in that, The process of obtaining the similarity factor for each similar tag includes: Extract the linear distance Li between the current incomplete label and the similar labels of the current incomplete label. When the linear distance Li is greater than the linear threshold, mark the similarity factor of the corresponding complete label as 0; when the linear distance Li is not greater than the linear threshold, determine the similarity factor YZi of the similar labels of the current incomplete label based on the linear distance Li; the similarity factor YZi satisfies formula (3): YZi=1 / exp(Li / (∑(Li / n)+Mode(Li)+P(Li)))(3); Where i is the number of the similar tag to the current incomplete tag, and the value of i is [1, n], and n is the maximum value of the similar tag number; ∑ is the summation symbol, and the summation range is [1, n]; Mode(Li) symbol is to take the mode of Li, and P(Li) is to take the percentile of Li.

9. The automatic label detection method according to claim 1, characterized in that, The process of filling in incomplete tags with content based on similar tags corresponding to similar factors and then reporting it to the administrator includes: Extract similarity factors from similar tags, mark the similar tag with the largest similarity factor as the substitute tag, fill in the missing content of incomplete tags based on the content of the substitute tag, and report the filled record to the administrator.

10. An automatic label detection system, used to run the automatic label detection method according to any one of claims 1 to 9, characterized in that, include: The label recognition module, as well as the data acquisition module and label information restoration module connected to the label recognition module; The data acquisition module is used to acquire the name and type of each label within the scanning area, as well as the computing power and bandwidth of the label scanning device. The tag recognition module is used to determine the resource allocation factor of the current tag scanning device based on computing power and bandwidth. The identification factors of each tag are determined based on the resource call factor, the name and type of each tag; the content of each tag is obtained by content recognition based on the identification factors; the content includes text, images, text and image layout, and the position of text and images in the tag; The tag information restoration module is used to classify the tags after content recognition into complete tags and incomplete tags; when there are incomplete tags in the scanning area, it obtains complete tags similar to each incomplete tag in the scanning area based on the content of the tags, and marks the complete tags as similar tags; it obtains the similarity factor of each similar tag, fills the content of the incomplete tags based on the similar tags corresponding to the similarity factor, and reports to the administrator.

Images