Man-machine cooperation rapid identification method for multi-batch products

Abstract

The invention relates to the field of machine vision, in particular to a multi-batch product man-machine cooperation rapid identification method, which comprises the following steps of: constructing a man-machine cooperation product identification process, acquiring appearance information of a plurality of packages of products in the same batch by utilizing an image acquisition module, and performing image denoising and feature enhancement through a preprocessing module; product mark features are extracted through a fusion recognition module, a manual auxiliary confirmation mechanism is combined, rapid and accurate distinguishing of different batches or the same batch of products is achieved, and meanwhile the stability and consistency of recognition results are ensured through feature fusion and a formulation calculation mechanism. The system part is composed of a data acquisition module, a feature processing module, a fusion calculation module and a result output module, a complete closed-loop identification process is formed, and correspondence and coherence of the method and the system are guaranteed.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of machine vision, and in particular to a man-machine cooperative rapid identification method for multiple batches of products. BACKGROUND

[0002] In the process of cargo storage and transportation, such as chemical raw materials, mine transportation, and pharmaceutical circulation industries, batch management and rapid identification are important links to ensure production safety and quality stability. In particular, during the storage and use of dangerous chemicals, different batches of materials often have performance differences, and if mixed, it will directly cause process failure or even serious safety accidents. Therefore, in the links of storage, transportation, wharf loading and unloading, and on-site operation, it is necessary to meet the dual needs of batch rapid registration and on-site manual confirmation. In the prior art, there are many labeling methods, such as printing batch information directly on the package or handwritten labels. Although these methods can meet the basic needs of manual identification, manual identification has low efficiency, is prone to errors, is obviously affected by color difference and lighting conditions, and is difficult to maintain stability in large-scale scenarios. At the same time, identification methods based on cameras, code guns, or machine vision have improved efficiency and accuracy to some extent, but relying completely on machines also has potential risks. Once the equipment fails or the algorithm misjudges, the on-site operator will not be able to immediately determine, especially in the management of dangerous chemicals, which will have serious consequences. Based on this, in practical applications, a man-machine coexistence identification mechanism is generally required, that is, the machine is responsible for efficient identification and traceability in the background, and the manual is responsible for on-site intuitive confirmation and emergency redundancy. However, when implementing man-machine coexistence, the prior art often needs to superimpose multiple information such as color, pattern, and code on the same package in order to consider the dual interpretation of the human eye and the algorithm. Although this method can achieve identification to some extent, it inevitably causes information overload, making manual identification too complex and increasing the cognitive burden of the operator. At the same time, it also leads to an increase in the computational load of machine identification, reducing overall efficiency. More importantly, manual identification can only be perceived at a macro scale, while machines can extract features at a pixel level or even a finer scale. This multi-scale identification difference can lead to inconsistent interpretation of the same sign between different identification subjects, resulting in distortion of batch information in cross-subject identification. Therefore, the technical problem to be solved in the field is how to ensure that the sign is simple and intuitive while also having multi-scale readability, so that it can meet the needs of manual rapid identification at a macro scale and support machine high-precision analysis at a micro scale, in order to achieve efficient, safe, and stable man-machine coexistence identification in typical scenarios such as dangerous chemical batch management. SUMMARY

[0003] The application provides a man-machine collaborative rapid identification method for multi-batch products, and solves the problems of low efficiency, easy confusion, color blindness limitation and insufficient accuracy in large batch scenarios in the prior art.

[0004] To achieve the above-mentioned purpose, the embodiments of the application disclose the following technical solutions:

[0005] In one aspect, the application discloses a man-machine collaborative rapid identification method for multi-batch products, comprising the following steps:

[0006] S1, a collection step, obtaining packaging surface mark information of target batch products, the mark information including color marks and pattern marks printed or attached to the packaging, the collection method including manual visual collection and imaging device collection, and obtaining initial mark data;

[0007] S2, a preprocessing step, performing normalization processing on the initial mark data, wherein the manual recognition part forms a naked-eye recognition result through direct observation, and the machine recognition part maps and decomposes the initial mark data through a color space mapping algorithm and a pattern decomposition recognition algorithm to obtain preprocessed mark data;

[0008] S3, a feature extraction step, taking the preprocessed mark data as input, using a multi-scale feature extraction algorithm to extract color and geometric shape features at a macro scale and extract texture and local pixel features at a micro scale, and obtaining multi-scale feature data;

[0009] S4, a fusion step, inputting the multi-scale feature data into an artificial readable sub-module and a machine readable sub-module, the artificial readable sub-module mapping macro features into intuitive recognition labels, and the machine readable sub-module converting micro features into digital feature vectors, and obtaining fusion recognition data;

[0010] S5, a comparison step, taking the fusion recognition data as input, comparing the intuitive recognition labels and the digital feature vectors with a preset batch mark library respectively, and obtaining comparison result data;

[0011] S6, a verification step, taking the comparison result data as input, performing man-machine cross verification, wherein the manual recognition result is used as a first verification path, the machine recognition result is used as a second verification path, batch confirmation information is generated when the recognition results of the two paths are consistent, a secondary recognition process is triggered when the recognition results of the two paths are inconsistent, and verification result data is obtained;

[0012] S7, an output step, taking the verification result data as input, generating a final batch identification result, and outputting to on-site operators and a background database, and completing the batch identification process.

[0013] In another aspect, the present solution discloses a human-machine collaborative rapid identification system for multi-batch products, comprising:

[0014] A data acquisition module for acquiring color pattern information carried on the surface of the product and converting it into a digital data stream;

[0015] An artificial identification auxiliary module for artificial visual identification of the color pattern information and converting the artificial identification result into a labeled input;

[0016] A preprocessing module for denoising, standardizing and feature enhancement processing of the digital data stream to obtain a first processing result;

[0017] An artificial input preprocessing module for structured processing of the labeled input to obtain an artificial processing result;

[0018] A fusion module for receiving the first processing result and the artificial processing result and performing information comparison, result weighting and difference determination to generate a fusion output;

[0019] A result generation module for generating a final identification result based on the fusion output and outputting the final identification result to a user end and a data storage end.

[0020] The present solution realizes the simultaneous improvement of identification efficiency and accuracy by designing a method for artificial rapid identification of multi-pack products in the same batch under a human-machine collaboration framework. First, the uniform setting and differential expression design of the logo information are introduced in the scheme, which makes the multi-pack products in the same batch have significant distinguishability in appearance, avoiding errors caused by similar colors or shape confusion in the identification process, thereby improving the intuitiveness and stability of artificial identification. Second, the scheme uses machine vision technology combined with a preprocessing module and a fusion module to complete rapid auxiliary identification based on artificial operation. The machine can automatically collect and analyze logo features and compare them with sample information stored in the database, effectively solving the problems of long time consumption and easy errors in artificial identification in large batch scenarios. By introducing formulaic feature matching rules, the present solution realizes the comprehensive calculation and optimization processing of multi-dimensional logo features, further improving the accuracy of machine-assisted identification. At the same time, artificial and machine form a complementary relationship in the operation process, artificial is responsible for preliminary logo confirmation and difference setting, machine is responsible for batch detection and rapid discrimination, which significantly reduces the repetitive labor and cognitive burden. Overall, the present solution takes into account the flexibility of artificial and the efficiency of machine, realizes the organic combination of human-machine collaboration, and ensures that multi-pack products in the same batch can be quickly and accurately distinguished and identified in complex and variable production and circulation scenarios, thereby improving the efficiency and reliability of production management. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1Method flowchart for embodiment one of the present invention;

[0022] Figure 2 System diagram for embodiment two of the present invention;

[0023] Figure 3 Module interaction diagram for embodiment two of the present invention. DETAILED DESCRIPTION

[0024] Reference will now be made in detail to the present embodiments of the application. While the application will be described in conjunction with these specific embodiments, it will be understood that it is not intended to limit the application to these specific embodiments. On the contrary, it is intended to cover alternatives, modifications, and equivalents, which can be included within the spirit and scope of the application as defined by the claims. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. The present application can be practiced without some or all of these specific details. In other instances, well known process operations have not been described in detail in order not to unnecessarily obscure the present application.

[0025] As used in this description and the accompanying claims, the singular forms "a," "an," and "the" include plural references unless the context clearly dictates otherwise. Unless defined otherwise, all technical and scientific terms used herein have the same meanings as commonly understood by one of ordinary skill in the art to which this application belongs.

[0026] Embodiment one

[0027] A man-machine collaborative rapid identification method for multi-batch products, comprising the following steps:

[0028] S1, a collection step, obtaining the packaging surface mark information of the target batch product, the mark information including color marks and pattern marks printed or attached to the packaging, the collection method including manual visual collection and imaging device collection, obtaining initial mark data;

[0029] S2, a preprocessing step, normalizing the initial mark data, wherein the manual recognition part forms a naked eye recognition result by direct observation, and the machine recognition part maps and decomposes the initial mark data through a color space mapping algorithm and a pattern decomposition recognition algorithm, obtaining preprocessed mark data;

[0030] S3, a feature extraction step, taking the preprocessed mark data as input, using a multi-scale feature extraction algorithm to extract color and geometric shape features at a macro scale and texture and local pixel features at a micro scale, obtaining multi-scale feature data;

[0031] S4, a fusion step, inputting the multi-scale feature data into an artificial readable sub-module and a machine readable sub-module, the artificial readable sub-module mapping macro features into intuitive identification labels, and the machine readable sub-module converting micro features into digital feature vectors, to obtain fusion identification data;

[0032] S5, a comparison step, inputting the fusion identification data, comparing the intuitive identification labels and the digital feature vectors respectively with a preset batch mark library, to obtain comparison result data;

[0033] S6, a verification step, inputting the comparison result data, performing human-machine cross verification, wherein the artificial identification result is a first verification path, and the machine identification result is a second verification path, when the identification results of the two paths are consistent, generating batch confirmation information, and when the identification results of the two paths are inconsistent, triggering a secondary identification process, to obtain verification result data;

[0034] S7, an output step, inputting the verification result data, generating a final batch identification result, and outputting to on-site operators and a background database, to complete the batch identification process.

[0035] In the embodiment, the overall process of the method includes seven main steps of collection, preprocessing, feature extraction, fusion, comparison, verification and output, which are completed by artificial operation and machine processing in actual scenarios. Taking the identification of batch products in an industrial site as an example, first in the S1 collection step, the operator directly identifies the color marks or pattern marks carried on the product surface by visual observation, and saves them in the form of brief symbols, written records or handheld terminal inputs; at the same time, imaging devices such as industrial cameras and line array scanners take pictures of the product surface to obtain high-resolution image data. The artificial collection method can ensure continuous operation under unstable lighting conditions or when the device cannot take pictures, while the machine collection method can quickly and batchingly obtain standardized digital image data.

[0036] In actual scenarios, product surface marks can take various forms, for example: color marks can be red, blue, green, black or yellow areas; shape marks can be circles, triangles, rectangles, diamonds or stars; symbol marks can be letters A, B, numbers 1, 2, or abstract symbols such as arrows and bars. These marks meet the needs of artificial intuitive identification and can also be converted into digital features through machine image processing. Through human-machine dual-channel collection, the system can ensure the integrity and robustness of the collection under various complex conditions

[0037] The scheme further proposes that the color space mapping algorithm is a mapping operation based on a multi-channel color vector, which is used to enhance the recognition stability under different lighting conditions.

[0038] In the S2 preprocessing step, the initial mark data collected needs to be normalized to eliminate differences caused by light, angle, lens distortion or material reflection. Part of the manual identification relies on the intuitive judgment of the operator, and the results are usually presented in the form of language description or symbols, such as "red circle" and "blue letter B". In order to ensure that the manual results can correspond to the machine processing results, the system uniformly converts the manual input into a structured description to form a standardized annotation input.

[0039] The machine recognition part adopts color space mapping algorithm and pattern decomposition recognition algorithm. Color space mapping reduces the deviation caused by light fluctuation by converting data in RGB or HSV channel into a standardized multi-dimensional vector; the pattern decomposition recognition algorithm distinguishes different shapes under the same color through boundary detection and texture decomposition. For example, when two batches use red marks, the manual may record as "red circle" and "red triangle", and the machine distinguishes the shape through boundary feature extraction and keeps the consistency of color recognition through normalization mapping. The preprocessing step realizes the unification of manual input and machine data, so that the subsequent feature extraction and fusion can run under the same benchmark.

[0040] The scheme further proposes that the pattern decomposition recognition algorithm includes shape boundary extraction and texture pattern decomposition, which is used to distinguish different batches under the same color condition.

[0041] In the S3 feature extraction step, the system performs multi-scale feature extraction on the preprocessed mark data. At the macro scale, the overall boundary of color histogram distribution and geometric shape is extracted; at the micro scale, the texture pattern and local pixel difference are extracted.

[0042] For example, if one batch is "red circle" and the other batch is "red triangle", the geometric boundary features at the macro scale can directly distinguish the two types of products; if one batch is "blue letter A" and the other batch is "black letter A", the color histogram at the macro scale can form obvious differences. For more complex cases, such as similar color and shape but different texture, the pixel difference feature at the micro scale can capture small differences, thereby avoiding the ambiguity of manual identification. This method ensures the accuracy and robustness of batch differentiation.

[0043] The scheme further proposes that the multi-scale feature extraction algorithm includes regional statistics at the macro scale and pixel difference at the micro scale, and the feature importance is weighted based on the following saliency score function:

[0044] ;

[0045] Where, is the a saliency score of a feature, denotes a feature function, denotes a corresponding weight parameter, is a feature dimension.

[0046] In the refinement process of the feature extraction algorithm, the macro scale adopts the regional statistical method to calculate the color proportion, geometric shape area, etc. to form a stable overall description; the micro scale adopts the pixel difference method to form the description of texture and local features through the difference of adjacent pixel gray values or color vectors. The final feature set is expressed by the formula:

[0047] ;

[0048] wherein, is a macro statistical feature set, is a micro pixel difference feature set. The feature set makes the system capture both overall saliency features and detect detail differences through the complement of the two types of information, thereby providing solid input for subsequent fusion.

[0049] The scheme further proposes that the fusion recognition data realizes the unified storage of the artificial readable label and the digital feature vector through a weighted mapping model, and the weighted mapping model meets the following optimization objectives:

[0050] ;

[0051] wherein, is a path cost function, is the collection priority weight of the i-th item, denotes the difference measure between the artificial label and the machine prediction value, is the number of comparison samples.

[0052] In the S4 fusion step, the artificial readable sub-module maps the macro features into intuitive labels such as "green circle" or "red letter A" and stores them as markers that can be directly understood by the operator; the machine readable sub-module converts the micro features into high-dimensional digital vectors for computer retrieval and comparison. Both are stored uniformly through a weighted mapping model, and the calculation formula is: ;

[0053] wherein, L is the artificial label vector, V is the machine feature vector, and α and β are the weighted parameters. The model ensures the balance between artificial intuitive understanding and machine high-precision feature representation. For example, when the artificial input is "blue triangle" and the machine extracted feature vector also shows high confidence blue and triangular boundary features, the fusion result Z can confirm the batch with high confidence, reducing the misjudgment caused by a single path. ​

[0054] The scheme further proposes that in the comparison step, the preset batch mark library includes a historical batch label library and a standard batch pattern library.

[0055] In the S5 comparison step, the fusion recognition data is input into the preset batch mark library. The library includes two parts: a historical batch label library and a standard batch pattern library. The historical library is used for traceable records, such as "Batch 1 in 2023 is a blue rectangle" and "Batch 2 in 2023 is a green circle"; the standard library is a fixed reference established by the enterprise, such as "black triangle symbol" or "yellow letter C". The comparison process includes both text comparison of manual labels and historical libraries and similarity calculation of feature vectors and standard libraries. Through this double-layer comparison method, recognition errors caused by a single database or a single path can be avoided, and the reliability of batch confirmation can be improved.

[0056] The scheme further proposes that the secondary recognition process of the verification step includes reacquiring mark information and limiting comparison at a microscopic scale.

[0057] In the S6 verification step, the system uses a cross-verification mechanism of human-machine dual paths. The manual recognition result is the first verification path, and the machine recognition result is the second verification path. When they are consistent, batch confirmation information is directly generated; if they are not consistent, a secondary recognition process is triggered. The secondary recognition process includes reacquiring product surface images and performing more detailed feature comparison at a microscopic scale. For example, when a person mistakenly identifies "dark blue rectangle" as "black rectangle", the microscopic pixel difference can distinguish the color features of the two, thereby correcting the human error. This mechanism effectively reduces the misjudgment rate and improves the robustness of the overall system.

[0058] The scheme further proposes that the output step provides the batch recognition result to the operator in an intuitive display manner and stores it in the background database in the form of digital data.

[0059] In the S7 output step, the final recognition result is presented in two forms at the same time: on the one hand, it is displayed in the form of intuitive text or symbols in the field operation terminal, such as "Batch confirmation: green triangle"; on the other hand, it is written in the form of digital data into the background database, realizing long-term storage and traceability of batch information. The field operator can quickly complete tasks such as sorting and checking according to the output information, and the background database provides reliable data support for production management and logistics tracking.

[0060] Embodiment Two

[0061] A human-machine collaborative rapid identification system for multiple batches of products, comprising:

[0062] A data acquisition module for acquiring color pattern information carried on the surface of the product and converting it into a digital data stream;

[0063] An artificial recognition auxiliary module is configured to manually visually recognize the color pattern information and convert the artificial recognition result into a label input;

[0064] A preprocessing module is configured to perform denoising, standardization, and feature enhancement processing on the digitized data stream to obtain a first processing result;

[0065] An artificial input preprocessing module is configured to perform structured processing on the label input to obtain an artificial processing result;

[0066] A fusion module is configured to receive the first processing result and the artificial processing result and perform information comparison, result weighting, and difference determination to generate a fusion output;

[0067] A result generation module is configured to generate a final recognition result based on the fusion output and output the final recognition result to a user end and a data storage end.

[0068] The system is composed of a data acquisition module, an artificial recognition auxiliary module, a preprocessing module, an artificial input preprocessing module, a fusion module, and a result generation module. The data acquisition module acquires images through an industrial camera and converts the images into a digital stream. The artificial recognition auxiliary module allows an operator to directly input a result in the form of a word or a symbol, such as “red circle” or “green letter B”. The preprocessing module performs denoising and normalization on machine data. The artificial input preprocessing module converts natural language input into a structured description. The fusion module performs comparison, weighting, and difference determination on artificial and machine data. The result generation module outputs a final result to a user end and a background database, thereby realizing batch recognition under human-machine collaboration.

[0069] The scheme further proposes that the fusion module includes an artificial readable sub-module and a machine readable sub-module. The artificial readable sub-module is configured to output a visual recognition label, and the machine readable sub-module is configured to output a digitized feature vector.

[0070] The fusion module is specifically divided into an artificial readable sub-module and a machine readable sub-module. The artificial readable sub-module directly converts macro features into an artificial understandable label, such as “blue circle”. The machine readable sub-module converts micro features into a vector form and stores the vector form in a database for batch retrieval. The result generation module combines the two to generate a final recognition result and synchronizes the final recognition result to a terminal and a database. For example, when the artificial input is consistent with the machine features, “batch confirmation: blue circle” is output. If the artificial input is inconsistent with the machine features, a secondary recognition process is triggered to ensure the accuracy and traceability of the result. This design makes the entire system meet the intuitive understanding of the operator and the digital storage and comparison requirements of the background data, thereby balancing the real-time performance of the industrial site and the standardization of the management system.

[0071] It should be pointed out finally that the above examples are only used to illustrate the technical solutions of the present application but not to limit it; although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the specific embodiments of the present application can be modified or some technical features can be replaced by equivalent ones without departing from the spirit of the technical solutions of the present application, and all of them should be covered in the technical solution range claimed by the present application.

Claims

1. A human-machine collaborative rapid identification method for multiple batches of products, characterized in that, Includes the following steps: S1. Collection step: Obtain the marking information on the packaging surface of the target batch of products. The marking information includes color markings and pattern markings printed or affixed to the packaging. The collection methods include manual visual collection and imaging equipment collection to obtain initial marking data. S2. Preprocessing step: Normalize the initial marker data. The manual recognition part forms a visual recognition result through intuitive observation, while the machine recognition part maps and decomposes the initial marker data through a color space mapping algorithm and a pattern decomposition recognition algorithm to obtain preprocessed marker data. S3. Feature extraction step: Using the preprocessed flag data as input, a multi-scale feature extraction algorithm is used to extract color and geometric shape features at the macro scale and texture and local pixel features at the micro scale to obtain multi-scale feature data. S4. Fusion step: Input the multi-scale feature data into the human-readable submodule and the machine-readable submodule. The human-readable submodule maps macroscopic features into intuitive recognition labels, and the machine-readable submodule converts microscopic features into digital feature vectors to obtain fused recognition data. S5. Comparison step: Using the fused identification data as input, the intuitive identification label and the digital feature vector are compared with the preset batch label library respectively to obtain the comparison result data. S6. Verification step: Using the comparison result data as input, perform human-machine cross-verification, wherein the human recognition result is used as the first verification path and the machine recognition result is used as the second verification path. When the recognition results of the two paths are consistent, batch confirmation information is generated. When the recognition results of the two paths are inconsistent, a secondary recognition process is triggered to obtain the verification result data. S7. Output step: Using the verification result data as input, generate the final batch recognition result and output it to the on-site operator and the background database to complete the batch recognition process.

2. The human-machine collaborative rapid identification method for multiple batches of products according to claim 1, characterized in that, The color space mapping algorithm is a mapping operation based on multi-channel color vectors, used to enhance recognition stability under different lighting conditions.

3. The human-machine collaborative rapid identification method for multiple batches of products according to claim 1, characterized in that, The pattern decomposition and recognition algorithm includes shape boundary extraction and texture pattern decomposition, which are used to distinguish different batches under the same color conditions.

4. The human-machine collaborative rapid identification method for multiple batches of products according to claim 1, characterized in that, The multi-scale feature extraction algorithm includes macro-scale region statistics and micro-scale pixel difference.

5. The human-machine collaborative rapid identification method for multiple batches of products according to claim 1, characterized in that, The fused identification data achieves unified storage of human-readable labels and digital feature vectors through a weighted mapping model.

6. The human-machine collaborative rapid identification method for multiple batches of products according to claim 1, characterized in that, In the comparison step, the preset batch mark library includes a historical batch label library and a standard batch pattern library.

7. The human-machine collaborative rapid identification method for multiple batches of products according to claim 1, characterized in that, The secondary identification process of the verification step includes re-collecting the marker information and comparing it at a microscopic scale.

8. The human-machine collaborative rapid identification method for multiple batches of products according to claim 1, characterized in that, The output step provides the batch identification results to the operator in an intuitive display manner, and at the same time stores them in the background database in the form of digital data.

9. A human-machine collaborative rapid identification system for multiple batches of products, characterized in that, include: The data acquisition module is used to collect color and pattern information carried on the product surface and convert it into a digital data stream; The manual recognition assistance module is used to perform manual and intuitive recognition of the color pattern information and convert the manual recognition results into labeled inputs. The preprocessing module is used to perform denoising, standardization, and feature enhancement on the digitized data stream to obtain a first processing result; The manual input preprocessing module is used to perform structured processing on the labeled input to obtain the manual processing result; The fusion module is used to receive the first processing result and the manual processing result and perform information comparison, result weighting and difference determination to generate fusion output; The result generation module is used to generate a final recognition result based on the fusion output and output the final recognition result to the user terminal and the data storage terminal.

10. The human-machine collaborative rapid identification system for multiple batches of products according to claim 9, characterized in that, The fusion module includes a human-readable submodule and a machine-readable submodule. The human-readable submodule is used to output intuitive identification labels, and the machine-readable submodule is used to output digital feature vectors.

Images