Intelligent automatic labeling method and system based on multi-variety mixed production

Abstract

The invention discloses an intelligent automatic labeling method and system based on multi-variety mixed production, and relates to the technical field of intelligent labeling, precise labeling is achieved through three core processes, the first step is that a hardware-software-data three-layer intelligent operation system is constructed, and posture and position analysis is conducted on SKU commodities transmitted to labeling equipment; and marking the to-be-labeled commodities meeting the conditions. 2, a to-be-labeled commodity is transmitted to a visual collection area, an algorithm reasoning mode is evaluated, and appearance characteristic parameters and dynamic parameters of a single-module reasoning commodity are accurately extracted; and step 3, analyzing and generating a label template, labeling position coordinates, labeling pressure and labeling speed according to the extracted parameters, and finally integrating into a labeling execution parameter set containing specific numerical values, constraint conditions and execution priorities, so as to meet the efficient and accurate labeling requirements of multi-variety mixed production.

Description

Technical Field

[0001] This invention relates to the field of intelligent labeling technology, and more specifically to an intelligent automatic labeling method and system based on multi-variety mixed production. Background Technology

[0002] As flexible manufacturing extends to the retail sector, supermarkets, as the core link in commodity circulation, face high-frequency, multi-category, and fast-turnover labeling demands. A single supermarket needs to process thousands of products from different brands and specifications every day. These products not only have large differences in specifications, but also have diverse materials, complex packaging forms, and dynamically changing label information. Therefore, there is a need for intelligent automatic labeling methods and systems based on multi-variety mixed production.

[0003] Currently, most supermarkets rely on manual labeling or semi-automated equipment. When faced with scenarios such as relabeling near-expiry products or adjusting prices for promotional activities, manual adjustment of label position and content is required for each item, which presents the following problems: 1. The strong reliance on manual operation in existing supermarket labeling technologies has become the core driving force for the research and development of intelligent automatic labeling systems.

[0004] 2. The shortcomings of existing labeling equipment in adapting to complex packaging directly drive the innovation of intelligent labeling methods for multi-variety mixed production.

[0005] 3. The lag in response of existing technologies to SKU switching and sudden demands highlights the necessity of intelligent reasoning models. Summary of the Invention

[0006] To address the aforementioned technical shortcomings, the purpose of this invention is to provide an intelligent automatic labeling method and system based on multi-variety mixed production.

[0007] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: In the first aspect, the present invention provides an intelligent automatic labeling method based on multi-variety mixed production, including: Step 1, SKU product labeling analysis: an automated labeling intelligent operation system is constructed in the labeling factory of the target supermarket, so that when each SKU product in the target production line is transmitted to the labeling equipment, the transmission posture and position of each SKU product are analyzed, and each SKU product that meets the labeling conditions is marked as the SKU product to be labeled.

[0008] Step 2, Image Acquisition and Feature Extraction: When each SKU to be labeled is transmitted to the visual acquisition area, the algorithm inference mode corresponding to each SKU to be labeled is evaluated, and then the appearance feature parameters and dynamic parameters corresponding to each single-module inference SKU are obtained.

[0009] Step 3: Generation of SKU labeling parameters: Based on the appearance feature parameters and dynamic parameters of the SKU products inferred by each single module, the label template, labeling position coordinates, labeling pressure and labeling speed of the SKU products inferred by each single module are analyzed, and then the labeling execution parameter set of the SKU products inferred by each single module is generated.

[0010] In a second aspect, the present invention provides an intelligent automatic labeling system based on multi-variety mixed production, comprising: an SKU product labeling analysis module: constructing an automated labeling intelligent operation system in the target labeling factory, thereby analyzing the transmission posture and position of each SKU product after each SKU product in the target production line is transmitted to the labeling equipment, and marking each SKU product that meets the labeling conditions as the SKU product to be labeled.

[0011] Image acquisition and feature extraction module: When each SKU to be labeled is transmitted to the visual acquisition area, the algorithm inference mode corresponding to each SKU to be labeled is evaluated, and then the appearance feature parameters and dynamic parameters corresponding to each SKU inferred by each single module are obtained.

[0012] SKU Labeling Parameter Generation Module: Based on the appearance feature parameters and dynamic parameters of the SKU products inferred from each single module, this module analyzes the label template, labeling position coordinates, labeling pressure, and labeling speed of the SKU products inferred from each single module, and then generates the labeling execution parameter set for the SKU products inferred from each single module.

[0013] The beneficial effects of this invention are as follows: 1. The embodiments of this invention, through a fully automated parameter generation and execution mechanism, completely change the traditional labeling method's strong reliance on manual operation. On the one hand, the system achieves full automation of attitude analysis, inference mode evaluation, and labeling parameter generation through a "hardware layer-software layer-data layer" architecture, eliminating the need for manual intervention in label calibration, parameter debugging, and other processes. This reduces the average daily labor cost per store by more than 60%, while also eliminating label misapplication and omissions caused by manual operation, reducing the rework rate from 30% to below 5%. On the other hand, the standardized parameter system and dynamic compensation mechanism ensure stable labeling accuracy, controlling label offset within ±0.5mm for flat products and reducing the wrinkle rate of curved products to less than 3%, significantly improving the aesthetics of product packaging and the standardization of information display.

[0014] 2. In this embodiment of the invention, addressing the limitations of traditional equipment in adapting to complex packaging, the solution constructs a dynamic mapping system of "appearance features - labeling parameters," significantly improving adaptability in multi-SKU mixed production scenarios. Through the collaborative acquisition of data from a 20-megapixel industrial camera and a laser contour sensor, the system can accurately extract appearance feature parameters of planar / curved surfaces and products of different materials, such as radius of curvature and surface reflectivity. Using technologies such as dynamic label template generation and rotation compensation coordinate calculation, it achieves accurate labeling of complex products such as curved bottles, flexible packaging, and transparent materials. In practical applications, the labeling qualification rate for multiple product varieties has increased from the traditional 85% to over 99%, completely resolving common problems such as label wrinkles, offsets, and recognition failures.

[0015] 3. In this embodiment of the invention, the adaptive algorithm inference mode and the second-level parameter generation mechanism significantly improve the dynamic response efficiency of the production line. The system automatically switches inference modes based on features such as the complexity of the product outline and the proportion of reflective area—simple products use algorithmic inference for rapid parameter calls, while complex products utilize single-module inference for refined parameter calculations. No manual debugging is required when switching SKUs, and parameter generation time is reduced from the traditional 5-15 minutes to seconds. Simultaneously, the edge computing gateway supports real-time storage of production logs and dynamic updates of the parameter mapping table. In response to sudden demands such as relabeling near-expiry products or promotional activities, the model and parameters can be quickly updated via a remote maintenance interface, with response latency controlled within 5 minutes. This effectively reduces the risk of product backlog and improves supermarket replenishment efficiency and promotional flexibility. Attached Figure Description

[0016] 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.

[0017] Figure 1 This is a flowchart illustrating the implementation steps of the method of the present invention.

[0018] Figure 2 This is a schematic diagram of the system module connections of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. 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.

[0020] Examples of embodiments of the present invention Figure 1 As shown, the intelligent automatic labeling method based on multi-variety mixed production includes the following steps: Step 1, SKU labeling analysis: An automated labeling intelligent operation system is built in the target supermarket labeling factory. When each SKU product in the target production line is transferred to the labeling equipment, the transfer posture and position of each SKU product are analyzed, and each SKU product that meets the labeling conditions is marked as the SKU product to be labeled.

[0021] In a specific embodiment, the construction process of the automated labeling intelligent operation system in the target labeling factory is as follows: A three-layer architecture of "hardware layer - software layer - data layer" is established. The hardware layer deploys industrial cameras, laser contour sensors, labeling actuators, and industrial control units, equipped with Intel i7 processors, 16GB of memory, and NVIDIA graphics cards with a computing power of ≥8.6 GHz, and achieves communication between devices via Ethernet. The software layer integrates a deep SKU recognition model library and a labeling parameter management module, and develops a remote operation and maintenance interface to support online model updates and function upgrades. The data layer constructs an SKU sample database, storing images and labeling data of various product types, labeling parameter mapping tables, and a production log library. Real-time data interaction and local storage are achieved through an edge computing gateway, forming a closed-loop automated labeling intelligent operation system of "perception-decision-execution-optimization".

[0022] In a specific embodiment, the analysis of the transmission posture and position corresponding to each SKU product is performed as follows: A1. When each SKU product is conveyed to the labeling equipment entrance area via the conveyor belt, the laser displacement sensor at the equipment entrance scans the surface of the conveyor belt in real time. When a product is detected to be blocking the laser signal, the visual pre-positioning device is triggered to start in conjunction with the conveyor belt speed control motor. The operating frequency of the speed control motor is dynamically adjusted according to the product spacing, and the calculation formula is as follows: ,in, The operating frequency of the speed-regulating motor, This is the inherent coefficient of the equipment. As the production line reference speed, For real-time detection of product spacing.

[0023] It should be noted that the inherent coefficient of the equipment is a fixed parameter preset by calibration and testing before the labeling equipment leaves the factory. It is determined through offline calibration based on the hardware characteristics of the equipment, such as the motor model and the transmission ratio of the conveyor belt, and is stored in the parameter configuration library of the industrial control unit. The production line reference speed is the standard operating speed preset for this labeling station. It is set by the production management system according to the daily production capacity requirements and is adjusted within the range of 0-60m / min through the remote operation and maintenance interface and synchronized to the control unit. The real-time detection of the product spacing is obtained by continuous scanning by the laser displacement sensor at the equipment entrance. The sensor emits a laser beam at a frequency of 100Hz to the surface of the conveyor belt. When a product passes by, it blocks the laser signal. The system calculates the real-time spacing value by combining the time interval between two adjacent blocking signals with the current conveyor belt speed and dynamically feeds it back to the speed control motor module.

[0024] A2. Each SKU is fed into the positioning and detection area via a speed-adjustable conveyor belt. Two sets of laser contour sensors are installed at 300mm intervals along the conveying direction in this area. The first set of laser contour sensors collects the front contour coordinates of each SKU. The second set of laser contour sensors outputs the coordinates of the rear contour of the product. .

[0025] A3. Import the two sets of contour coordinate data for each SKU into the attitude analysis algorithm: Using the center line of the conveyor belt as the reference Y-axis, calculate the horizontal deviation between the center point of the front contour of each SKU and the reference axis: ,in, (referring to the X-axis as the reference axis), the horizontal deviation between the center point of the rear profile of each SKU and the reference axis. If a certain SKU product If the transmission posture is correct, the SKU is deemed to be acceptable; otherwise, it is marked as unacceptable.

[0026] A4. Calculate the positional parameters of the product's transport direction using the two sets of contour coordinates for each SKU; calculate the actual length of the product using the distance formula between two points. The estimated time for the product to reach the labeling station is calculated by combining the conveyor belt speed. ,in, The distance from the positioning area to the labeling station is calculated; simultaneously, the longitudinal difference between the two sets of coordinates is calculated. If a certain SKU product If the position is correct, it is considered compliant; otherwise, it is marked as a position offset.

[0027] In a specific embodiment, the step of marking each SKU that meets the labeling conditions as a SKU to be labeled is carried out as follows: If the posture of a certain SKU is qualified and the position is compliant, it is directly transported to the labeling preparation area by the conveyor belt to wait for labeling; if the posture of a certain SKU is unqualified or the position is non-compliant, the pneumatic push-alignment device on the side of the positioning area is triggered to push the product to rotate until the posture of the SKU is qualified and the position is compliant; and each SKU with qualified posture and compliant position is marked as a SKU to be labeled and proceeds to the next process.

[0028] Step 2, Image Acquisition and Feature Extraction: When each SKU to be labeled is transmitted to the visual acquisition area, the algorithm inference mode corresponding to each SKU to be labeled is evaluated, and then the appearance feature parameters and dynamic parameters corresponding to each single-module inference SKU are obtained.

[0029] In a specific embodiment, the algorithm reasoning mode for obtaining and evaluating each SKU product to be labeled is specifically evaluated as follows: B1. Obtain product feature data for each SKU product to be labeled. The product feature data includes appearance contour complexity, surface reflective area ratio, and material hardness level. Compare the appearance contour complexity, surface reflective area ratio, and material hardness level of each SKU product to be labeled with the corresponding score table to obtain the appearance contour complexity score, surface reflective area ratio score, and material hardness level score of each SKU product to be labeled.

[0030] It should be noted that the complexity of the product's appearance is determined by capturing 360-degree images of the product using a 20-megapixel industrial camera. Edge detection algorithms are used to extract the number of outline lines, corner density, and irregularity parameters. Products with more than 50 outline lines, a corner density greater than 10 / cm², or an irregularity greater than 0.6 are defined as high complexity; otherwise, they are classified as medium to low complexity. A pre-defined scoring table is then used to generate an appearance complexity score: 6-10 points for high complexity, 3-5 points for medium complexity, and 1-2 points for low complexity. The surface reflective area ratio is determined by capturing images of the product's surface using a camera equipped with a polarizing filter. A grayscale threshold segmentation algorithm is used to identify reflective areas, specifically pixels with a grayscale value greater than 220. The ratio of the reflective area to the total area of ​​the labeling area is calculated, categorized as ≤5%, 5%-20%, and >20%. The reflective area ratio scores correspond to 1-3, 4-7, and 8-10 points, respectively. The material hardness level is determined by collecting the surface hardness value of the product through a contact hardness sensor and combining it with the material database matching results. Metal / hard plastic, soft plastic, and paper / soft materials are assigned material hardness level scores of 8-10, 4-7, and 1-3 points, respectively. All three types of scores are stored in the SKU feature database of the data layer for subsequent inference mode evaluation.

[0031] B2. Then compare the scores for the appearance complexity, surface reflective area ratio, and material hardness level of each SKU to be labeled with the corresponding score thresholds.

[0032] B3. If any one of the following scores for a given SKU—appearance complexity, surface reflective area ratio, and material hardness grade—is greater than the corresponding threshold, then the algorithm reasoning mode for that SKU is single-module reasoning. Conversely, if all three scores are less than or equal to the corresponding threshold, then the algorithm reasoning mode for that SKU is algorithm scheme reasoning. SKUs with a single-module reasoning mode are then designated as single-module reasoning SKUs.

[0033] B4. If the algorithm reasoning mode corresponding to a certain SKU to be labeled is algorithm scheme reasoning, then there is no need to adjust the labeling parameters of the SKU to be labeled. It is only necessary to complete the parameter preparation before labeling through the preset standardized labeling parameters. There is no intermediate parameter adjustment step in the whole process. If the algorithm reasoning mode corresponding to a certain SKU to be labeled is single module reasoning, then the labeling parameters of the SKU to be labeled need to be adjusted.

[0034] It should be noted that during the algorithm's inference process, the data layer calls pre-stored standardized labeling parameters, including the label size baseline value, the default offset of the labeling position, the fixed pressure value of 0.2MPa, and the basic speed of 30m / min. These parameters are generated by aggregating the optimal labeling data of similar simple products in history and are fixed in the parameter mapping table. If any feature score > 5 points triggers single-module inference, the system starts the dynamic parameter calculation process, calls the real-time appearance feature parameters and dynamic parameters acquired by vision, and regenerates personalized parameters adapted to the product through the label template generation formula, position compensation algorithm, and pressure and speed calculation formula.

[0035] Step 3: Generation of SKU labeling parameters: Based on the appearance feature parameters and dynamic parameters of the SKU products inferred by each single module, the label template, labeling position coordinates, labeling pressure and labeling speed of the SKU products inferred by each single module are analyzed, and then the labeling execution parameter set of the SKU products inferred by each single module is generated.

[0036] In a specific embodiment, the specific generation process of the label template corresponding to each single-module inference SKU product is as follows: C1. When a single-module inference SKU product is a planar product, a high-definition image of the product labeling area is acquired using a 20-megapixel industrial camera with a ring light source, and the physical width and height of the labeling area are extracted; the maximum allowable size of the label is generated based on the size constraint rules, and the calculation formula is: the label width of the single-module inference SKU product ≤ the width of the product labeling area × 80%, and the label height ≤ the height of the product labeling area × 80%, where the 80% ratio is set to reserve 5% edge margin on each side to avoid the label exceeding the labeling area boundary.

[0037] C2. If a single-module inference SKU is a curved surface product, activate the laser contour sensor in the positioning detection area to scan the surface contour of the product and generate three-dimensional point cloud data to obtain the radius of curvature of the single-module inference SKU. Based on the formula for the circumference of a cylinder, derive the label unfolding length limit. The calculation formula is: label unfolding length ≤ π × (radius of curvature of the single-module inference SKU × 2) × 70%. The 70% proportional constraint can prevent the label from wrinkling due to stretching or overlapping when it is attached to the curved surface, ensuring that the label is flat and attached.

[0038] C3. Generate a label template for each inference SKU product, including size parameters, edge allowance, and layout baseline, based on the label width, height, or unfolded length calculated for each individual module inference SKU product.

[0039] In a specific embodiment, the analysis of the label position coordinates corresponding to each single-module inference SKU product is carried out as follows: The label position coordinates of each single-module inference SKU product are calculated based on the relative offset of its bounding box. The calculation formula is: the actual label X coordinate of each single-module inference SKU product = its bounding box reference X coordinate + ΔX, and the actual label Y coordinate of each single-module inference SKU product = its bounding box reference Y coordinate + ΔY. ΔX and ΔY are calculated by converting the design drawing data of the SKU product in the single module inference with the dynamic parameters of visual acquisition: ΔX = (X coordinate of the product design label - physical coordinate of the left edge of the product) - (X coordinate of the bounding box center - pixel coordinate of the left edge of the image) × pixel physical size. The pixel physical size is obtained through camera calibration. Similarly, ΔY is calculated. At the same time, for the attitude fluctuation of each SKU product in the single module inference during transmission, a rotation compensation formula is introduced: after correction, ΔX = ΔX × cosθ - ΔY × sinθ, after correction, ΔY = ΔX × sinθ + ΔY × cosθ, where θ is the maximum rotation angle fluctuation value of the product acquired through dynamic parameters.

[0040] In a specific embodiment, the analysis of the label position coordinates corresponding to each single-module inference SKU product is carried out as follows: The label position coordinates of each single-module inference SKU product are calculated based on the relative offset of its bounding box. The calculation formula is: the actual label X coordinate of each single-module inference SKU product = its bounding box reference X coordinate + ΔX, and the actual label Y coordinate of each single-module inference SKU product = its bounding box reference Y coordinate + ΔY. ΔX and ΔY are calculated by converting the design drawing data of the SKU product in the single module inference with the dynamic parameters of visual acquisition: ΔX = (X coordinate of the product design label - physical coordinate of the left edge of the product) - (X coordinate of the bounding box center - pixel coordinate of the left edge of the image) × pixel physical size. The pixel physical size is obtained through camera calibration. Similarly, ΔY is calculated. At the same time, for the attitude fluctuation of each SKU product in the single module inference during transmission, a rotation compensation formula is introduced: after correction, ΔX = ΔX × cosθ - ΔY × sinθ, after correction, ΔY = ΔX × sinθ + ΔY × cosθ, where θ is the maximum rotation angle fluctuation value of the product acquired through dynamic parameters.

[0041] In a specific embodiment, the labeling pressure of each single-module reasoning SKU is analyzed as follows: The labeling pressure of each single-module reasoning SKU is calculated by combining appearance feature parameters and dynamic parameters. The calculation formula is: Labeling pressure of each single-module reasoning SKU = (its material hardness coefficient × 0.3) + (its roughness coefficient × 0.1) - (its label adhesion coefficient × 0.1).

[0042] It should be noted that the material hardness coefficient is determined based on the material type in the product appearance feature parameters. Metal or hard plastic materials are assigned a value of 1.0 after visual recognition and matching with the material database, while soft plastic or paper materials are assigned a value of 0.5. The determination result is synchronized to the labeling parameter management module. The roughness coefficient is obtained by collecting three-dimensional point cloud data of the product labeling area surface through a laser contour sensor to obtain the surface roughness Ra value. If Ra > 1.6μm, it is assigned a value of 1.0, and if Ra ≤ 0.8μm, it is assigned a value of 0.5. Real-time data is transmitted to the industrial control unit via Ethernet. The label adhesion coefficient is determined based on the label initial tack test result in the dynamic parameters. High-adhesion labels with an initial tack > 20N / 25mm are assigned a value of 1.0, and medium-adhesion labels (initial tack 5-20N / 25mm) are assigned a value of 0.5. The test data is pre-stored in the parameter mapping table of the data layer along with the label type information, and the corresponding coefficient is automatically matched when called.

[0043] In a specific embodiment, the labeling speed of each single-module inference SKU product is analyzed as follows: The labeling speed of each single-module inference SKU product needs to be matched with the production line speed and time window. The calculation formula is: Initial labeling speed of each single-module inference SKU product = Production line transmission speed ÷ Product real-time spacing × 1.2. At the same time, a size correction coefficient needs to be introduced: Corrected labeling speed = Initial labeling speed × (50 / Product label length) × (50mm / Product label width).

[0044] The label templates, labeling position coordinates, labeling pressure, and labeling speed corresponding to the SKU products in each single module are integrated into a dimensional structure to form a labeling execution parameter set containing specific values, constraints, and execution priorities.

[0045] Examples of embodiments of the present invention Figure 2 As shown, the intelligent automatic labeling system based on multi-variety mixed production includes: SKU product labeling analysis module: an automated labeling intelligent operation system is built in the target labeling factory, so that when each SKU product in the target production line is transmitted to the labeling equipment, the transmission posture and position of each SKU product are analyzed, and each SKU product that meets the labeling conditions is marked as the SKU product to be labeled.

[0046] Image acquisition and feature extraction module: When each SKU to be labeled is transmitted to the visual acquisition area, the algorithm inference mode corresponding to each SKU to be labeled is evaluated, and then the appearance feature parameters and dynamic parameters corresponding to each SKU inferred by each single module are obtained.

[0047] SKU Labeling Parameter Generation Module: Based on the appearance feature parameters and dynamic parameters of the SKU products inferred from each single module, this module analyzes the label template, labeling position coordinates, labeling pressure, and labeling speed of the SKU products inferred from each single module, and then generates the labeling execution parameter set for the SKU products inferred from each single module.

[0048] The above description is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined in this specification, they should all fall within the protection scope of the present invention.

Claims

1. A smart automatic labeling method and system based on multi-variety mixed production, characterized in that, include: Step 1: SKU Labeling Analysis: Build an automated labeling intelligent operation system in the target supermarket labeling factory. When each SKU product in the target production line is transferred to the labeling equipment, analyze the transfer posture and position of each SKU product, and mark each SKU product that meets the labeling conditions as the SKU product to be labeled. Step 2, Image Acquisition and Feature Extraction: When each SKU to be labeled is transmitted to the visual acquisition area, the algorithm inference mode corresponding to each SKU to be labeled is evaluated, and then the appearance feature parameters and dynamic parameters corresponding to each single module inference SKU are obtained. Step 3: Generation of SKU labeling parameters: Based on the appearance feature parameters and dynamic parameters of the SKU products inferred by each single module, the label template, labeling position coordinates, labeling pressure and labeling speed of the SKU products inferred by each single module are analyzed, and then the labeling execution parameter set of the SKU products inferred by each single module is generated.

2. The intelligent automatic labeling method based on multi-variety mixed production as described in claim 1, characterized in that, The specific construction process for building an automated labeling intelligent operation system in the target labeling factory is as follows: A three-layer architecture of "hardware layer - software layer - data layer" is constructed. The hardware layer deploys industrial cameras, laser contour sensors, labeling actuators, and industrial control units, equipped with Intel i7 processors, 16GB of memory, and computing power ≥ The NVIDIA graphics card is 8.6 GHz, enabling communication between devices via Ethernet; the software layer integrates a deep SKU recognition model library and a labeling parameter management module, and develops a remote operation and maintenance interface to support online model updates and function upgrades; The data layer constructs a SKU sample database, storing images and labeling data of various products, labeling parameter mapping tables, and production log databases. Real-time data interaction and local storage are achieved through an edge computing gateway, forming an automated labeling intelligent operation system with a closed loop of "perception-decision-execution-optimization".

3. The intelligent automatic labeling method based on multi-variety mixed production as described in claim 2, characterized in that, The analysis of the transmission attitude and location corresponding to each SKU is as follows: A1. When each SKU is conveyed to the labeling equipment entrance area via the conveyor belt, the laser displacement sensor at the equipment entrance scans the conveyor belt surface in real time. When a product is detected obstructing the laser signal, the vision pre-positioning device and the conveyor belt speed control motor are triggered to start in conjunction. The operating frequency of the speed control motor is dynamically adjusted according to the product spacing. The calculation formula is as follows: ,in, The operating frequency of the speed-regulating motor, This is the inherent coefficient of the equipment. As the production line reference speed, For real-time detection of product spacing; A2. Each SKU is fed into the positioning and detection area via a speed-adjustable conveyor belt. Two sets of laser contour sensors are installed at 300mm intervals along the conveying direction in this area. The first set of laser contour sensors collects the front contour coordinates of each SKU. The second set of laser contour sensors outputs the coordinates of the rear contour of the product. ; A3. Import the two sets of contour coordinate data for each SKU into the attitude analysis algorithm: Using the center line of the conveyor belt as the reference Y-axis, calculate the horizontal deviation between the center point of the front contour of each SKU and the reference axis: ,in, (referring to the X-axis as the reference axis), the horizontal deviation between the center point of the rear profile of each SKU and the reference axis. If a certain SKU product If the transmission posture of the SKU is qualified, it is determined that the transmission posture is qualified; otherwise, it is marked as unqualified. A4. Calculate the positional parameters of the product's transport direction using the two sets of contour coordinates for each SKU; calculate the actual length of the product using the distance formula between two points. The estimated time for the product to reach the labeling station is calculated by combining the conveyor belt speed. ,in, The distance from the positioning area to the labeling station is calculated; simultaneously, the longitudinal difference between the two sets of coordinates is calculated. If a certain SKU product If the position is correct, it is considered compliant; otherwise, it is marked as a position offset.

4. The intelligent automatic labeling method based on multi-variety mixed production as described in claim 3, characterized in that, Each SKU that meets the labeling conditions is recorded as a SKU to be labeled. The specific labeling process is as follows: If the posture and position of a certain SKU are correct, it is directly conveyed to the labeling preparation area via conveyor belt to wait for labeling; if the posture or position of a certain SKU is incorrect, the pneumatic push-alignment device on the side of the positioning area is triggered to push the product to rotate until the posture and position of the SKU are correct; and each SKU with correct posture and position is marked as a product to be labeled and proceeds to the next process.

5. The intelligent automatic labeling method based on multi-variety mixed production as described in claim 4, characterized in that, The algorithmic reasoning pattern for obtaining and evaluating each SKU to be labeled is specifically evaluated as follows: B1. Obtain the product feature data corresponding to each SKU to be labeled. The product feature data includes the complexity of the appearance outline, the proportion of surface reflective area, and the material hardness level. Compare the complexity of the appearance outline, the proportion of surface reflective area, and the material hardness level of each SKU to be labeled with the corresponding score table to obtain the score of the complexity of the appearance outline, the proportion of surface reflective area, and the material hardness level of each SKU to be labeled. B2. Then compare the scores for the appearance complexity, surface reflective area ratio, and material hardness level of each SKU to be labeled with the corresponding score thresholds. B3. If any one of the following scores for a certain SKU to be labeled—appearance complexity, surface reflective area ratio, and material hardness grade—is greater than the corresponding threshold, it indicates that the algorithm reasoning mode for that SKU is single-module reasoning. Conversely, if all three scores are less than or equal to the corresponding threshold, it indicates that the algorithm reasoning mode for that SKU is algorithm scheme reasoning. SKUs with a single-module reasoning algorithm reasoning mode are then recorded as single-module reasoning SKUs. B4. If the algorithm reasoning mode corresponding to a certain SKU to be labeled is algorithm scheme reasoning, then there is no need to adjust the labeling parameters of the SKU to be labeled. It is only necessary to complete the parameter preparation before labeling through the preset standardized labeling parameters. There is no intermediate parameter adjustment step in the whole process. If the algorithm reasoning mode corresponding to a certain SKU to be labeled is single module reasoning, then the labeling parameters of the SKU to be labeled need to be adjusted.

6. The intelligent automatic labeling method based on multi-variety mixed production as described in claim 5, characterized in that, The specific generation process for the tag templates corresponding to the SKU products in each individual module is as follows: C1. When a single-module inference SKU is a planar product, a high-resolution image of the product's labeling area is captured using a 20-megapixel industrial camera and a ring light source. The physical width and height of the labeling area are extracted. Based on size constraint rules, the maximum allowable size of the label is generated. The calculation formula is: the label width of the single-module inference SKU is ≤ the width of the product's labeling area × 80%, and the label height is ≤ the height of the product's labeling area × 80%. The 80% ratio is set to reserve 5% edge margin on each side to prevent the label from exceeding the labeling area boundary. C2. If a single-module inference SKU is a curved surface product, activate the laser contour sensor in the positioning detection area to scan the product surface contour and generate 3D point cloud data to obtain the radius of curvature of the single-module inference SKU. Based on the cylinder circumference formula, derive the label unfolding length limit. The calculation formula is: label unfolding length ≤ π × (radius of curvature of the single-module inference SKU × 2) × 70%. The 70% proportional constraint can prevent the label from wrinkling due to stretching or overlapping when it is attached to the curved surface, ensuring that the label is flat and attached. C3. Generate a label template for each inference SKU product, including size parameters, edge allowance, and layout baseline, based on the label width, height, or unfolded length calculated for each individual module inference SKU product.

7. The intelligent automatic labeling method based on multi-variety mixed production as described in claim 6, characterized in that, The analysis of the label location coordinates corresponding to the SKU products in each single module is as follows: The labeling position coordinates of each single-module inference SKU are calculated based on the relative offset of its bounding box. The calculation formula is: the actual labeling X coordinate of each single-module inference SKU = its bounding box reference X coordinate + ΔX, and the actual labeling Y coordinate of each single-module inference SKU = its bounding box reference Y coordinate + ΔY. ΔX and ΔY are obtained by converting the design drawing data of the single-module inference SKU with the dynamic parameters of visual acquisition: ΔX = (the design labeling X coordinate of the product - the physical coordinate of its left edge) - (the X coordinate of its bounding box center - the pixel coordinate of the left edge of the image) × pixel physical size. The pixel physical size is obtained through camera calibration. Similarly, ΔY is calculated in the same way. At the same time, for the attitude fluctuation of each single-module inference SKU during transmission, a rotation compensation formula is introduced: the corrected ΔX = ΔX × cosθ - ΔY × sinθ, and the corrected ΔY = ΔX × sinθ + ΔY × cosθ, where θ is the maximum rotation angle fluctuation value of the product acquired through dynamic parameters.

8. The intelligent automatic labeling method based on multi-variety mixed production as described in claim 7, characterized in that, The analysis of the labeling pressure corresponding to each single-module inference SKU product is as follows: The labeling pressure of each single-module inference SKU is calculated by combining appearance feature parameters and dynamic parameters. The calculation formula is: Labeling pressure of each single-module inference SKU = (its material hardness coefficient × 0.3) + (its roughness coefficient × 0.1) - (its label adhesion coefficient × 0.1).

9. The intelligent automatic labeling method based on multi-variety mixed production as described in claim 8, characterized in that, The analysis of the labeling speed corresponding to the SKU products in each single module is as follows: The labeling speed of each inference SKU in a single module must be matched with the production line speed and time window. The calculation formula is: Initial labeling speed of each inference SKU in a single module = Production line transmission speed ÷ Real-time product spacing × 1.

2. At the same time, a size correction factor needs to be introduced: Corrected labeling speed = Initial labeling speed × (50 / Product label length) × (50mm / Product label width). The label templates, labeling position coordinates, labeling pressure, and labeling speed corresponding to the SKU products in each single module are integrated into a dimensional structure to form a labeling execution parameter set containing specific values, constraints, and execution priorities.

10. An intelligent automatic labeling system based on multi-variety mixed production, implementing the intelligent automatic labeling method based on multi-variety mixed production as described in any one of claims 1-9, characterized in that, include: SKU Labeling Analysis Module: Builds an automated labeling intelligent operation system in the target labeling factory. When each SKU product in the target production line is transferred to the labeling equipment, the corresponding transfer posture and position of each SKU product are analyzed, and each SKU product that meets the labeling conditions is marked as the SKU product to be labeled. Image acquisition and feature extraction module: When each SKU to be labeled is transmitted to the visual acquisition area, the algorithm inference mode corresponding to each SKU to be labeled is evaluated, and then the appearance feature parameters and dynamic parameters corresponding to each SKU inferred by each single module are obtained. SKU Labeling Parameter Generation Module: Based on the appearance feature parameters and dynamic parameters of the SKU products inferred from each single module, this module analyzes the label template, labeling position coordinates, labeling pressure, and labeling speed of the SKU products inferred from each single module, and then generates the labeling execution parameter set for the SKU products inferred from each single module.

Images