Cigarette case surface printing quality evaluation method and system based on intaglio printing

Abstract

The invention discloses a cigarette case surface printing quality evaluation method and system based on intaglio printing, and relates to printing quality inspection. The method is applied to a surface quality detection system and comprises the following steps: acquiring a surface image of a to-be-evaluated cigarette case; performing feature extraction and texture residual analysis on the surface image of the to-be-evaluated cigarette case to obtain a deep semantic feature map, a physical attribute feature map and a texture residual difference map; based on the texture residual difference graph, fusing the deep semantic feature graph and the physical attribute feature graph to obtain a to-be-recognized surface defect graph; defect identification is carried out on the to-be-identified surface defect graph, a plurality of defect judgment results are obtained, and the defect judgment results comprise defect information and confidence coefficients; and the rendering sensitivity of the defect judgment result with low confidence is calculated, and whether the defect belongs to a real defect or not is determined according to the rendering sensitivity, so that the problems of missed judgment and misjudgment of the defect in cigarette case surface quality evaluation are solved.

Description

Technical Field

[0001] This application relates to the technical field of printing quality inspection, specifically to a method and system for evaluating the surface printing quality of cigarette boxes based on gravure printing. Background Technology

[0002] As the core packaging carrier of tobacco products, cigarette boxes serve multiple functions, including brand communication, product protection, and compliance display. The quality of their surface printing directly affects the product's market competitiveness and industry compliance. Gravure printing, with its advantages of strong ability to reproduce fine lines, thick and uniform ink layers, and good adaptability of anti-counterfeiting patterns, has become the mainstream process for fine printing of cigarette boxes. However, it also places higher demands on surface quality assessment.

[0003] Traditional surface quality assessment of gravure-printed cigarette boxes primarily involves acquiring printed images of the box surface and then using low-level feature algorithms such as edge detection and threshold segmentation to identify obvious defects. However, the patterns and textures in gravure printing are extremely fine, with very small differences between defective and normal features, resulting in low defect identification accuracy. To address this issue, the industry employs ultra-high-definition industrial cameras combined with integrated lighting systems to amplify the differences between defective and normal features in the images, thereby improving the accuracy of defect identification on cigarette box surfaces.

[0004] However, the above solutions require extremely high stability of the detection environment. Even slight changes in ambient light, or minor variations in the lighting system or camera angle, can cause minute defects in the printed image to become difficult to detect again, leading to misjudgment or missed detection by the algorithm. Summary of the Invention

[0005] To address the issues of missed and incorrect defect assessments in current cigarette box surface quality evaluation methods, this application provides a method and system for evaluating cigarette box surface printing quality based on gravure printing.

[0006] In a first aspect, this application provides a method for evaluating the surface printing quality of cigarette boxes based on gravure printing, applied to a surface quality inspection system, the method comprising: Obtain an image of the surface of the cigarette box to be evaluated; Feature extraction and texture residual analysis were performed on the surface image of the cigarette box to be evaluated to obtain a deep semantic feature map, a physical attribute feature map, and a texture residual difference map. Based on the texture residual difference map, the deep semantic feature map and the physical attribute feature map are fused to obtain the surface defect map to be identified; Defect identification is performed on the surface defect map to be identified, and multiple defect judgment results are obtained. The defect judgment results include defect information and confidence level. If the confidence level of the first defect judgment result is less than or equal to the preset confidence threshold, the defect information of the first defect judgment result is input into the physical rendering model to obtain the rendering sensitivity of the first defect judgment result. The first defect judgment result is any one of the multiple defect judgment results. If the rendering sensitivity of the first defect judgment result is greater than or equal to the preset rendering sensitivity, then the first defect judgment result is output as a false defect; if the rendering sensitivity of the first defect judgment result is less than the preset rendering sensitivity, then the first defect judgment result is output as a real defect.

[0007] Optionally, the step of performing feature extraction and texture residual analysis on the surface image of the cigarette box to be evaluated to obtain a deep semantic feature map, a physical attribute feature map, and a texture residual difference map specifically includes: The surface image of the cigarette box to be evaluated is input into a pre-trained semantic feature extraction network to obtain the deep semantic feature map; The surface image of the cigarette box to be evaluated is input into a pre-trained physical property decoupling network for feature decoupling to obtain the physical property feature map. The physical property decoupling network is a neural network model with an encoder-decoder architecture, which is used to decouple the underlying physical properties that are independent of illumination. The surface image of the cigarette box to be evaluated is input into a pre-trained texture autoencoder to obtain a texture reconstruction image; The pixel-level residual between the reconstructed texture image and the surface image of the cigarette box to be evaluated is calculated to obtain the texture residual difference map.

[0008] Optionally, the texture autoencoder is trained from defect-free cigarette box surface samples and non-printing defect cigarette box surface samples.

[0009] Optionally, the process of fusing the deep semantic feature map and the physical attribute feature map based on the texture residual difference map to obtain the surface defect map to be identified specifically involves: The texture residual difference map is normalized to obtain an initial spatial attention weight map; The initial spatial attention weight map is upsampled to obtain multiple spatial attention weight maps of different scales; The saliency of each of the spatial attention weight maps is evaluated to obtain a saliency score for each of the spatial attention weight maps. The target spatial attention weight map is determined based on the saliency scores of multiple spatial attention weight maps. Based on the target space attention weight map, the deep semantic feature map and the physical attribute feature map are enhanced respectively to obtain the deep semantic feature map enhancement map and the physical attribute feature map enhancement map; The deep semantic feature map enhancement map and the physical property feature map enhancement map are concatenated by channels to obtain the surface defect map to be identified.

[0010] Optionally, the significance assessment includes noise assessment and physical property consistency assessment.

[0011] Optionally, the step of performing defect identification on the surface defect map to obtain multiple defect judgment results is as follows: The surface defect image to be identified is input into the cigarette box surface defect identification model to obtain multiple defect information; The multiple defect information is input into the evidence depth model to obtain the evidence vector of the multiple defect information; Calculate the confidence level of each defect information based on the evidence vectors of the multiple defect information; Based on the multiple defect information and their respective confidence levels, multiple defect judgment results are obtained.

[0012] Optionally, the step of inputting the defect information of the first defect judgment result into the physical rendering model to obtain the rendering sensitivity of the first defect judgment result specifically involves: Based on the defect information of the first defect judgment result, extract the physical attribute feature region of the defect information of the first defect judgment result in the physical attribute feature map. Based on the physical attribute feature region of the first defect judgment result, configure the rendering target parameters in the physical rendering model; Based on the confidence level of the real scene lighting parameters and the first defect judgment result, the rendering parameters of the physical rendering model are configured, including the rendering trajectory and light intensity of the virtual light source; After the physical rendering model completes the rendering of the physical attribute feature region of the first defect judgment result, it reads the appearance feature value change curve of the physical attribute feature region of the first defect judgment result. The standard deviation of the appearance feature value change curve of the physical attribute feature region of the first defect judgment result is used as the rendering sensitivity of the first defect judgment result.

[0013] Secondly, this application provides a surface printing quality assessment system for cigarette boxes based on gravure printing. The system is a surface quality detection system, comprising an acquisition module, a processing module, and an output module, wherein: The acquisition module is used to acquire an image of the surface of the cigarette box to be evaluated; The processing module is used to perform feature extraction and texture residual analysis on the surface image of the cigarette box to be evaluated, to obtain a deep semantic feature map, a physical attribute feature map, and a texture residual difference map; based on the texture residual difference map, the deep semantic feature map and the physical attribute feature map are fused to obtain a surface defect map to be identified; the surface defect map to be identified is used to identify defects, to obtain multiple defect judgment results, the defect judgment results including defect information and confidence level; if the confidence level of the first defect judgment result is less than or equal to a preset confidence threshold, the defect information of the first defect judgment result is input into the physical rendering model to obtain the rendering sensitivity of the first defect judgment result, the first defect judgment result being any one of the multiple defect judgment results; The output module is configured to output that the first defect judgment result is a false defect if the rendering sensitivity of the first defect judgment result is greater than or equal to a preset rendering sensitivity, and to output that the first defect judgment result is a real defect if the rendering sensitivity of the first defect judgment result is less than the preset rendering sensitivity.

[0014] Thirdly, this application provides an electronic device including a processor, a memory, a user interface, and a network interface. The memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method as described in any one of the first aspects.

[0015] Fourthly, this application provides a computer-readable storage medium storing instructions that, when executed, perform the method described in any one of the first aspects.

[0016] In summary, one or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: This application first extracts the deep semantic feature map, physical attribute feature map, and texture residual difference map from the surface image of the cigarette box to be evaluated. The deep semantic feature map is a feature map containing semantic information of the surface image of the cigarette box to be evaluated. The physical attribute feature map is a feature map containing low-level physical attributes (normals, roughness, etc.) that are independent of illumination. The texture residual difference map reflects the degree of texture change in each region of the surface image of the cigarette box to be evaluated. Because the gravure printing pattern is fine, areas with drastic texture changes often correspond to potential defects. Highlighting these areas can reduce the possibility of minor defects being overlooked. Therefore, the texture residual difference map is used to enhance both the deep semantic feature map and the physical attribute feature map at the same time, thereby highlighting the areas with drastic changes in the texture residual difference map in the deep semantic feature map and the physical attribute feature map. Then, the enhanced deep semantic feature map and the physical attribute feature map are concatenated to restore a surface defect map that is independent of illumination and can be used for defect identification. At this point, the surface defect map to be identified has eliminated illumination interference and highlighted the areas with drastic changes, thereby solving the problem of missed defect detection in the surface image of the cigarette box to be evaluated. Then, because changes in the detection environment can make some subtle defect features very similar to normal features, for multiple detected defects, the confidence levels of multiple defects are calculated. For defects with low confidence levels, this application inputs them into a physically based rendering model for rendering simulation to obtain the rendering sensitivity of the defect. The physically based rendering model is a simulation model that simulates the surface changes of an object under different lighting conditions. Therefore, when the rendering sensitivity of a defect is high, it indicates that the appearance of the defect changes drastically with lighting, conforming to optical artifacts and belonging to false defects. When the rendering sensitivity of a defect is low, it indicates that the appearance of the defect does not change with lighting, representing a physical and essential flaw and belonging to real defects. This solves the problem of misjudging defects in the surface image of the cigarette box to be evaluated. Thus, it solves the problem of difficulty in capturing subtle defects in the printed images of the cigarette box surface caused by gravure printing, thereby improving the accuracy of cigarette box surface printing quality evaluation. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating a method for evaluating the surface printing quality of cigarette boxes based on gravure printing, as provided in an embodiment of this application.

[0018] Figure 2 This is a schematic diagram of a cigarette box surface printing quality assessment system based on gravure printing, provided in an embodiment of this application.

[0019] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0020] Explanation of reference numerals in the attached drawings: 1. Acquisition module; 2. Processing module; 3. Output module; 300. Electronic device; 301. Processor; 302. Communication bus; 303. User interface; 304. Network interface; 305. Memory. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0022] Currently, there are various printing methods for cigarette box surface patterns, such as gravure printing, offset printing, and screen printing. However, with the continuous upgrading of counterfeit cigarette technology and the intensification of market competition, the requirements for the printing quality of cigarette box surface patterns are also increasing. Therefore, gravure printing, due to its advantages such as strong ability to reproduce fine lines, thick and uniform ink layer, and good adaptability of anti-counterfeiting patterns, has gradually become the mainstream printing method for cigarette box surface patterns. However, with the increasing sophistication of cigarette box surface patterns, higher requirements are also being placed on the inspection of cigarette box surface quality.

[0023] Currently, to highlight subtle defects in printed images on cigarette packs, the industry uses ultra-high-definition industrial cameras combined with lighting systems to amplify the differences between defective and normal features in the image, thereby improving the accuracy of defect identification. However, this method requires extremely high stability of the detection environment. Even slight changes in ambient light, or minor adjustments to the lighting system or camera angle, can cause subtle defects in the printed image to become difficult to capture again, leading to misjudgment or missed detection by the algorithm.

[0024] Therefore, to address this problem, this application provides a method for evaluating the surface printing quality of cigarette boxes based on gravure printing. This method is applied to surface quality inspection systems, such as... Figure 1 As shown, the method includes steps S101 to S106, which are as follows: S101. Obtain an image of the surface of the cigarette box to be evaluated.

[0025] In the above steps, firstly, a suitable lighting environment and industrial camera shooting angle are set up on the production line according to the printed pattern on the cigarette box surface. Then, as the production line runs, the printed image on the cigarette box surface is captured in real time. The printed image on the cigarette box surface is then preprocessed with distortion correction, ROI area alignment, and brightness normalization to obtain a standardized image of the cigarette box surface to be evaluated.

[0026] S102. Perform feature extraction and texture residual analysis on the surface image of the cigarette box to be evaluated to obtain deep semantic feature map, physical attribute feature map and texture residual difference map.

[0027] In the above steps, if defect identification is directly performed on the surface image of the cigarette box to be evaluated, some defects whose features are not obvious due to changes in the detection environment will be difficult to identify, leading to missed or false judgments. Therefore, this application first inputs the surface image of the cigarette box to be evaluated into a pre-trained semantic feature extraction network to obtain a deep semantic feature map. The deep semantic feature map carries rich category semantic information and global context information, distinguishing the complex texture of the cigarette box surface from a semantic level, thereby reducing the influence of the detection environment on the low-level attributes of the image (RGB values, brightness values, etc.). The semantic feature extraction network is a model based on a deep convolutional neural network architecture, used to extract feature representations with high-level semantic information from the input cigarette box surface image, such as special characters used to distinguish logos, text, backgrounds, etc. The semantic feature extraction network is trained from a large number of pre-annotated semantic feature samples of cigarette box surface images. The preferred network architecture in this application is the YOLOv10 model feature extraction network structure.

[0028] Then, the surface image of the cigarette box to be evaluated is input into a pre-trained physical property decoupling network for feature decoupling, resulting in a physical property feature map. In the physical property feature map, each pixel is decoupled from the underlying physical properties independent of illumination, such as normals and roughness. Since these underlying physical properties reflect the texture changes of the printed image from the perspective of the real physical structure, they can reduce the influence of illumination on the image to a certain extent. The physical property decoupling network is a neural network model based on an encoder-decoder architecture, consisting of a shared encoder and multiple parallel decoders. It is trained through self-supervised learning using qualified cigarette box surface image samples under multiple illumination angles, and is used to decouple multiple underlying physical properties independent of illumination from a single RGB image.

[0029] Finally, the image of the cigarette box surface to be evaluated is input into a pre-trained texture autoencoder to obtain a texture reconstruction image. Specifically, during the texture reconstruction process, each pixel is reconstructed into a standard texture conforming to normal gravure printing processes, such as uniform dot distribution and smooth gradient transitions. Since these standard textures are statistical prior knowledge learned from defect-free samples, they can amplify the difference between abnormal textures and normal backgrounds to a certain extent. The texture autoencoder is also a neural network model based on an encoder-decoder architecture, which is trained unsupervised from defect-free cigarette box surface samples to reconstruct an ideal texture map conforming to normal statistical laws from a single RGB image. Then, the pixel-level residual between the texture reconstruction image and the cigarette box surface image to be evaluated is calculated to obtain a texture residual difference map. The texture residual difference map reflects the degree of change in image texture of the cigarette box surface image to be evaluated compared to the qualified standard image. The residual calculation method is a conventional method for those skilled in the art and will not be elaborated on here.

[0030] In one possible implementation, the surface of the cigarette box contains not only printing defects but also non-printing defects, such as dust, fibers, and other adhering substances. These non-printing defects are not actually surface defects of the cigarette box, but the texture autoencoder will treat them as anomalies that cannot be reconstructed, thus generating a strong response in the residual map and leading to misjudgment. Therefore, this application also introduces non-printing defect cigarette box surface samples for training during the training process, allowing the model to encounter and learn these common non-printing defects during the training phase, guiding it to focus on reconstructing the inherent texture patterns of gravure printing, so that the final generated texture residual difference map can more accurately highlight the real defects caused by the printing process itself.

[0031] S103. Based on the texture residual difference map, the deep semantic feature map and the physical attribute feature map are fused to obtain the surface defect map to be identified.

[0032] In the above steps, because the gravure printing pattern is intricate, areas with drastic texture changes often correspond to potential defects. Highlighting these areas can reduce the chances of subtle defects being overlooked. Therefore, the texture residual difference map is used to simultaneously enhance both the deep semantic feature map and the physical attribute feature map, thereby highlighting the areas with drastic changes in the texture residual difference map in both the deep semantic feature map and the physical attribute feature map. Specifically: First, the texture residual difference map is normalized to obtain an initial spatial attention weight map. This initial spatial attention weight map contains weight values ​​for multiple pixels; a larger weight value indicates a more drastic texture change at that pixel, thus requiring special attention in subsequent image enhancement. Then, the initial spatial attention weight map is upsampled to further uncover more subtle texture changes, resulting in multiple spatial attention weight maps at different scales (resolutions). Next, the prominence of defective and normal regions in each spatial attention weight map at different scales is evaluated, specifically including noise assessment and physical property consistency assessment. Noise assessment determines whether defective and normal textures are sufficiently prominent in the image, while physical property consistency assessment determines whether defective textures are sufficiently prominent within normal textures. This results in a spatial attention weight map with low noise impact and sufficiently prominent defective textures. Specifically: For noise assessment, this application calculates the information entropy of each spatial attention weight map. The larger the information entropy, the more chaotic the texture, which in turn indicates higher noise interference. At this time, the information entropy of each spatial attention weight map can be normalized to obtain the noise assessment score of each spatial attention weight map. For the physical property consistency assessment, this application calculates the correlation coefficient between the gradient maps of each spatial attention weight map and the physical property feature map. The greater the correlation, the more the texture change area and the physical deformation area overlap, which indicates that the anomaly is very likely caused by a real physical defect. At this time, the calculated correlation coefficient can be normalized to obtain the physical property consistency assessment score of each spatial attention weight map.

[0033] It should be noted that the above-mentioned methods for calculating information entropy and correlation coefficient are conventional techniques used by those skilled in the art. For example, information entropy can be calculated using the principle of Shannon entropy, and correlation coefficient can be calculated using the principle of Pearson correlation coefficient. Therefore, the specific calculation process will not be elaborated on here.

[0034] Then, the noise evaluation score and physical attribute consistency evaluation score of each spatial attention weight map are weighted and summed to obtain the saliency score of each spatial attention weight map. For spatial attention weight maps with high saliency scores, they can better highlight the defect texture. Therefore, the spatial attention weight map with the highest saliency score among multiple spatial attention weight maps is selected as the target spatial attention weight map. Then, the spatial size of the target spatial attention weight map is unified with the spatial size of the deep semantic feature map and the spatial size of the physical attribute feature map. At this time, the target spatial attention weight map after spatial size unification is multiplied element-wise with the deep semantic feature map and the physical attribute feature map respectively to obtain the deep semantic feature map enhancement map and the physical attribute feature map enhancement map.

[0035] Finally, the deep semantic feature map enhancement map and the physical attribute feature map enhancement map are concatenated to restore a surface defect map to be identified that is independent of illumination and has prominent defect features.

[0036] S104. Perform defect identification on the surface defect map to be identified, and obtain multiple defect judgment results. The defect judgment results include defect information and confidence level.

[0037] In the above steps, the surface defect image to be identified is input into the cigarette box surface defect recognition model to obtain multiple defect information in the surface defect image, including the type and location of the defect. Since the surface defect image to be identified has undergone defect magnification processing, the detection accuracy of the cigarette box surface defect recognition model is greatly improved, thereby reducing the number of missed detections. The cigarette box surface defect recognition model is a neural network model trained from cigarette box surface defect samples.

[0038] In actual working conditions, influenced by the design of the cigarette box surface pattern, some printing defects can closely resemble normal textures. For example, the cigarette box surface pattern often features color gradient areas. Gravure printing itself uses cell transfer of ink to form intricate patterns. If some cells become slightly clogged, the ink transfer will be slightly insufficient, resulting in a slight lightening of the color in the color area, creating a color difference. Since the color area itself is a color gradient area, this color difference will be highly similar to the color gradient area. Therefore, although the cigarette box surface defect recognition model can detect a potential defect, it may be mistakenly classified as normal texture due to its high confidence level. To address this, this application performs a second defect verification for defect information with low confidence levels to reduce the possibility of misjudgment. Specifically, multiple defect information points are input into an evidence depth model to obtain evidence vectors for each defect. The evidence depth model is a deep learning model incorporating DS evidence theory, capable of quantifying and outputting uncertainty. In this application, it is used to detect multiple pieces of evidence supporting the existence of a defect, as well as the prior probabilities of these pieces of evidence. For example, the existence of a certain piece of evidence gives a 20% probability that the defect is true. The vector formed by the prior probabilities of multiple pieces of evidence is the evidence vector. Then, the evidence vectors for each defect are constructed using Dirichlet distributions, thereby elevating point estimation to distribution estimation. Finally, based on the Dirichlet distributions of the multiple evidence vectors, the defect uncertainty of the multiple defect information points is calculated. This uncertainty is then normalized, and the complement of the normalization result is calculated to obtain the confidence level of the multiple defect information points. Defect uncertainty can be understood as the degree of uncertainty of the defect information. The formula for calculating defect uncertainty is designed as follows:

[0039] Where S represents the defect uncertainty. Let be the prior probability of the k-th piece of evidence in the evidence vector, n be the total number of pieces of evidence in the evidence vector, and N be the number of types of evidence in the evidence vector.

[0040] In the above formula, when a certain piece of evidence has a high proportion in the Dirichlet distribution and a high prior probability, it indicates that the defect is more likely to be a real defect, and its defect uncertainty is smaller.

[0041] Then, the confidence levels of multiple defect information are compared with preset confidence thresholds. If the confidence level of defect information is greater than the preset confidence threshold, the defect information is directly output.

[0042] S105. If the confidence level of the first defect judgment result is less than or equal to the preset confidence threshold, the defect information of the first defect judgment result is input into the physical rendering model to obtain the rendering sensitivity of the first defect judgment result. The first defect judgment result is any one of multiple defect judgment results.

[0043] S106. If the rendering sensitivity of the first defect judgment result is greater than or equal to the preset rendering sensitivity, the first defect judgment result is output as a false defect. If the rendering sensitivity of the first defect judgment result is less than the preset rendering sensitivity, the first defect judgment result is output as a real defect.

[0044] In steps S105 to S106 above, defect information with a confidence level less than or equal to a preset confidence threshold does not necessarily mean it is not a defect. This application proposes calculating the rendering sensitivity of this defect information. Rendering sensitivity can be understood as the degree to which the defect is sensitive to changes in the lighting environment. A higher sensitivity indicates that the appearance of the defect changes more drastically with lighting, consistent with optical artifacts, and is more likely to be a false defect. A lower sensitivity indicates that the appearance of the defect does not change with lighting, and is a physical, inherent flaw, more likely to be a real defect. Specifically: First, based on the location in the defect information, the corresponding physical attribute feature region in the physical attribute feature map is located. Then, based on the physical attribute feature region, the relevant structural parameters of the rendering target are constructed in the physical rendering model. Next, based on the real scene lighting parameters, the illumination intensity of the virtual light source in the rendering parameters is configured. Then, based on the confidence level of the defect information, the rendering trajectory of the virtual light source in the rendering parameters is configured. In this application, the rendering trajectory consists of multiple Fibonacci spiral sampling points evenly distributed on a hemisphere centered on the rendering target. Specifically, the confidence level of the defect information is matched with the rendering trajectory configuration table to obtain the number of Fibonacci spiral sampling points corresponding to the confidence level of each defect information. These points are then evenly distributed on a preset hemisphere centered on the rendering target. The lower the confidence level, the more sampling points there are. This provides more refined data support for the rendering sensitivity calculation of low-confidence defect information, thereby improving the accuracy of the rendering sensitivity of each low-confidence defect information.

[0045] After the physical rendering model completes the rendering of the target according to the rendering parameters, it reads the average grayscale value of the target surface during each rendering and uses it as the appearance feature value to obtain the overall brightness of the target under different lighting conditions. Then, it calculates the difference between the appearance feature values ​​in the rendering order to generate the appearance feature value change curve. At this time, it calculates the standard deviation of the appearance feature value change curve and uses it as the rendering sensitivity of defect information.

[0046] If the rendering sensitivity of the defect information is greater than or equal to the preset rendering sensitivity, the defect is output as a false defect. If the rendering sensitivity of the defect information is less than the preset rendering sensitivity, the defect information is output as a real defect. This effectively solves the problem of misjudging defects on the surface of cigarette boxes.

[0047] Reference Figure 2 This application also provides a surface printing quality assessment system for cigarette boxes based on gravure printing. The system is a surface quality detection system, comprising an acquisition module 1, a processing module 2, and an output module 3, wherein: The acquisition module 1 is used to acquire an image of the surface of the cigarette box to be evaluated; The processing module 2 is used to perform feature extraction and texture residual analysis on the surface image of the cigarette box to be evaluated, to obtain a deep semantic feature map, a physical attribute feature map, and a texture residual difference map; based on the texture residual difference map, the deep semantic feature map and the physical attribute feature map are fused to obtain a surface defect map to be identified; the surface defect map to be identified is used to identify defects, to obtain multiple defect judgment results, the defect judgment results including defect information and confidence level; if the confidence level of the first defect judgment result is less than or equal to a preset confidence threshold, the defect information of the first defect judgment result is input into the physical rendering model to obtain the rendering sensitivity of the first defect judgment result, the first defect judgment result being any one of the multiple defect judgment results; The output module 3 is configured to output the first defect judgment result as a false defect if the rendering sensitivity of the first defect judgment result is greater than or equal to a preset rendering sensitivity, and to output the first defect judgment result as a real defect if the rendering sensitivity of the first defect judgment result is less than the preset rendering sensitivity.

[0048] It should be noted that the above embodiments of the apparatus are only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0049] This application also discloses an electronic device. (See reference...) Figure 3 , Figure 3 This is a schematic diagram of the structure of an electronic device disclosed in an embodiment of this application. The electronic device 300 may include: at least one processor 301, at least one network interface 304, a user interface 303, a memory 305, and at least one communication bus 302.

[0050] The communication bus 302 is used to enable communication between these components.

[0051] The user interface 303 may include a display screen and a camera. Optionally, the user interface 303 may also include a standard wired interface and a wireless interface.

[0052] The network interface 304 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0053] The processor 301 may include one or more processing cores. The processor 301 connects to various parts of the server using various interfaces and lines, and performs various server functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in memory 305, and by calling data stored in memory 305. Optionally, the processor 301 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 301 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display; and the modem handles wireless communication. It is understood that the modem may also not be integrated into the processor 301 and may be implemented as a separate chip.

[0054] The memory 305 may include random access memory (RAM) or read-only memory. Optionally, the memory 305 may include a non-transitory computer-readable storage medium. The memory 305 may be used to store instructions, programs, code, code sets, or instruction sets. The memory 305 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 305 may also be at least one storage device located remotely from the aforementioned processor 301. (Refer to...) Figure 3 The memory 305, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and an application program for evaluating the surface printing quality of cigarette boxes based on gravure printing.

[0055] exist Figure 3In the illustrated electronic device 300, the user interface 303 is mainly used to provide an input interface for the user and acquire user input data; while the processor 301 can be used to call an application program stored in the memory 305 for evaluating the surface printing quality of cigarette boxes based on gravure printing. When executed by one or more processors 301, the electronic device 300 performs one or more of the methods described in the above embodiments. It should be noted that, for the foregoing method embodiments, for the sake of simplicity, they are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, because according to this application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also understand that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0056] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0057] In the various embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some service interface; the indirect coupling or communication connection between apparatuses or units may be electrical or other forms.

[0058] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0059] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0060] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, portable hard drives, magnetic disks, or optical disks.

[0061] The above description is merely an exemplary embodiment of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Other embodiments of this disclosure will be readily apparent to those skilled in the art upon consideration of the specification and the disclosure of practical truths.

[0062] This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described in this disclosure. The specification and embodiments are to be considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

Claims

1. A method for evaluating the surface printing quality of cigarette boxes based on gravure printing, characterized in that, The method, applied in a surface quality inspection system, includes: Obtain an image of the surface of the cigarette box to be evaluated; Feature extraction and texture residual analysis were performed on the surface image of the cigarette box to be evaluated to obtain a deep semantic feature map, a physical attribute feature map, and a texture residual difference map. Based on the texture residual difference map, the deep semantic feature map and the physical attribute feature map are fused to obtain the surface defect map to be identified; Defect identification is performed on the surface defect map to be identified, and multiple defect judgment results are obtained. The defect judgment results include defect information and confidence level. If the confidence level of the first defect judgment result is less than or equal to the preset confidence threshold, the defect information of the first defect judgment result is input into the physical rendering model to obtain the rendering sensitivity of the first defect judgment result. The first defect judgment result is any one of the multiple defect judgment results. If the rendering sensitivity of the first defect judgment result is greater than or equal to the preset rendering sensitivity, then the first defect judgment result is output as a false defect; if the rendering sensitivity of the first defect judgment result is less than the preset rendering sensitivity, then the first defect judgment result is output as a real defect.

2. The method according to claim 1, characterized in that, The process of extracting features and performing texture residual analysis on the surface image of the cigarette box to be evaluated, resulting in a deep semantic feature map, a physical attribute feature map, and a texture residual difference map, specifically includes: The surface image of the cigarette box to be evaluated is input into a pre-trained semantic feature extraction network to obtain the deep semantic feature map; The surface image of the cigarette box to be evaluated is input into a pre-trained physical property decoupling network for feature decoupling to obtain the physical property feature map. The physical property decoupling network is a neural network model with an encoder-decoder architecture, which is used to decouple the underlying physical properties that are independent of illumination. The surface image of the cigarette box to be evaluated is input into a pre-trained texture autoencoder to obtain a texture reconstruction image; The pixel-level residual between the reconstructed texture image and the surface image of the cigarette box to be evaluated is calculated to obtain the texture residual difference map.

3. The method according to claim 2, characterized in that, The texture autoencoder is trained from surface samples of defect-free cigarette boxes and surface samples of cigarette boxes without printing defects.

4. The method according to claim 1, characterized in that, The process of fusing the deep semantic feature map and the physical attribute feature map based on the texture residual difference map to obtain the surface defect map to be identified is as follows: The texture residual difference map is normalized to obtain an initial spatial attention weight map; The initial spatial attention weight map is upsampled to obtain multiple spatial attention weight maps of different scales; The saliency of each of the spatial attention weight maps is evaluated to obtain a saliency score for each of the spatial attention weight maps. The target spatial attention weight map is determined based on the saliency scores of multiple spatial attention weight maps. Based on the target space attention weight map, the deep semantic feature map and the physical attribute feature map are enhanced respectively to obtain the deep semantic feature map enhancement map and the physical attribute feature map enhancement map; The deep semantic feature map enhancement map and the physical property feature map enhancement map are concatenated by channels to obtain the surface defect map to be identified.

5. The method according to claim 4, characterized in that, The significance assessment includes noise assessment and physical property consistency assessment.

6. The method according to claim 1, characterized in that, The defect identification process for the surface defect map to be identified yields multiple defect judgment results, specifically as follows: The surface defect image to be identified is input into the cigarette box surface defect identification model to obtain multiple defect information; The multiple defect information is input into the evidence depth model to obtain the evidence vector of the multiple defect information; Calculate the confidence level of each defect information based on the evidence vectors of the multiple defect information; Based on the multiple defect information and their respective confidence levels, multiple defect judgment results are obtained.

7. The method according to claim 1, characterized in that, The step of inputting the defect information of the first defect judgment result into the physical rendering model to obtain the rendering sensitivity of the first defect judgment result is as follows: Based on the defect information of the first defect judgment result, extract the physical attribute feature region of the defect information of the first defect judgment result in the physical attribute feature map. Based on the physical attribute feature region of the first defect judgment result, configure the rendering target parameters in the physical rendering model; Based on the confidence level of the real scene lighting parameters and the first defect judgment result, the rendering parameters of the physical rendering model are configured, including the rendering trajectory and light intensity of the virtual light source; After the physical rendering model completes the rendering of the physical attribute feature region of the first defect judgment result, it reads the appearance feature value change curve of the physical attribute feature region of the first defect judgment result. The standard deviation of the appearance feature value change curve of the physical attribute feature region of the first defect judgment result is used as the rendering sensitivity of the first defect judgment result.

8. A system for evaluating the surface printing quality of cigarette boxes based on gravure printing, characterized in that, The system is a surface quality detection system, which includes an acquisition module (1), a processing module (2), and an output module (3), wherein: The acquisition module (1) is used to acquire an image of the surface of the cigarette box to be evaluated; The processing module (2) is used to perform feature extraction and texture residual analysis on the surface image of the cigarette box to be evaluated, and obtain a deep semantic feature map, a physical attribute feature map and a texture residual difference map; based on the texture residual difference map, the deep semantic feature map and the physical attribute feature map are fused to obtain a surface defect map to be identified; the surface defect map to be identified is used to identify defects, and multiple defect judgment results are obtained. The defect judgment results include defect information and confidence level; if the confidence level of the first defect judgment result is less than or equal to a preset confidence threshold, the defect information of the first defect judgment result is input into the physical rendering model to obtain the rendering sensitivity of the first defect judgment result. The first defect judgment result is any one of the multiple defect judgment results. The output module (3) is used to output the first defect judgment result as a false defect if the rendering sensitivity of the first defect judgment result is greater than or equal to the preset rendering sensitivity, and to output the first defect judgment result as a real defect if the rendering sensitivity of the first defect judgment result is less than the preset rendering sensitivity.

9. An electronic device, characterized in that, The device includes a processor (301), a memory (305), a user interface (303), and a network interface (304). The memory (305) is used to store instructions. The user interface (303) and the network interface (304) are used to communicate with other devices. The processor (301) is used to execute the instructions stored in the memory (305) to cause the electronic device (300) to perform the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed, perform the method as described in any one of claims 1 to 7.

Images