Method and system for visually detecting surface defects of plastic product

Abstract

The invention discloses a method and a system for visually detecting surface defects of a plastic product, and relates to the technical field of defect detection. The method comprises the following steps: performing function identification according to design information to obtain a shell function partition; determining defect identification requirements and position distribution according to functional characteristics of the shell function partitions; collecting image information of the plastic shell product; according to the corresponding relation between defect identification requirements and position distribution, partition defect identification parameter positioning is conducted on the image information, partition identification parameters are obtained, and when multiple partition identification parameters exist, an identification parameter weight exists; configuring a defect identification module according to the partition identification parameters; and carrying out partition defect identification on the corresponding image information by utilizing a defect identification module to obtain a defect identification result. The technical problem that in the prior art, surface defect detection efficiency and recognition precision are low is solved, defect detection parameters are optimized through function partition features, and the technical effect of improving detection efficiency and recognition precision is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of defect detection, and particularly to a method and system for visually detecting surface defects of plastic products. Background Art

[0002] As an important part of household appliances, the surface quality of the outer shell of household appliances directly affects the appearance of the product and the user experience. However, the outer shell of household appliances is usually made of plastic materials and processed by processes such as injection molding. Due to the complex process and various material characteristics involved, defects such as bubbles, scratches, depressions, and deformations are likely to appear on the surface of the outer shell. These defects not only affect the aesthetics of the product but may also lead to a decline in the functionality of the outer shell, such as insufficient fixing strength or reduced sealing performance. Existing defect detection technologies usually rely on traditional visual detection systems to uniformly detect by collecting the overall image of the outer shell surface. However, due to the different design characteristics and defect proneness of different functional areas of the household appliance outer shell (such as the operation panel, joint parts, corners, etc.), the unified detection method cannot be optimized according to the characteristics of each area, resulting in insufficient efficiency and accuracy of defect identification. Summary of the Invention

[0003] This application provides a method and system for visually detecting surface defects of plastic products, which solves the technical problem of low efficiency and recognition accuracy in surface defect detection in the prior art.

[0004] In view of the above problems, this application provides a method and system for visually detecting surface defects of plastic products.

[0005] In the first aspect of this application, a method for visually detecting surface defects of plastic products is provided. The method includes: Obtain the design information of the household appliance outer shell, perform function recognition according to the design information to obtain the outer shell function partition; determine the defect recognition requirements and position distribution according to the function characteristics of the outer shell function partition; based on the position distribution, collect the image information of the plastic outer shell product, and the image information has a position distribution identifier; perform partition defect recognition parameter positioning on the image information according to the corresponding relationship between the defect recognition requirements and the position distribution to obtain partition recognition parameters, and when there are multiple partition recognition parameters, there are recognition parameter weights; configure a defect recognition module according to the partition recognition parameters, and the defect recognition module includes an integrated learning model that matches the partition recognition parameters and the recognition parameter weights; use the defect recognition module to perform partition defect recognition on the corresponding image information to obtain a defect recognition result.

[0006] In the second aspect of this application, a system for visually detecting surface defects of plastic products is provided. The system includes: Function recognition component: Obtain the design information of the household appliance shell, perform function recognition according to the design information, and obtain the shell function partition; Defect recognition requirement determination component: Determine the defect recognition requirements and location distribution according to the function characteristics of the shell function partition; Image acquisition component: Based on the location distribution, collect the image information of the plastic shell product, and the image information has a location distribution identifier; Parameter positioning component: According to the corresponding relationship between the defect recognition requirements and the location distribution, perform partition defect recognition parameter positioning on the image information to obtain partition recognition parameters, and when there are multiple partition recognition parameters, there are recognition parameter weights; Module configuration component: Configure the defect recognition module according to the partition recognition parameters, and the defect recognition module includes an integrated learning model that matches the partition recognition parameters and the recognition parameter weights; Defect recognition execution component: Use the defect recognition module to perform partition defect recognition on the corresponding image information to obtain the defect recognition result.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: First, obtain the design information of the household appliance shell, perform function recognition according to the design information, and obtain the shell function partition. Next, determine the defect recognition requirements and location distribution according to the function characteristics of the shell function partition. Further, based on the location distribution, collect the image information of the plastic shell product, and the image information has a location distribution identifier. Then, according to the corresponding relationship between the defect recognition requirements and the location distribution, perform partition defect recognition parameter positioning on the image information to obtain partition recognition parameters, and when there are multiple partition recognition parameters, there are recognition parameter weights. Next, configure the defect recognition module according to the partition recognition parameters, and the defect recognition module includes an integrated learning model that matches the partition recognition parameters and the recognition parameter weights. Finally, use the defect recognition module to perform partition defect recognition on the corresponding image information to obtain the defect recognition result. Solve the technical problem of low surface defect detection efficiency and recognition accuracy in the prior art, and optimize the defect detection parameters through the function partition characteristics, achieving the technical effect of improving the visual detection efficiency and recognition accuracy. Description of the Drawings

[0008] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0009] Figure 1 It is a schematic flowchart of a method for visually detecting surface defects of plastic products provided by an embodiment of this application; Figure 2Schematic structural diagram of a system for visually detecting surface defects of plastic products provided by an embodiment of the present application.

[0010] Explanation of reference numerals: Function recognition component 11, Defect recognition requirement determination component 12, Image acquisition component 13, Parameter positioning component 14, Module configuration component 15, Defect recognition execution component 16. Detailed implementation manners

[0011] By providing a method and a system for visually detecting surface defects of plastic products, the present application solves the technical problems of low detection efficiency and low recognition accuracy of surface defects in the prior art.

[0012] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0013] It should be noted that the terms "including" and "having" are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server including a series of steps or units is not necessarily limited to those clearly listed steps or units, but may include other steps or modules that are not clearly listed or are inherent to these processes, methods, products, or devices.

[0014] Embodiment 1, as Figure 1 shown, the present application provides a method for visually detecting surface defects of plastic products, wherein the method includes: Obtain the design information of the household appliance shell, and perform function recognition according to the design information to obtain the shell function partition.

[0015] By reading the design information of the household appliance shell (such as CAD model, 3D drawing or specification document), analyze the functional structure characteristics of each part of the shell. Specifically, by analyzing the shape, size and geometric structure of the shell, determine the uses of different components. For example, the flat area may be the display panel or button area, and the protruding part may be the decorative part or fixed structure; use the material characteristics recorded in the design information to distinguish the uses of different functional areas. For example, the flexible plastic part may be the sealing area, while the hard plastic part may be the structural support area; according to the geometric characteristics, functional requirements and structural layout of the shell, divide the shell into several functional partitions, such as the decoration area (the frame or painted part of the shell), the operation area (buttons, touch screens), the structural support area (used to connect or support other components), etc.

[0016] Determine the defect identification requirements and location distribution based on the functional characteristics of the housing functional partitions.

[0017] Combined with the uses of each functional partition and the possible defect types, determine the defect identification requirements for each partition and the location distribution where these defects may occur. For example, the operation area is prone to defects such as scratches, dents, and uneven coatings; the decoration area is prone to defects such as color difference, uneven brightness, and coating peeling; the structural support area is prone to defects such as cracks, deformation, and injection molding residues; the edge area is prone to defects such as burrs and irregular warping. Combine the spatial layout of the partitions and the overall design drawing of the housing to determine the specific location of each partition on the housing.

[0018] Furthermore, determine the defect identification requirements and location distribution based on the functional characteristics of the housing functional partitions, including: Conduct partition function analysis on the housing functional partitions to determine the functional characteristics, which include decoration characteristics, structural strength characteristics, dimensional accuracy characteristics, assembly area characteristics, and heat dissipation area characteristics; based on the functional characteristics, conduct identification and analysis of core functional characteristics and auxiliary functional characteristics; mark the location distribution of the functional characteristics according to the identified locations of the core functional characteristics and auxiliary functional characteristics; based on the functional characteristics and functional coordination relationship of the core functional characteristics and auxiliary functional characteristics, determine the defect identification requirements and establish the corresponding relationship between the defect identification requirements and the location distribution of the functional characteristics.

[0019] Conduct partition function analysis on the housing functional partitions to determine the functional characteristics, including decoration characteristics, structural strength characteristics, dimensional accuracy characteristics, assembly area characteristics, and heat dissipation area characteristics. Specifically, the decoration area: directly exposed to the field of vision, with aesthetic requirements; the structural strength area: provides support and strength, affecting the overall stability of the product; the dimensional accuracy area: key interaction parts, such as buttons, openings, etc.; the assembly area: the edges or interfaces for assembling or connecting other components. Conduct feature analysis on each partition to determine its core functional requirements; among them, the decoration characteristics: surface smoothness, color consistency, no scratches or bubbles; the structural strength characteristics: crack resistance, small deformation, no local depression; the dimensional accuracy characteristics: geometric dimensions meet the design requirements, no burrs or irregular openings; the assembly area characteristics: smooth edges, no warping, seamless assembly; the heat dissipation area characteristics: no blockage of heat dissipation holes, no scratches or cracks on the material surface.

[0020] The core functional characteristics are the characteristics that play a key role in the product function and are sensitive to defects. For example, the surface smoothness of the operation panel and the accuracy of the touch area, the mechanical strength of the structural support area and the prevention and control of crack defects. The auxiliary functional characteristics are the functional characteristics that support or assist the core function, such as the ventilation efficiency of the heat dissipation area, the fastening degree of the assembly area, etc.

[0021] Establish a three-dimensional spatial distribution model according to the partition location of functional features, and use a coordinate system to mark the spatial positions of core and auxiliary functional features. By analyzing the collaborative relationship between core functional features and auxiliary functional features, determine the defect identification requirements for each partition. For example, the core feature area requires high-precision detection of defects such as scratches and cracks, while the auxiliary feature area can perform medium-precision detection of color difference, aperture, etc. At the same time, establish a corresponding relationship between the defect identification requirements and the position distribution of functional features.

[0022] Furthermore, according to the functional features and functional collaborative relationship of the core functional features and auxiliary functional features, determine the defect identification requirements, including: Based on the functional features, analyze the surface consistency, transition gradient, and auxiliary installation collaborative relationship of the core functional features and auxiliary functional features to determine the collaborative functional relationship; according to the functional features and the collaborative functional relationship, analyze the defect identification requirements of each functional feature to obtain the defect identification requirements.

[0023] The surface quality requirements of the core functional features and auxiliary functional features are consistent in the connection or adjacent areas. For example, there should be no obvious defects or differences in the surface transition between the decoration area and the operation panel, and the coating in the heat dissipation hole area should be consistent with the surrounding area.

[0024] For the transition design at the junction of different functional features (such as the decoration area and the structural support area), analyze whether the gradient change is smooth and whether there are defects such as sharp edges, protrusions, or depressions. For example, the boundary between the assembly area and the support area needs to ensure a smooth transition to avoid poor assembly or appearance defects.

[0025] The auxiliary functional features need to provide support for the core function. For example, the dimensional accuracy of the assembly hole and the structural strength work together to ensure the stability and sealing of the assembly process, while avoiding assembly failure caused by position deviation or deformation.

[0026] Through the above analysis, determine the collaborative relationship between the core functional features and auxiliary functional features, including physical connection relationship (such as whether the surface flatness and smoothness meet the assembly requirements), functional continuity (such as the aperture distribution and color consistency between the heat dissipation area and the decoration area), and mechanical strength collaboration (such as the strength matching between the structural support and dimensional accuracy features).

[0027] For each functional feature, determine potential defect types according to its specific functional requirements and performance criteria, such as scratches, dents, color differences, excessive assembly gaps, etc. Consider the interactions and dependencies between functional features and identify defects that may be caused by collaborative function failures, such as overheating due to poor heat dissipation, performance degradation due to loose assembly, etc. For example, in the decorative area: surface scratch detection, color difference detection; in the heat dissipation area: aperture integrity detection, surface flatness detection; in the assembly area: mounting hole size detection, edge defect detection; in areas with complex collaborative relationships, multiple detection indicators need to be integrated. For example, in the assembly area, dimensional accuracy, smoothness of edge transition, and crack defects may need to be detected simultaneously.

[0028] Based on the position distribution, collect image information of the plastic shell product, and the image information has a position distribution identifier.

[0029] Use a high-resolution industrial camera or other imaging device to collect image information according to the spatial position distribution of partitions. The position distribution identifier is attached to the image information so that subsequent processing can be matched with the functional partitions.

[0030] According to the corresponding relationship between the defect identification requirements and the position distribution, perform partition defect identification parameter positioning on the image information to obtain partition identification parameters. When there are multiple partition identification parameters, there are identification parameter weights.

[0031] According to the defect identification requirements and position distribution identifiers of each functional partition, perform partition processing on the collected image information, divide the image into multiple regions, which respectively correspond to the functional partitions. For each partition, combine the specific defect types in this region (such as scratch detection of the operation panel, aperture integrity detection of the heat dissipation area) and detection requirements (such as detection accuracy, target feature size) to determine the partition identification parameters. The partition identification parameters include color, texture, shape, size, etc., which specifically depend on the defect identification requirements. When there are multiple partition identification parameters, assign weights to each parameter, and the weights are set according to the partition importance and defect feature complexity.

[0032] Furthermore, according to the corresponding relationship between the defect identification requirements and the position distribution, perform partition defect identification parameter positioning on the image information to obtain partition identification parameters, including: Perform defect image parameter conversion according to the defect identification requirements to obtain image parameter features; perform partition coordinate positioning on the image information according to the position distribution, perform functional partition identification mapping on the image information to determine the functional partitions in the image information; based on the corresponding relationship between the position distribution and the defect identification requirements, project the image parameter features into the functional partitions in the image information to obtain the partition identification parameters.

[0033] Specifically, according to the defect recognition requirements of each functional area, the image information is processed and analyzed to extract image parameter features related to defects, including color, texture, shape, edge, brightness, contrast, etc. For example, surface texture, edge features, or color distribution features are extracted through image processing algorithms, and the extracted image parameter features are converted into formats or standards that can be used for defect recognition, such as numerical values, vectors, matrices, etc.

[0034] According to the pre-established partition position distribution, the image information is located by coordinates according to the functional areas, that is, through the position distribution identification and the coordinate system of the image information, the regional ranges of each functional area in the image are calibrated one by one (such as the operation area, the heat dissipation area), and then, the image information is mapped and matched with the functional areas to clarify the image area corresponding to each functional area, so as to conduct targeted defect recognition.

[0035] Based on the corresponding relationship between the position distribution and the defect recognition requirements, the extracted image parameter features are projected into the functional areas in the image information. Specifically, the image parameter features (such as edge sharpness or texture integrity) are associated with the corresponding functional areas (such as the operation panel or the decorative border) through the mapping relationship, so as to determine the specific defect recognition parameters of each functional area. According to the projected image parameter features, partition recognition parameters are determined for each functional area, and these parameters may include the number of pixels in a specific color range, the statistical value of texture features, the matching degree of shape features, etc.

[0036] Configure the defect recognition module according to the partition recognition parameters, and the defect recognition module includes an integrated learning model that matches the partition recognition parameters and the recognition parameter weights; Use an integrated learning model (such as multi-model fusion based on random forest or deep learning) that matches the partition recognition parameters and the recognition parameter weights to optimize the configuration of the defect recognition module, so that it has the efficient recognition ability for different partition characteristics.

[0037] Furthermore, before configuring the defect recognition module according to the partition recognition parameters, it also includes: Obtain a surface defect sample set of the household appliance shell; perform defect category aggregation according to the surface defect sample set to obtain a defect category cluster; use the defect category cluster to construct a training sample set for each defect category, and conduct recognition model training for each defect category through the training sample set to obtain multiple defect category recognition models; extract the image recognition parameters of the defect category recognition models, and use the image recognition parameters as model labels to identify the multiple defect category recognition models.

[0038] First, obtain a surface defect sample set of the household appliance shell. The surface defect sample set includes various defect image data (such as scratches, bubbles, color differences, cracks, etc.) collected in actual production and their corresponding annotation information. Subsequently, conduct an aggregation analysis on the defect data in the surface defect sample set, and classify the defects into several categories according to defect characteristics (such as shape, size, color change) to generate defect category clusters, such as scratch category, hole category, coating abnormality category, etc. Based on these defect category clusters, further construct training sample sets for each category to ensure that each training sample set contains a sufficient number and diverse samples to support the robustness of the model. Next, use these training sample sets to train independent recognition models for each defect category. Through deep learning or other machine learning methods (such as convolutional neural network or random forest), generate exclusive defect category recognition models for each category. For example, the scratch detection model focuses on the extraction of edge sharpness features, while the color difference detection model focuses on the analysis of color distribution. After training, extract the image recognition parameters (such as feature weights, detection thresholds, etc.) from each defect category recognition model, and use these parameters as model labels to assign unique identifiers to each model. In this way, the defect category recognition models not only have classification functions but also can be quickly matched and called according to the model labels, supporting the flexible configuration of defect recognition modules according to the partition recognition parameters in the follow-up.

[0039] Furthermore, configuring the defect recognition module according to the partition recognition parameters includes: Match the partition recognition parameters with the image recognition parameters of the model label to determine the matching defect category recognition model; set the ensemble learning coefficient of the matching defect category recognition model according to the recognition parameter weight value; conduct ensemble learning on the matching defect category recognition model according to the ensemble learning coefficient to obtain an ensemble learning model, and generate the defect recognition module based on the ensemble learning model.

[0040] First, the image recognition features extracted from the partition recognition parameters (such as edge sharpness, brightness gradient, texture frequency, etc.) are matched one by one with the model labels (including image recognition parameters) of multiple defect category recognition models generated during the training phase to determine the defect category recognition model that best fits the partition requirements. For example, for the partition recognition parameters of the heat dissipation hole area, an aperture consistency detection model can be matched; for the partition recognition parameters of the decoration area, a color difference detection model can be matched. Then, according to the recognition parameter weights in the partition recognition parameters, corresponding ensemble learning coefficients are assigned to the matched defect category recognition models, and the ensemble learning coefficients reflect the importance of the models in the final recognition task. For example, scratch detection in the operation area may be assigned a higher learning coefficient, while the learning coefficient for coating detection is relatively low to highlight the high-priority detection targets. Subsequently, based on the set ensemble learning coefficients, ensemble learning is performed on the matched multiple defect category recognition models. Through weighted ensemble methods (such as weighted average, Boosting, Bagging, etc.), the advantages of each model are combined to improve the accuracy and robustness of defect recognition. The ensemble learning process adjusts the weight relationship between the models so that the recognition results can effectively balance various defect types within the partition and avoid misjudgments of a single model. Finally, the defect recognition module for a specific partition is generated using the results of ensemble learning. Each module contains an optimized ensemble learning model and has the ability to accurately recognize various potential defect types within the partition.

[0041] Furthermore, the training of the recognition models for each defect category is performed through the training sample set, including: According to the defect category cluster, perform an analysis of the recognition anomaly probability to determine the adversarial image parameters, where the adversarial image parameters are the image parameters when the recognition anomaly probability reaches a preset threshold; perform a defect parameter compensation analysis according to the adversarial image parameters to determine the image enhancement parameters; use the image enhancement parameters to perform image enhancement processing on the training sample set, and perform enhanced training of the recognition model through the training sample set after image enhancement processing to obtain the defect category recognition model.

[0042] First, according to the defect category clusters, analyze the recognition anomaly probability for each defect category in the training sample set. The recognition anomaly probability refers to the likelihood of misjudgment of certain images by the model during preliminary training or testing. By evaluating the anomaly probabilities of these image parameters (such as brightness gradient, edge features, or texture distribution), determine the adversarial image parameters, which are the key image feature parameters that make the recognition anomaly probability reach a preset threshold. Then, conduct defect parameter compensation analysis for the adversarial image parameters and calculate the enhancement requirements for optimizing these image parameters. For example, for samples misjudged due to lighting changes, the brightness parameter can be compensated; for samples misjudged due to blurred edges, the sharpening parameter can be optimized. After analysis, generate corresponding image enhancement parameters, such as brightness adjustment coefficients, sharpening radii, contrast enhancement intensities, etc. Subsequently, use the image enhancement parameters to perform image enhancement processing on the training sample set to increase the diversity and robustness of the sample set. The enhancement processing includes adding simulated defect scenarios (such as noise, lighting changes), expanding the distribution range of training data, and strengthening the feature regions that are easily misjudged. For example, perform contrast enhancement or edge sharpening on samples for scratch detection, and adjust the color distribution range for samples for color difference detection. Finally, perform enhanced training on the recognition model using the training sample set after image enhancement processing. During the enhanced training phase, based on the rich and diverse sample characteristics, optimize the model parameters to improve the model's ability to distinguish between real defects and adversarial samples, thereby obtaining a more robust defect category recognition model.

[0043] Furthermore, setting the ensemble learning coefficient for the matching defect category recognition model according to the recognition parameter weights further includes: Obtain the unmatched defect category recognition models among the multiple defect category recognition models and configure the ensemble learning coefficient of the unmatched defect category recognition models to the minimum preset value.

[0044] On the basis of determining the matching defect category recognition models and setting their ensemble learning coefficients according to the recognition parameter weights, for the unmatched defect category recognition models, configure their ensemble learning coefficients to the minimum preset value. The unmatched defect category recognition models refer to the models that are not explicitly called in the recognition requirements of the current partition but may still have an auxiliary role in specific situations. Specifically, when configuring the ensemble learning coefficients, first, according to the importance of the matching defect category recognition models (such as the partition function features and recognition requirements with high priority), allocate higher learning coefficients. For example, a higher weight can be assigned to the scratch detection model in the core partition. For the unmatched models, uniformly allocate a learning coefficient with the minimum preset value to ensure that they will not have a significant impact on the final decision during the ensemble learning process but still retain their basic participation.

[0045] Use the defect recognition module to perform partition defect recognition on the corresponding image information to obtain defect recognition results.

[0046] According to the partition recognition parameters and their weights, call the configured defect recognition module, input the image information marked by functional partitions into the corresponding defect recognition module, and the defect recognition module gradually detects the input partitioned image information, and finally outputs the defect recognition results, including defect type, defect location, defect severity, etc.

[0047] In summary, the embodiments of the present application have at least the following technical effects: First, obtain the design information of the household appliance shell, perform function recognition according to the design information to obtain the shell functional partitions. Then, according to the functional characteristics of the shell functional partitions, determine the defect recognition requirements and location distributions. Further, based on the location distribution, collect the image information of the plastic shell product, and the image information has a location distribution identifier. Then, according to the correspondence between the defect recognition requirements and the location distribution, perform partition defect recognition parameter positioning on the image information to obtain partition recognition parameters, and there are recognition parameter weights when there are multiple partition recognition parameters. Next, configure the defect recognition module according to the partition recognition parameters, and the defect recognition module includes an integrated learning model that matches the partition recognition parameters and the recognition parameter weights. Finally, use the defect recognition module to perform partition defect recognition on the corresponding image information to obtain the defect recognition results. This solves the technical problem of low surface defect detection efficiency and recognition accuracy in the prior art, optimizes the defect detection parameters through the functional partition characteristics, and achieves the technical effect of improving the visual detection efficiency and recognition accuracy.

[0048] Embodiment 2, based on the same inventive concept as the method for visually detecting surface defects of plastic products in the foregoing embodiment, as Figure 2 shown, the present application provides a system for visually detecting surface defects of plastic products, wherein the system includes: A function recognition component 11: obtain the design information of the household appliance shell, perform function recognition according to the design information to obtain the shell functional partitions; a defect recognition requirement determination component 12: determine the defect recognition requirements and location distributions according to the functional characteristics of the shell functional partitions; an image acquisition component 13: based on the location distribution, collect the image information of the plastic shell product, and the image information has a location distribution identifier; a parameter positioning component 14: according to the correspondence between the defect recognition requirements and the location distribution, perform partition defect recognition parameter positioning on the image information to obtain partition recognition parameters, and there are recognition parameter weights when there are multiple partition recognition parameters; a module configuration component 15: configure a defect recognition module according to the partition recognition parameters, and the defect recognition module includes an integrated learning model that matches the partition recognition parameters and the recognition parameter weights; a defect recognition execution component 16: use the defect recognition module to perform partition defect recognition on the corresponding image information to obtain the defect recognition results.

[0049] Further, the defect recognition requirement determination component 12 is used to execute the following method: Perform partition function analysis according to the housing function partition to determine functional features, where the functional features include decorative features, structural strength features, dimensional accuracy features, assembly area features, and heat dissipation area features; perform core functional feature and auxiliary functional feature recognition and analysis according to the functional features; mark the functional feature position distribution according to the recognition positions of the core functional features and auxiliary functional features; determine the defect recognition requirements according to the functional features and functional coordination relationships of the core functional features and auxiliary functional features, and establish the corresponding relationship between the defect recognition requirements and the functional feature position distribution.

[0050] Further, the defect recognition requirement determination component 12 is used to execute the following method: Based on the functional features, perform surface consistency, transition gradient, and auxiliary installation coordination relationship analysis on the core functional features and auxiliary functional features to determine the coordination functional relationship; perform defect recognition requirement analysis for each functional feature according to the functional features and the coordination functional relationship to obtain the defect recognition requirements.

[0051] Further, the parameter positioning component 14 is used to execute the following method: Perform defect image parameter conversion according to the defect recognition requirements to obtain image parameter features; perform partition coordinate positioning on the image information according to the position distribution, perform functional partition recognition mapping on the image information to determine the functional partitions in the image information; project the image parameter features into the functional partitions in the image information based on the corresponding relationship between the position distribution and the defect recognition requirements to obtain the partition recognition parameters.

[0052] Further, the module configuration component 15 is used to execute the following method: Obtain the surface defect sample set of the home appliance housing; perform defect category aggregation according to the surface defect sample set to obtain defect category clusters; use the defect category clusters to construct training sample sets for each defect category, and perform recognition model training for each defect category through the training sample sets to obtain multiple defect category recognition models; extract the image recognition parameters of the defect category recognition models, and use the image recognition parameters as model labels to identify the multiple defect category recognition models.

[0053] Further, the module configuration component 15 is used to execute the following method: Match the partition recognition parameters with the image recognition parameters of the model tags to determine a matching defect category recognition model; set the ensemble learning coefficient of the matching defect category recognition model according to the recognition parameter weights; perform ensemble learning on the matching defect category recognition model according to the ensemble learning coefficient to obtain an ensemble learning model, and generate the defect recognition module based on the ensemble learning model.

[0054] Further, the module configuration component 15 is used to execute the following method: Perform recognition anomaly probability analysis according to the defect category cluster to determine adversarial image parameters, where the adversarial image parameters are image parameters with a recognition anomaly probability reaching a preset threshold; perform defect parameter compensation analysis according to the adversarial image parameters to determine image enhancement parameters; use the image enhancement parameters to perform image enhancement processing on the training sample set, and perform recognition model enhancement training through the training sample set after image enhancement processing to obtain a defect category recognition model.

[0055] Further, the module configuration component 15 is used to execute the following method: Obtain the unmatched defect category recognition models among the multiple defect category recognition models, and configure the ensemble learning coefficient of the unmatched defect category recognition models to the minimum preset value.

[0056] It should be noted that the above sequence of embodiments of the present application is only for description and does not represent the superiority or inferiority of the embodiments. And the above description of specific embodiments of this specification has been made. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0057] The above are only the preferred embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0058] This specification and the drawings are only exemplary descriptions of the present application and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of the present application. Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalent technologies, the present application is intended to include these changes and modifications.

Claims

1. A method for visually detecting surface defects of plastic products, characterized in that: The method comprises: Acquire design information of the housing of the home appliance, perform function identification according to the design information, and obtain functional partitions of the housing; Determine defect identification requirements and location distribution according to the functional characteristics of the shell functional partitions; Based on the position distribution, collecting image information of the plastic housing product, wherein the image information has a position distribution mark; According to the corresponding relationship between the defect recognition requirement and the position distribution, the image information is partitioned and defect recognition parameters are located to obtain partition recognition parameters, and when there are multiple partition recognition parameters, there are recognition parameter weights; configuring a defect recognition module according to the partition recognition parameters, wherein the defect recognition module includes an integrated learning model that matches the partition recognition parameters and recognition parameter weights; The defect recognition module is used to perform partition defect recognition on the corresponding image information to obtain a defect recognition result.

2. The method for visually inspecting surface defects of plastic products according to claim 1, characterized in that: According to the functional characteristics of the shell functional partition, the defect identification requirements and location distribution are determined, including: Performing zoning function analysis according to the functional zoning of the shell to determine functional features, wherein the functional features include decorative features, structural strength features, dimensional accuracy features, assembly area features, and heat dissipation area features; According to the functional characteristics, identification and analysis of core functional characteristics and auxiliary functional characteristics are performed; Marking the functional feature position distribution according to the identification positions of the core functional features and the auxiliary functional features; According to the functional characteristics and functional coordination relationship of the core functional characteristics and the auxiliary functional characteristics, the defect identification requirements are determined, and a corresponding relationship between the defect identification requirements and the functional characteristic position distribution is established.

3. The method for visually inspecting surface defects of plastic products according to claim 2, characterized in that: Determine the defect identification requirements based on the functional characteristics and functional synergy of the core functional characteristics and auxiliary functional characteristics, including: Based on the functional features, analyzing the surface consistency, transition gradient, and auxiliary installation synergistic relationship of the core functional features and the auxiliary functional features to determine the synergistic functional relationship; Defect identification requirements of each functional feature are analyzed according to the functional features and the collaborative functional relationships to obtain the defect identification requirements.

4. The method for visually inspecting surface defects of plastic products according to claim 1, characterized in that: According to the correspondence between the defect recognition requirement and the position distribution, the image information is partitioned and defect recognition parameters are located to obtain partition recognition parameters, including: Perform defect image parameter conversion according to the defect recognition requirements to obtain image parameter features; Performing partition coordinate positioning on the image information according to the position distribution, performing functional partition identification mapping on the image information, and determining the functional partitions in the image information; Based on the corresponding relationship between the position distribution and the defect recognition requirement, the image parameter features are projected into the functional partitions in the image information to obtain the partition recognition parameters.

5. The method for visually inspecting surface defects of plastic products according to claim 1, characterized in that: The defect recognition module is configured according to the partition recognition parameters, and the method also includes: Obtain a sample set of surface defects of household appliance casings; Aggregating defect categories according to the surface defect sample set to obtain defect category clusters; Using the defect category cluster to construct a training sample set for each defect category, and using the training sample set to train a recognition model for each defect category to obtain multiple defect category recognition models; The image recognition parameters of the defect category recognition model are extracted, and the image recognition parameters are used as model labels to identify the multiple defect category recognition models.

6. The method for visually inspecting surface defects of plastic products according to claim 5, characterized in that: Configuring a defect recognition module according to the partition recognition parameters includes: According to the partition recognition parameters, the image recognition parameters of the model label are matched to determine a matching defect category recognition model; According to the recognition parameter weights, setting the ensemble learning coefficient of the matching defect category recognition model; The matching defect category recognition model is subjected to ensemble learning according to the ensemble learning coefficient to obtain an ensemble learning model, and the defect recognition module is generated based on the ensemble learning model.

7. The method for visually inspecting surface defects of plastic products according to claim 5, characterized in that: The identification model training of each defect category is performed using the training sample set, including: According to the defect category cluster, an abnormality recognition probability analysis is performed to determine adversarial image parameters, where the adversarial image parameters are image parameters whose abnormality recognition probability reaches a preset threshold; Perform defect parameter compensation analysis according to the countermeasure image parameters to determine image enhancement parameters; The image enhancement parameters are used to perform image enhancement processing on the training sample set, and recognition model enhancement training is performed on the training sample set after the image enhancement processing to obtain a defect category recognition model.

8. The method for visually inspecting surface defects of plastic products according to claim 6, characterized in that: According to the recognition parameter weights, setting the ensemble learning coefficient of the matching defect category recognition model also includes: An unmatched defect category recognition model among the multiple defect category recognition models is obtained, and an ensemble learning coefficient of the unmatched defect category recognition model is configured to be a minimum preset value.

9. A system for visually inspecting surface defects of plastic products, characterized in that: A method for visually inspecting surface defects of plastic products according to any one of claims 1 to 8, the system comprising: Function identification component: obtains the design information of the housing of the home appliance, performs function identification based on the design information, and obtains the functional partition of the housing; Defect identification requirement determination component: determining defect identification requirements and location distribution according to functional characteristics of the functional partitions of the shell; Image acquisition component: based on the position distribution, collects image information of the plastic housing product, the image information has a position distribution mark; Parameter positioning component: based on the correspondence between the defect recognition requirement and the position distribution, the image information is partitioned and defect recognition parameters are positioned to obtain partition recognition parameters, and when there are multiple partition recognition parameters, there are recognition parameter weights; Module configuration component: configuring a defect recognition module according to the partition recognition parameters, wherein the defect recognition module includes an integrated learning model matching the partition recognition parameters and recognition parameter weights; Defect recognition execution component: utilizes the defect recognition module to perform partition defect recognition on the corresponding image information to obtain defect recognition results.

Images