Plastic product quality inspection method and system based on machine vision

Abstract

The present application provides a method and system for quality inspection of plastic products based on machine vision, which relates to the field of optical intelligent inspection technology. The method includes: extracting information from an image data set through a contour recognition channel to determine the contour of an identification component; using the contour of the identification component as a constraint, performing fitting and filling of a point cloud data stream, generating a point cloud of the identification component, and fusing and building a plastic mesh model; performing comparison based on the plastic mesh model to locate dimensional deviation information; extracting defects from the plastic mesh model based on defect recognition elements, determining abnormal defect information, and generating a quality inspection report. This application can solve the technical problems that the existing methods have insufficient precision, accuracy, and completeness in detecting surface defects and dimensional deviations, and have a low degree of automation, resulting in poor detection accuracy, comprehensiveness, and detection efficiency. It can improve the automation of quality inspection and achieve the effect of improving the accuracy, comprehensiveness, and detection efficiency of inspection.

Description

Technical Field

[0001] The present application relates to the field of optical intelligent detection technology, and in particular to a plastic product quality detection method and system based on machine vision. Background Art

[0002] Plastic product quality inspection is a comprehensive and detailed inspection and evaluation of plastic products to ensure that their quality, performance and safety meet relevant standards and requirements. The traditional method of manual sampling inspection has low inspection efficiency and poor coverage. At present, the existing plastic product quality inspection method mainly uses image recognition to conduct quality inspection. This method of three-dimensional dimensional deviation analysis through two-dimensional images has large errors and cannot accurately capture the internal surface defects of plastic products, resulting in poor accuracy and integrity of internal surface defect detection.

[0003] In summary, the existing single plastic product quality inspection method based on image recognition has insufficient precision, accuracy and completeness in surface defect detection and dimensional deviation detection, and a low degree of automation, resulting in poor accuracy and comprehensiveness in plastic product quality inspection, as well as technical problems such as low detection efficiency. Summary of the invention

[0004] The purpose of this application is to provide a plastic product quality inspection method and system based on machine vision, so as to solve the technical problems that the existing single plastic product quality inspection method through image recognition has insufficient precision, accuracy and completeness in surface defect detection and dimensional deviation detection, and has a low degree of automation, resulting in poor accuracy and comprehensiveness of plastic product quality inspection, and low detection efficiency.

[0005] In view of the above problems, the present application provides a plastic product quality inspection method and system based on machine vision.

[0006] In a first aspect, the present application provides a plastic product quality inspection method based on machine vision, and the method is implemented by a plastic product quality inspection system based on machine vision, wherein the method comprises: based on a three-dimensional scanner and an image sensor, collecting information on multiple components of a plastic object to be inspected, generating a point cloud data stream and an image data set; extracting contour information from the image data set through a convergent contour recognition channel, and determining multiple identification component contours; taking the multiple identification component contours as constraints, respectively performing fitting filling of the point cloud data stream, generating multiple identification component point clouds that meet the expected fitting constraints, and building a plastic mesh model based on the fusion of the multiple identification component contours and the multiple identification component point clouds; The plastic mesh model is compared and analyzed with the standard mesh model in the standard model library to locate the size deviation information, wherein the size deviation information includes the deviation value and the deviation coordinates; the preset defect identification elements are obtained, wherein the preset defect identification elements include the point cloud distribution topology and the point cloud mesh density threshold, and correspond one to one with the preset defect type; the surface point cloud traversal screening and abnormal defect extraction of the plastic mesh model are performed based on the preset defect identification elements to determine the abnormal defect information, wherein the abnormal defect information includes the defect type, defect intensity and defect coordinates; a target quality inspection report is generated based on the size deviation information and the abnormal defect information, and subsequent processing of the plastic object to be inspected is performed according to the target quality inspection report.

[0007] In the second aspect, the present application also provides a plastic product quality inspection system based on machine vision, which is used to execute a plastic product quality inspection method based on machine vision as described in the first aspect, wherein the system includes: an information acquisition module, the information acquisition module is used to collect information on multiple components of the plastic object to be inspected based on a three-dimensional scanner and an image sensor, and generate a point cloud data stream and an image data set; a contour information extraction module, the contour information extraction module is used to extract contour information from the image data set through a convergent contour recognition channel, and determine multiple identification component contours; a plastic mesh model building module, the plastic mesh model building module is used to use the multiple identification component contours as constraints, respectively perform fitting filling of the point cloud data stream, generate multiple identification component point clouds that meet the expected fitting constraints, and build a plastic mesh model based on the fusion of the multiple identification component contours and the multiple identification component point clouds; a size deviation information positioning module, the size deviation information positioning module The bit module is used to compare and analyze the plastic mesh model with the standard mesh model in the standard model library to locate the size deviation information, and the size deviation information includes the deviation value and the deviation coordinates; the preset defect identification element acquisition module, the preset defect identification element acquisition module is used to obtain the preset defect identification element, and the preset defect identification element includes the point cloud distribution topology and the point cloud mesh density threshold, and corresponds to the preset defect type one by one; the abnormal defect information determination module, the abnormal defect information determination module is used to perform surface point cloud traversal screening and abnormal defect extraction on the plastic mesh model based on the preset defect identification element, and determine the abnormal defect information, and the abnormal defect information includes the defect type, defect intensity and defect coordinates; the target quality inspection report generation module, the target quality inspection report generation module is used to generate a target quality inspection report based on the size deviation information and the abnormal defect information, and perform subsequent processing of the plastic object to be inspected according to the target quality inspection report.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0009] 1. Based on a three-dimensional scanner and an image sensor, information is collected on multiple parts of the plastic object to be inspected to generate a point cloud data stream and an image data set; contour information is extracted from the image data set through a convergent contour recognition channel to determine multiple identification component contours; with the multiple identification component contours as constraints, the point cloud data stream is fitted and filled respectively to generate multiple identification component point clouds that meet the expected fitting constraints, and a plastic mesh model is built based on the multiple identification component contours and the multiple identification component point clouds; the plastic mesh model is compared and analyzed with a standard mesh model in a standard model library to locate size deviation information, and the size deviation information includes a deviation value and a deviation coordinate; a preset defect identification element is obtained, and the preset defect identification element includes a point cloud distribution topology and a point cloud mesh density threshold, and corresponds to a preset defect type one by one; based on the preset defect identification element, the plastic mesh model is traversed and screened and abnormal defects are extracted to determine abnormal defect information, and the abnormal defect information includes defect type, defect intensity and defect coordinates; a target quality inspection report is generated based on the size deviation information and the abnormal defect information, and subsequent processing of the plastic object to be inspected is performed according to the target quality inspection report. That is to say, by collecting images and point cloud data of multiple parts of the plastic object to be inspected, and determining the contour of the part based on image analysis; then fitting and filling the point cloud data based on the part contour to generate a part point cloud; then using the part contour as a framework and the part point cloud as the filling data fusion to build a plastic mesh model; further comparing the size deviation of the plastic mesh model to determine the size deviation value and positioning deviation coordinates; on the other hand, based on preset defect recognition elements, traversing and screening the surface point cloud and extracting abnormal defects of the plastic mesh model to determine the surface abnormal defect information, including defect type, defect intensity and defect coordinates; finally, combining the size deviation information and the abnormal defect information to form a quality inspection report, the precision, accuracy and completeness of surface defect detection and size deviation detection of plastic products can be improved, and at the same time, the automation and intelligence of quality inspection can be improved, thereby achieving the technical effect of improving the accuracy, comprehensiveness and efficiency of quality inspection.

[0010] 2. By using the component outline as the framework and the component point cloud as the filling data fusion to build a plastic solid model, and meshing the plastic solid model to obtain a plastic mesh model, the precision and accuracy of the plastic mesh model can be improved; then, based on the comparison and analysis of the plastic mesh model and the standard mesh model, the dimensional deviation information can be determined, which can improve the intelligence of the dimensional deviation analysis, and at the same time improve the precision, accuracy and efficiency of the dimensional deviation analysis.

[0011] 3. By obtaining preset defect identification elements based on preset surface defect type analysis, and performing surface point cloud feature screening and abnormal defect feature extraction based on the preset defect identification elements, the accuracy and comprehensiveness of abnormal defect information can be improved, while improving the efficiency of surface defect identification.

[0012] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented according to the contents of the specification, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are specifically cited below. It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easy to understand through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions in the present application or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0014] Figure 1 A schematic diagram of a process for detecting quality of plastic products based on machine vision in this application;

[0015] Figure 2 A schematic diagram of a flow chart of obtaining a convergent contour recognition channel in a plastic product quality inspection method based on machine vision in this application;

[0016] Figure 3 This is a structural schematic diagram of a plastic product quality inspection system based on machine vision for this application.

[0017] Description of reference numerals:

[0018] Information collection module 11, contour information extraction module 12, plastic mesh model building module 13, size deviation information positioning module 14, preset defect recognition element acquisition module 15, abnormal defect information determination module 16, target quality inspection report generation module 17. DETAILED DESCRIPTION

[0019] This application provides a plastic product quality inspection method and system based on machine vision, which solves the existing single plastic product quality inspection method through image recognition, which has insufficient precision, accuracy and completeness in surface defect detection and dimensional deviation detection, and low automation, resulting in poor precision and comprehensiveness of plastic product quality inspection, and low detection efficiency. The precision, accuracy and completeness of plastic product surface defect detection and dimensional deviation detection can be improved, and the automation and intelligence of quality inspection can be improved, thereby achieving the technical effect of improving the precision, comprehensiveness and detection efficiency of quality inspection.

[0020] Below, the technical solutions in the present application will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments of the present application. It should be understood that the present application is not limited to the example embodiments described herein. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present application. It should also be noted that, for the convenience of description, only the parts related to the present application are shown in the accompanying drawings, rather than all of them.

[0021] Embodiment 1

[0022] Please see attached Figure 1 The present application provides a plastic product quality inspection method based on machine vision, wherein the method is applied to a plastic product quality inspection system based on machine vision, the system is communicatively connected with a three-dimensional scanner and an image sensor, and the method specifically comprises the following steps:

[0023] Step 1: Based on the 3D scanner and image sensor, collect information on multiple parts of the plastic object to be inspected, and generate point cloud data streams and image data sets;

[0024] Specifically, the method provided in the present application is used to optimize the existing plastic product quality inspection method to achieve the technical goal of improving the automation and intelligence of quality inspection, and to achieve the technical effect of improving the accuracy, comprehensiveness and inspection efficiency of quality inspection. The method is specifically implemented in a plastic product quality inspection system based on machine vision, wherein the system can receive point cloud data and image data collected by a three-dimensional scanner and an image sensor through data transmission, wherein the three-dimensional scanner is used to collect point cloud data of plastic products, such as a three-dimensional laser scanner, which calculates the three-dimensional coordinate information of each point on the surface of an object by emitting a laser beam and receiving its reflected signal, thereby realizing fast and accurate three-dimensional measurement of the surface of the object; the image sensor is used to collect image data of plastic products, such as a CCD image sensor, etc.

[0025] First, point cloud data collection and image collection are performed on multiple parts of the inspected plastic object through a 3D scanner and an image sensor to obtain a point cloud data stream and an image data set. The point cloud data stream contains multiple part point cloud data, and the image data set contains multiple part images, and the part point cloud data and part images have a mapping association. The point cloud data stream and image data set obtained through collection provide raw data support for the next step of building a plastic mesh model.

[0026] Step 2: extracting contour information from the image data set through a convergent contour recognition channel to determine contours of multiple identification components;

[0027] Specifically, a convergent contour recognition channel is constructed based on a convolutional neural network. The convergent contour recognition channel is a neural network model that can be iteratively optimized in machine learning and is trained to a convergent state through a training data set. Then the image data set is input into the convergent contour recognition channel for contour recognition, and multiple identified component contours are output. By constructing a convergent contour recognition channel to extract component contour information, the accuracy and efficiency of obtaining the identified component contour can be improved.

[0028] Step 3: Taking the multiple identification component contours as constraints, respectively performing fitting filling of the point cloud data streams to generate multiple identification component point clouds that meet the expected fitting constraints, and building a plastic mesh model based on the fusion of the multiple identification component contours and the multiple identification component point clouds;

[0029] Specifically, the outlines of the multiple identified components are used as framework constraints, and the point cloud data of the same components in the point cloud data stream are randomly fitted and filled, and multiple identified component point clouds that meet the expected fitting constraints are obtained. The expected fitting constraints are preset point cloud ratio thresholds, which can be set by technicians in this field according to actual conditions. Then, in the three-dimensional simulation platform, based on the component correspondence, the outlines of the multiple identified components are used as component frameworks, and the point clouds of the identified components are used as framework filling content to fuse and build a plastic solid model; then, the plastic solid model is meshed according to the preset grid size to obtain a plastic mesh model. By using the component outline as the framework, the component point cloud as the filling data to fuse and build a plastic solid model, and meshing the plastic solid model to obtain a plastic mesh model, the precision and accuracy of the plastic mesh model can be improved, and at the same time, a basis for the next step of size deviation comparison and positioning size deviation information is provided.

[0030] Step 4: Compare and analyze the plastic mesh model with the standard mesh model in the standard model library to locate size deviation information, where the size deviation information includes a deviation value and a deviation coordinate;

[0031] Specifically, call a standard model library, which includes standard grid models of multiple sample plastic objects; obtain the type information or ID identification of the target object to be inspected, and perform model matching in the standard model library based on the type information or ID identification to obtain a standard grid model, wherein the standard grid model is of the same type and size as the plastic grid model, and has the same grid size. Then, the plastic grid model and the standard grid model are traversed and compared and analyzed, and the size deviation information is located based on the comparison and analysis results, wherein the size deviation information includes a deviation value and a deviation coordinate, wherein the deviation value is the difference between the size data of the plastic grid model and the size data of the corresponding position of the standard grid model; the deviation coordinate refers to the position coordinate corresponding to the size deviation value. By determining the size deviation information based on the comparison and analysis of the plastic grid model and the standard grid model, the intelligence of the size deviation analysis can be improved, while the precision, accuracy and efficiency of the size deviation analysis can be improved.

[0032] Step 5: Obtaining preset defect identification elements, wherein the preset defect identification elements include point cloud distribution topology and point cloud grid density threshold, and correspond one to one with preset defect types;

[0033] Specifically, a preset defect identification element is obtained, and the preset defect identification element corresponds to the preset defect type one by one. The preset defect type includes at least bubble defects, crack defects and scratch defects. The technicians in this field can also add the required defect types according to actual needs; the preset defect identification element includes point cloud distribution topology and point cloud mesh density threshold, wherein the point cloud distribution topology refers to the point cloud distribution characteristics of the plastic object surface corresponding to the defect type, and the point cloud mesh density threshold refers to the point cloud density range of the plastic object surface corresponding to the defect type. By obtaining the preset defect identification element, a basis is provided for the next step of abnormal defect identification of the surface point cloud of the plastic mesh model, and the accuracy and intelligence of surface abnormal defect identification can be improved.

[0034] Step 6: Based on the preset defect identification elements, the surface point cloud of the plastic mesh model is traversed and screened and abnormal defects are extracted to determine abnormal defect information, where the abnormal defect information includes defect type, defect intensity and defect coordinates;

[0035] Specifically, the surface point cloud traversal screening and abnormal defect extraction of the plastic mesh model are performed based on the preset defect identification elements, that is, the point cloud data that meets the preset defect identification elements is extracted, and the position coordinates of the point cloud data are located, the defect coordinates are determined, and the defect type and defect intensity are determined based on the extracted point cloud data, wherein the defect intensity represents the degree of defect, and the greater the degree of defect, the greater the defect intensity; finally, the abnormal defect information is obtained by combining the defect type, the defect intensity and the defect coordinates. By performing surface point cloud traversal screening and abnormal defect extraction on the plastic mesh model based on the preset defect identification elements, the accuracy, comprehensiveness and intelligence of surface abnormal defect analysis can be improved.

[0036] Step seven: Generate a target quality inspection report based on the size deviation information and the abnormal defect information, and perform subsequent processing of the plastic object to be inspected according to the target quality inspection report.

[0037] Specifically, a target quality inspection report of the plastic object to be inspected is generated based on the size deviation information and the abnormal defect information, and then the plastic object to be inspected is subsequently processed according to the target quality inspection report. For example, whether defects can be repaired is analyzed based on the target quality inspection report, and plastic objects that can be repaired with low repair costs are recycled, and plastic objects with high repair costs or plastic objects that cannot be repaired are treated as waste.

[0038] The plastic product quality inspection method based on machine vision is applied to a plastic product quality inspection system based on machine vision, which can solve the existing single plastic product quality inspection method through image recognition, which has insufficient precision, accuracy and completeness in surface defect detection and dimensional deviation detection, and low degree of automation, resulting in poor precision and comprehensiveness of plastic product quality inspection, and low detection efficiency. Technical problems. First, based on a three-dimensional scanner and an image sensor, information is collected from multiple components of the plastic object to be inspected to generate a point cloud data stream and an image data set; then, contour information is extracted from the image data set through a convergent contour recognition channel to determine multiple identification component contours; then, with the multiple identification component contours as constraints, the point cloud data stream is fitted and filled respectively to generate multiple identification component point clouds that meet the expected fitting constraints, and a plastic mesh model is built based on the fusion of the multiple identification component contours and the multiple identification component point clouds; next, the plastic mesh model is compared and analyzed with the standard mesh model in the standard model library, and the positioning ruler is used to locate the plastic mesh model. The dimension deviation information includes the deviation value and the deviation coordinates; in addition, the preset defect identification elements are obtained, and the preset defect identification elements include the point cloud distribution topology and the point cloud mesh density threshold, and correspond to the preset defect type one by one; further, the surface point cloud traversal screening and abnormal defect extraction of the plastic mesh model are performed based on the preset defect identification elements to determine the abnormal defect information, and the abnormal defect information includes the defect type, defect intensity and defect coordinates; finally, a target quality inspection report is generated based on the dimension deviation information and the abnormal defect information, and subsequent processing of the plastic object to be inspected is performed according to the target quality inspection report. By collecting images and point cloud data of multiple parts of the plastic object to be inspected, and determining the contour of the part based on image analysis; then fitting and filling the point cloud data based on the part contour to generate the part point cloud; then using the part contour as the framework and the part point cloud as the filling data fusion to build a plastic mesh model; further comparing the size deviation of the plastic mesh model to determine the size deviation value and positioning deviation coordinates; on the other hand, based on the preset defect recognition elements, the surface point cloud traversal screening and abnormal defect extraction of the plastic mesh model to determine the surface abnormal defect information, including the defect type, defect intensity and defect coordinates; finally, combining the size deviation information and the abnormal defect information to form a quality inspection report, it can improve the precision, accuracy and completeness of surface defect detection and size deviation detection of plastic products, and at the same time improve the automation and intelligence of quality inspection, thereby achieving the technical effect of improving the accuracy, comprehensiveness and efficiency of quality inspection.

[0039] Furthermore, before extracting contour information from the image data set through the convergent contour recognition channel, as shown in the attached Figure 2 As shown, step 2 of this application includes:

[0040] Taking the acquisition accuracy of the image sensor as a device constraint and the plastic object to be inspected as an object constraint, a sample identification data set is retrieved and acquired, wherein the sample identification data set includes different types of sample component images and sample identification contours in a variety of different environments;

[0041] Performing contour fitting on the sample identification contour and the sample component image, extracting sample identification data whose contour fitting degree meets the fitting degree threshold to form a sample training set;

[0042] Using the sample training set, supervised learning is performed on the contour recognition channel constructed based on the convolutional neural network, and the output accuracy is verified when a preset number of training times is met;

[0043] Determine whether the output accuracy satisfies a predetermined training constraint, and if so, set the contour recognition channel obtained at this time as the convergent contour recognition channel;

[0044] If not, the sample training set is expanded based on the style transfer algorithm to generate a new expanded training set, and the contour recognition channel is incrementally learned using the new expanded training set until a converged contour recognition channel that meets the predetermined training constraints is obtained.

[0045] Specifically, first, the acquisition accuracy of the image sensor is used as the equipment constraint, and the plastic object to be inspected is used as the object constraint. A sample identification data set is obtained based on the historical inspection log of the plastic object, and the sample identification data set includes different types of sample component images and sample identification contours under a variety of different environments. Then, the sample identification contour and the sample component image are contour fitted, and the contour fitting degree is determined based on the contour fitting result, wherein the greater the contour repetition ratio in the contour fitting result, the higher the corresponding contour fitting degree; a fitting degree threshold is obtained, and the fitting degree threshold can be set by a person skilled in the art according to the sample accuracy requirement, wherein the higher the sample accuracy requirement, the larger the fitting degree threshold; and sample identification data with a contour fitting degree greater than or equal to the fitting degree threshold are further extracted to form a sample training set.

[0046] A contour recognition channel is constructed based on a convolutional neural network, wherein the contour recognition channel is a neural network model that can be iteratively optimized in machine learning and is obtained through iterative supervised training; the contour recognition channel includes an input layer, multiple hidden layers and an output layer, wherein the input data of the input layer is a component image, and the output data is a component contour; the sample training set is used as training data, and iterative supervised learning is performed on the contour recognition channel until the current number of iterations reaches a preset number of training times, and the preset number of training times can be set by a technician in this field according to actual needs, and the output result of the contour recognition channel is verified to obtain the output accuracy.

[0047] A predetermined training constraint is obtained, wherein the predetermined training constraint is an expected output accuracy index; the output accuracy is judged based on the predetermined training constraint, and if the output accuracy is greater than or equal to the predetermined training constraint, the contour recognition channel obtained at this time is set as the convergent contour recognition channel; if the output accuracy is less than the predetermined training constraint, the sample component images in the sample training set are expanded based on the style transfer algorithm, wherein the style transfer algorithm is a technology that separates the content and style of an image and applies different styles to the same content. It can keep the content of a photo unchanged while giving it different artistic styles to obtain a newly added and expanded training set. Incremental learning is an online learning method that can learn and update new data without retraining the entire model. It has the advantages of saving time and resources, strong adaptability, and the ability to handle data imbalance problems. It enables the machine learning system to continuously learn new knowledge from new data while retaining the memory of old knowledge, thereby continuously improving its own performance. Further based on the principle of incremental learning, the contour recognition channel is supervised and trained using the newly added and expanded training set until a convergent contour recognition channel that meets the predetermined training constraints is obtained.

[0048] Further, the fitting and filling of the point cloud data streams are respectively performed with the contours of the multiple identification components as constraints to generate multiple identification component point clouds that meet the expected fitting constraints. Step three of the present application includes:

[0049] Selecting a first identified component contour from among the plurality of identified component contours, and extracting first point cloud data from the point cloud data stream based on a component mapping association;

[0050] Using the first identification component outline to perform random fitting filling in the first point cloud data to generate a first fitting identification component point cloud;

[0051] Determine whether the first fitting identification component point cloud satisfies a preliminary fitting constraint, where the preliminary fitting constraint is a threshold value of a ratio of the preliminary point cloud filled into the first identification component contour;

[0052] If satisfied, adding the first fitting identification component point cloud to the preliminary fitting identification component point cloud set;

[0053] If not, the first fitting identification component point cloud is discarded, and the first identification component outline is continued to be used to iteratively and randomly fit and fill in the first point cloud data until a preset number of fitting times is reached, and a preliminary fitting identification component point cloud set is output;

[0054] Based on the first identification component outline, random fitting and filling are performed again in the preliminary fitting identification component point cloud set to obtain a first identification component point cloud, and the first identification component point cloud is added to the multiple identification component point clouds.

[0055] Specifically, first, a first identification component contour is selected from the multiple identification component contours, the first identification component contour is any one of the multiple identification component contours, and the first point cloud data in the point cloud data stream is extracted based on the component mapping association, the first point cloud data and the first identification component contour are data of the same component.

[0056] Then, the first identification component outline is used to perform random fitting and filling in the first point cloud data, that is, the first point cloud data is randomly placed in the first identification component outline to obtain a first fitting identification component point cloud; a preliminary fitting constraint is obtained, and the preliminary fitting constraint is a preliminary point cloud ratio threshold filled into the first identification component outline. Those skilled in the art can set it according to actual conditions, for example: setting the preliminary point cloud ratio threshold to 70%; further judging whether the first point cloud ratio of the first fitting identification component point cloud is greater than or equal to the preliminary point cloud ratio threshold, and the first point cloud ratio is the ratio of the number of point clouds of the first fitting identification component point cloud to the number of point clouds of the first point cloud data.

[0057] If the first point cloud ratio is greater than or equal to the preliminary point cloud ratio threshold, the first fitting identification component point cloud is added to the preliminary fitting identification component point cloud set; if the first point cloud ratio is less than the preliminary point cloud ratio threshold, the first fitting identification component point cloud is discarded; then the first identification component contour is continued to be used to iteratively randomly fit and fill in the first point cloud data until a preset number of fits is reached, and the preset number of fits can be set based on actual conditions, and a preliminary fitting identification component point cloud set is output.

[0058] Then, based on the first identification component outline, random fitting and filling are performed again in the preliminary fitting identification component point cloud set to obtain a first identification component point cloud that satisfies the selected fitting constraints, and then the first identification component point cloud is added to multiple identification component point clouds to obtain the multiple identification component point clouds.

[0059] Furthermore, the present application also includes the following steps:

[0060] Performing point cloud fusion on the preliminary fitting identification component point cloud set to obtain first selected point cloud data;

[0061] Performing random fitting and filling in the first selected point cloud data based on the first identified component outline to generate a first selected fitting component point cloud;

[0062] Determining whether the first selected fitting component point cloud satisfies a selected fitting constraint, wherein the selected fitting constraint is a selected point cloud ratio threshold filled into the first identification component outline, and the selected point cloud ratio threshold is greater than the preliminary point cloud ratio threshold;

[0063] If satisfied, output the first selected fitting component point cloud and set it as the first identified component point cloud;

[0064] If not, iterative refined fitting is performed until the first identification component point cloud satisfying the refined fitting constraint is obtained.

[0065] Specifically, first, the preliminary fitting identification component point cloud set is point cloud fused to obtain the first selected point cloud data. Then, based on the first identification component outline, random fitting filling is performed in the first selected point cloud data to obtain the first selected fitting component point cloud; then, it is determined whether the first selected point cloud ratio of the first selected fitting component point cloud is greater than or equal to the selected fitting constraint, the first selected point cloud ratio is the ratio of the number of points of the first selected fitting component point cloud to the number of points of the first selected point cloud data, the selected fitting constraint is the selected point cloud ratio threshold filled into the first identification component outline, and the selected point cloud ratio threshold is greater than the preliminary point cloud ratio threshold, for example, the selected point cloud ratio threshold can be set to 90%. If the first selected point cloud ratio is greater than or equal to the selected fitting constraint, the first selected fitting component point cloud is set as the first identification component point cloud; if the first selected point cloud ratio is less than the selected fitting constraint, iterative selected fitting is performed until the first identification component point cloud that satisfies the selected fitting constraint is obtained.

[0066] By setting preliminary fitting constraints to preliminarily screen the fitting results and fuse them to obtain preliminary point cloud data; and by setting selected fitting constraints to accurately screen the preliminary point cloud data, the fitting speed can be improved in the early stage of point cloud fitting and filling, and the fitting accuracy can be improved in the later stage of fitting and filling, thereby improving the accuracy and efficiency of the point cloud fitting results.

[0067] Furthermore, before comparing and analyzing the plastic mesh model with the standard mesh model in the standard model library, step 4 of the present application includes:

[0068] Determine a plurality of predetermined processing indicators for a plurality of sample plastic objects, wherein the predetermined processing indicators include a plurality of component coordinates and a plurality of size tolerance thresholds, and the component coordinates correspond to the size tolerance thresholds one by one;

[0069] Modeling the plurality of sample plastic objects in a three-dimensional simulation platform to obtain a plurality of standard three-dimensional models;

[0070] Selecting a first standard three-dimensional model of a first sample plastic object and obtaining a first processing accuracy constraint;

[0071] Setting a first mesh window based on the first machining accuracy constraint, and meshing the first standard three-dimensional model based on the first mesh window to generate a first mesh model;

[0072] The first predetermined processing index of the first sample plastic object is read, and the first predetermined processing index is rendered to the first grid model to generate a first standard grid model, and the standard model library is constructed based on the first standard grid model.

[0073] Specifically, first, multiple predetermined processing indicators of multiple sample plastic objects are determined, and the predetermined processing indicators can be set based on actual conditions, wherein the predetermined processing indicators correspond to the sample plastic objects one by one; the predetermined processing indicators include multiple component coordinates and multiple size tolerance thresholds, wherein the size tolerance threshold refers to the allowable range of size deviation, wherein each component coordinate corresponds to a size tolerance threshold, and the size tolerance threshold is set based on the importance of the component coordinates, wherein the importance of the component coordinates is inversely proportional to the size tolerance threshold, that is, the greater the importance of the component coordinates, the smaller the size tolerance threshold.

[0074] The three-dimensional simulation platform is a software that can integrate various different data and realize three-dimensional simulation display and data visualization. In the three-dimensional simulation platform, the multiple sample plastic objects are subjected to three-dimensional simulation modeling to generate multiple standard three-dimensional models; further, a first standard three-dimensional model of a first sample plastic object is randomly selected from the multiple standard three-dimensional models, the first sample plastic object is any one of the multiple sample plastic objects, and a first processing accuracy constraint of the first sample plastic object is obtained, wherein the first processing accuracy constraint can be set according to production requirements.

[0075] Then, a first grid window is set based on the first processing accuracy constraint, wherein the first grid window refers to the grid size, wherein the grid is a square, and the first processing accuracy constraint is inversely proportional to the first grid window, i.e., the higher the processing accuracy requirement, the higher the requirement for the refinement of the deviation comparison analysis, and the smaller the corresponding grid size; then, the first standard three-dimensional model is grid segmented based on the first grid window to generate a first grid model. Then, a first predetermined processing index of the first sample plastic object is read, wherein the first predetermined processing index includes a plurality of first component coordinates and a plurality of first size tolerance thresholds; based on the plurality of first component coordinates, the plurality of first size tolerance thresholds are mapped and rendered to the first grid model to generate a first standard grid model. Then, a plurality of standard grid models are sequentially acquired using the same method as that for acquiring the first standard grid model, and the standard model library is constructed based on the plurality of standard grid models.

[0076] By mapping and rendering the predetermined processing indicators to the grid model to obtain a standard grid model, the automation and intelligence of the model size deviation comparison can be improved, thereby improving the efficiency and accuracy of the size deviation comparison.

[0077] Furthermore, the step 5 of obtaining the preset defect identification elements in this application includes:

[0078] Reading a preset defect type, wherein the preset defect type includes at least a bubble defect, a crack defect, and a scratch defect;

[0079] Taking the preset defect type as a constraint, obtaining a plurality of sample defect surface point cloud sets;

[0080] Selecting a first sample defect surface point cloud set of a first preset defect type from the plurality of sample defect surface point cloud sets;

[0081] Performing point cloud feature extraction on the first sample defect surface point cloud set based on a preset identification factor to generate a first extracted point cloud distribution topology set and a first extracted point cloud grid density set;

[0082] Selecting a mode in the first extracted point cloud distribution topology set as a first point cloud distribution topology, and obtaining a plurality of first extracted point cloud grid densities of the first point cloud distribution topology;

[0083] A first point cloud mesh density threshold is determined based on the multiple first extracted point cloud mesh densities, and a first preset defect identification element is generated according to the first preset defect type, the first point cloud distribution topology and the point cloud mesh density threshold mapping, and added to the preset defect identification element.

[0084] Specifically, first, the preset defect type is read, wherein the preset defect type includes at least bubble defects, crack defects and scratch defects, and those skilled in the art can also add defect types according to actual needs; then, with the preset defect type as a constraint, multiple sample defect surface point cloud sets are retrieved and obtained, wherein the sample defect surface point cloud sets and the preset defect types correspond one to one. Then, a first sample defect surface point cloud set of a first preset defect type is selected from the multiple sample defect surface point cloud sets, and the first preset defect type is any one of the preset defect types.

[0085] Obtain a preset identification factor, which includes a point cloud distribution topology and a point cloud mesh density; perform point cloud feature extraction on the first sample defect surface point cloud set based on the preset identification factor to obtain a first extracted point cloud distribution topology set and a first extracted point cloud mesh density set. Then, select the point cloud distribution topology with the highest frequency in the first extracted point cloud distribution topology set as the first point cloud distribution topology, and obtain multiple first extracted point cloud mesh densities corresponding to the first point cloud distribution topology. Further extract the minimum density value among the multiple first extracted point cloud mesh densities as the lower threshold, and use the maximum density value as the upper threshold to generate a first point cloud mesh density threshold. Then, based on the mapping association of the first preset defect type, the first point cloud distribution topology and the point cloud mesh density threshold, construct a first preset defect identification element, and add the first preset defect identification element to the preset defect identification element to obtain the preset defect identification element.

[0086] Further, the surface point cloud traversal screening and abnormal defect extraction of the plastic mesh model are performed based on the preset defect identification elements to determine abnormal defect information. Step six of this application includes:

[0087] Prioritize the preset defect identification elements based on the preset defect types to obtain defect identification element priorities;

[0088] Based on the priority of the defect identification elements, the surface point cloud of the plastic mesh model is traversed and screened in sequence. If the surface point cloud distribution characteristics meet the point cloud distribution topology and the surface point cloud distribution density meets the point cloud mesh density threshold, abnormal defect data is located. The abnormal defect data includes defect type, defect coordinates, defect area and defect point cloud density deviation. The defect point cloud density deviation is the density difference between the intermediate value of the surface point cloud distribution density and the point cloud mesh density threshold.

[0089] The defect degree analysis is performed based on the defect area and the defect point cloud density deviation to determine the defect intensity, and the abnormal defect information is obtained by combining the defect type and the defect coordinates.

[0090] Specifically, the preset defect identification elements are prioritized based on the preset defect types, wherein the greater the degree of influence of the defect type on the overall quality of the plastic object to be inspected, the higher the priority of the corresponding defect identification element, and the defect identification element priority is obtained. Then, according to the order of priority of the defect identification elements, the surface point cloud of the plastic mesh model is traversed and screened in sequence. If the surface point cloud distribution characteristics conform to the point cloud distribution topology in the defect identification element, and the surface point cloud distribution density is within the point cloud mesh density threshold in the defect identification element, then the position is located as the surface abnormal defect position, and the surface point cloud data of the surface abnormal defect position is located as the point cloud defect data; then, based on the point cloud defect data, the defect type, defect coordinates, defect area and defect point cloud density deviation are determined, wherein the defect point cloud density deviation is the density difference between the intermediate value of the surface point cloud distribution density and the point cloud mesh density threshold.

[0091] Then, the defect degree analysis is performed based on the defect area and the defect point cloud density deviation, wherein the larger the defect area and the larger the defect point cloud density deviation, the greater the defect degree and the corresponding defect intensity, and the defect intensity is determined, and finally the abnormal defect information is obtained based on the defect type, the defect coordinates and the defect intensity. By determining the defect type, defect location coordinates and defect intensity of the surface defect, the accuracy and comprehensiveness of the surface defect information can be improved.

[0092] In summary, the plastic product quality inspection method based on machine vision provided by this application has the following technical effects:

[0093] 1. Through collecting images and point cloud data of multiple parts of the plastic object to be inspected, and determining the contour of the part based on image analysis; then fitting and filling the point cloud data based on the contour of the part to generate the point cloud of the part; then taking the contour of the part as the framework and the point cloud of the part as the filling data fusion to build a plastic mesh model; further comparing the size deviation of the plastic mesh model to determine the size deviation value and positioning deviation coordinates; on the other hand, based on the preset defect recognition elements, traversing and screening the surface point cloud of the plastic mesh model and extracting abnormal defects, determining the surface abnormal defect information, including the defect type, defect intensity and defect coordinates; finally, combining the size deviation information and the abnormal defect information to form a quality inspection report, it can improve the precision, accuracy and completeness of surface defect detection and size deviation detection of plastic products, and at the same time improve the automation and intelligence of quality inspection, so as to achieve the technical effect of improving the accuracy, comprehensiveness and efficiency of quality inspection.

[0094] 2. By determining the dimensional deviation information based on the comparative analysis of the plastic mesh model and the standard mesh model, the intelligence of the dimensional deviation analysis can be improved, and the precision, accuracy and efficiency of the dimensional deviation analysis can be improved; by traversing and screening the surface point cloud and extracting abnormal defects of the plastic mesh model based on preset defect identification elements, the accuracy, comprehensiveness and intelligence of the surface abnormal defect analysis can be improved.

[0095] 3. By setting preliminary fitting constraints, the fitting results are preliminarily screened and fused to obtain preliminary point cloud data; and by setting selected fitting constraints, the preliminary point cloud data is accurately screened, which can improve the fitting speed in the early stage of point cloud fitting and filling, and improve the fitting accuracy in the later stage of fitting and filling, thereby improving the accuracy and efficiency of the point cloud fitting results.

[0096] Embodiment 2

[0097] Based on the same inventive concept as the plastic product quality detection method based on machine vision in the aforementioned embodiment, the present application also provides a plastic product quality detection system based on machine vision, please refer to the attached Figure 3 , the system comprising:

[0098] An information collection module 11, which is used to collect information on multiple parts of the plastic object to be inspected based on a three-dimensional scanner and an image sensor, and generate a point cloud data stream and an image data set;

[0099] A contour information extraction module 12, the contour information extraction module 12 is used to extract contour information from the image data set through a convergent contour recognition channel to determine contours of multiple identification components;

[0100] A plastic mesh model building module 13, the plastic mesh model building module 13 is used to respectively perform fitting filling of the point cloud data stream with the multiple identification component contours as constraints, generate multiple identification component point clouds that meet the expected fitting constraints, and build a plastic mesh model based on the fusion of the multiple identification component contours and the multiple identification component point clouds;

[0101] A size deviation information positioning module 14, which is used to compare and analyze the plastic mesh model with a standard mesh model in a standard model library to locate size deviation information, wherein the size deviation information includes a deviation value and a deviation coordinate;

[0102] A preset defect identification element acquisition module 15, wherein the preset defect identification element acquisition module 15 is used to acquire preset defect identification elements, wherein the preset defect identification elements include point cloud distribution topology and point cloud grid density threshold, and correspond to preset defect types one by one;

[0103] An abnormal defect information determination module 16, the abnormal defect information determination module 16 is used to perform surface point cloud traversal screening and abnormal defect extraction on the plastic mesh model based on the preset defect identification elements to determine abnormal defect information, the abnormal defect information including defect type, defect intensity and defect coordinates;

[0104] The target quality inspection report generating module 17 is used to generate a target quality inspection report based on the dimension deviation information and the abnormal defect information, and perform subsequent processing of the plastic object to be inspected according to the target quality inspection report.

[0105] Furthermore, the contour information extraction module 12 in the system is also used for:

[0106] Taking the acquisition accuracy of the image sensor as a device constraint and the plastic object to be inspected as an object constraint, a sample identification data set is retrieved and acquired, wherein the sample identification data set includes different types of sample component images and sample identification contours in a variety of different environments;

[0107] Performing contour fitting on the sample identification contour and the sample component image, extracting sample identification data whose contour fitting degree meets the fitting degree threshold to form a sample training set;

[0108] Using the sample training set, supervised learning is performed on the contour recognition channel constructed based on the convolutional neural network, and the output accuracy is verified when a preset number of training times is met;

[0109] Determine whether the output accuracy satisfies a predetermined training constraint, and if so, set the contour recognition channel obtained at this time as the convergent contour recognition channel;

[0110] If not, the sample training set is expanded based on the style transfer algorithm to generate a new expanded training set, and the contour recognition channel is incrementally learned using the new expanded training set until a converged contour recognition channel that meets the predetermined training constraints is obtained.

[0111] Furthermore, the plastic grid model building module 13 in the system is also used for:

[0112] Selecting a first identified component contour from among the plurality of identified component contours, and extracting first point cloud data from the point cloud data stream based on a component mapping association;

[0113] Using the first identification component outline to perform random fitting filling in the first point cloud data to generate a first fitting identification component point cloud;

[0114] Determine whether the first fitting identification component point cloud satisfies a preliminary fitting constraint, where the preliminary fitting constraint is a threshold value of a ratio of the preliminary point cloud filled into the first identification component contour;

[0115] If satisfied, adding the first fitting identification component point cloud to the preliminary fitting identification component point cloud set;

[0116] If not, the first fitting identification component point cloud is discarded, and the first identification component outline is continued to be used to iteratively and randomly fit and fill in the first point cloud data until a preset number of fitting times is reached, and a preliminary fitting identification component point cloud set is output;

[0117] Based on the first identification component outline, random fitting and filling are performed again in the preliminary fitting identification component point cloud set to obtain a first identification component point cloud, and the first identification component point cloud is added to the multiple identification component point clouds.

[0118] Furthermore, the plastic grid model building module 13 in the system is also used for:

[0119] Performing point cloud fusion on the preliminary fitting identification component point cloud set to obtain first selected point cloud data;

[0120] Performing random fitting and filling in the first selected point cloud data based on the first identified component outline to generate a first selected fitting component point cloud;

[0121] Determining whether the first selected fitting component point cloud satisfies a selected fitting constraint, wherein the selected fitting constraint is a selected point cloud ratio threshold filled into the first identification component outline, and the selected point cloud ratio threshold is greater than the preliminary point cloud ratio threshold;

[0122] If satisfied, output the first selected fitting component point cloud and set it as the first identified component point cloud;

[0123] If not, iterative refined fitting is performed until the first identification component point cloud satisfying the refined fitting constraint is obtained.

[0124] Furthermore, the size deviation information positioning module 14 in the system is also used for:

[0125] Determine a plurality of predetermined processing indicators for a plurality of sample plastic objects, wherein the predetermined processing indicators include a plurality of component coordinates and a plurality of size tolerance thresholds, and the component coordinates correspond to the size tolerance thresholds one by one;

[0126] Modeling the plurality of sample plastic objects in a three-dimensional simulation platform to obtain a plurality of standard three-dimensional models;

[0127] Selecting a first standard three-dimensional model of a first sample plastic object and obtaining a first processing accuracy constraint;

[0128] Setting a first mesh window based on the first machining accuracy constraint, and meshing the first standard three-dimensional model based on the first mesh window to generate a first mesh model;

[0129] The first predetermined processing index of the first sample plastic object is read, and the first predetermined processing index is rendered to the first grid model to generate a first standard grid model, and the standard model library is constructed based on the first standard grid model.

[0130] Furthermore, the preset defect identification element acquisition module 15 in the system is also used for:

[0131] Reading a preset defect type, wherein the preset defect type includes at least a bubble defect, a crack defect, and a scratch defect;

[0132] Taking the preset defect type as a constraint, obtaining a plurality of sample defect surface point cloud sets;

[0133] Selecting a first sample defect surface point cloud set of a first preset defect type from the plurality of sample defect surface point cloud sets;

[0134] Performing point cloud feature extraction on the first sample defect surface point cloud set based on a preset identification factor to generate a first extracted point cloud distribution topology set and a first extracted point cloud grid density set;

[0135] Selecting a mode in the first extracted point cloud distribution topology set as a first point cloud distribution topology, and obtaining a plurality of first extracted point cloud grid densities of the first point cloud distribution topology;

[0136] A first point cloud mesh density threshold is determined based on the multiple first extracted point cloud mesh densities, and a first preset defect identification element is generated according to the first preset defect type, the first point cloud distribution topology and the point cloud mesh density threshold mapping, and added to the preset defect identification element.

[0137] Furthermore, the abnormal defect information determination module 16 in the system is also used for:

[0138] Prioritize the preset defect identification elements based on the preset defect types to obtain defect identification element priorities;

[0139] Based on the priority of the defect identification elements, the surface point cloud of the plastic mesh model is traversed and screened in sequence. If the surface point cloud distribution characteristics meet the point cloud distribution topology and the surface point cloud distribution density meets the point cloud mesh density threshold, abnormal defect data is located. The abnormal defect data includes defect type, defect coordinates, defect area and defect point cloud density deviation. The defect point cloud density deviation is the density difference between the intermediate value of the surface point cloud distribution density and the point cloud mesh density threshold.

[0140] The defect degree analysis is performed based on the defect area and the defect point cloud density deviation to determine the defect intensity, and the abnormal defect information is obtained by combining the defect type and the defect coordinates.

[0141] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The plastic product quality inspection method based on machine vision and the specific examples in the aforementioned embodiment 1 are also applicable to the plastic product quality inspection system based on machine vision in this embodiment. Through the aforementioned detailed description of the plastic product quality inspection method based on machine vision, those skilled in the art can clearly understand the plastic product quality inspection system based on machine vision in this embodiment, so for the sake of brevity of the specification, it will not be described in detail here. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0142] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

[0143] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application belong to the scope of the present application and its equivalent technology, the present application is also intended to include these modifications and variations.

Claims

1. A plastic product quality inspection method based on machine vision, characterized in that: The method is applied to a plastic product quality detection system based on machine vision, the system is communicatively connected with a three-dimensional scanner and an image sensor, and the method comprises: Based on 3D scanners and image sensors, information of multiple parts of the inspected plastic object is collected to generate point cloud data streams and image data sets; Extracting contour information from the image data set through a convergent contour recognition channel to determine contours of multiple identification components; Taking the multiple identification component contours as constraints, respectively performing fitting filling of the point cloud data streams to generate multiple identification component point clouds that meet the expected fitting constraints, and building a plastic mesh model based on the fusion of the multiple identification component contours and the multiple identification component point clouds; Compare and analyze the plastic mesh model with a standard mesh model in a standard model library to locate size deviation information, wherein the size deviation information includes a deviation value and a deviation coordinate; Obtaining preset defect identification elements, wherein the preset defect identification elements include point cloud distribution topology and point cloud grid density threshold, and correspond one to one with preset defect types; Based on the preset defect identification elements, the surface point cloud of the plastic mesh model is traversed and screened and abnormal defects are extracted to determine abnormal defect information, wherein the abnormal defect information includes defect type, defect intensity and defect coordinates; A target quality inspection report is generated based on the size deviation information and the abnormal defect information, and subsequent processing of the plastic object to be inspected is performed according to the target quality inspection report.

2. The method according to claim 1, characterized in that The extracting of contour information from the image data set through a convergent contour recognition channel also includes: Taking the acquisition accuracy of the image sensor as a device constraint and the plastic object to be inspected as an object constraint, a sample identification data set is retrieved and acquired, wherein the sample identification data set includes different types of sample component images and sample identification contours in a variety of different environments; Performing contour fitting on the sample identification contour and the sample component image, extracting sample identification data whose contour fitting degree meets the fitting degree threshold to form a sample training set; Using the sample training set, supervised learning is performed on the contour recognition channel constructed based on the convolutional neural network, and the output accuracy is verified when a preset number of training times is met; Determine whether the output accuracy satisfies a predetermined training constraint, and if so, set the contour recognition channel obtained at this time as the convergent contour recognition channel; If not, the sample training set is expanded based on the style transfer algorithm to generate a new expanded training set, and the contour recognition channel is incrementally learned using the new expanded training set until a converged contour recognition channel that meets the predetermined training constraints is obtained.

3. The method according to claim 1, characterized in that The step of performing fitting filling of the point cloud data streams respectively based on the contours of the multiple identification components as constraints to generate multiple identification component point clouds satisfying expected fitting constraints includes: Selecting a first identified component contour from among the plurality of identified component contours, and extracting first point cloud data from the point cloud data stream based on a component mapping association; Using the first identification component outline to perform random fitting filling in the first point cloud data to generate a first fitting identification component point cloud; Determine whether the first fitting identification component point cloud satisfies a preliminary fitting constraint, where the preliminary fitting constraint is a threshold value of a ratio of the preliminary point cloud filled into the first identification component contour; If satisfied, adding the first fitting identification component point cloud to the preliminary fitting identification component point cloud set; If not, the first fitting identification component point cloud is discarded, and the first identification component outline is continued to be used to iteratively and randomly fit and fill in the first point cloud data until a preset number of fitting times is reached, and a preliminary fitting identification component point cloud set is output; Based on the first identification component outline, random fitting and filling are performed again in the preliminary fitting identification component point cloud set to obtain a first identification component point cloud, and the first identification component point cloud is added to the multiple identification component point clouds.

4. The method according to claim 3, characterized in that The step of performing random fitting and filling again in the preliminary fitting identification component point cloud set based on the first identification component outline to obtain a first identification component point cloud comprises: Performing point cloud fusion on the preliminary fitting identification component point cloud set to obtain first selected point cloud data; Performing random fitting and filling in the first selected point cloud data based on the first identified component outline to generate a first selected fitting component point cloud; Determining whether the first selected fitting component point cloud satisfies a selected fitting constraint, wherein the selected fitting constraint is a selected point cloud ratio threshold filled into the first identification component outline, and the selected point cloud ratio threshold is greater than the preliminary point cloud ratio threshold; If satisfied, output the first selected fitting component point cloud and set it as the first identified component point cloud; If not, iterative refined fitting is performed until the first identification component point cloud satisfying the refined fitting constraint is obtained.

5. The method according to claim 1, characterized in that The comparing and analyzing the plastic mesh model with the standard mesh model in the standard model library also includes: Determine a plurality of predetermined processing indicators for a plurality of sample plastic objects, wherein the predetermined processing indicators include a plurality of component coordinates and a plurality of size tolerance thresholds, and the component coordinates correspond to the size tolerance thresholds one by one; Modeling the plurality of sample plastic objects in a three-dimensional simulation platform to obtain a plurality of standard three-dimensional models; Selecting a first standard three-dimensional model of a first sample plastic object and obtaining a first processing accuracy constraint; Setting a first mesh window based on the first machining accuracy constraint, and meshing the first standard three-dimensional model based on the first mesh window to generate a first mesh model; The first predetermined processing index of the first sample plastic object is read, and the first predetermined processing index is rendered to the first grid model to generate a first standard grid model, and the standard model library is constructed based on the first standard grid model.

6. The method according to claim 1, characterized in that The obtaining of preset defect identification elements includes: Reading a preset defect type, wherein the preset defect type includes at least a bubble defect, a crack defect, and a scratch defect; Taking the preset defect type as a constraint, obtaining a plurality of sample defect surface point cloud sets; Selecting a first sample defect surface point cloud set of a first preset defect type from the plurality of sample defect surface point cloud sets; Performing point cloud feature extraction on the first sample defect surface point cloud set based on a preset identification factor to generate a first extracted point cloud distribution topology set and a first extracted point cloud grid density set; Selecting a mode in the first extracted point cloud distribution topology set as a first point cloud distribution topology, and obtaining a plurality of first extracted point cloud grid densities of the first point cloud distribution topology; A first point cloud mesh density threshold is determined based on the multiple first extracted point cloud mesh densities, and a first preset defect identification element is generated according to the first preset defect type, the first point cloud distribution topology and the point cloud mesh density threshold mapping, and added to the preset defect identification element.

7. The method according to claim 6, characterized in that The step of performing surface point cloud traversal screening and abnormal defect extraction on the plastic mesh model based on the preset defect identification elements to determine abnormal defect information includes: Prioritize the preset defect identification elements based on the preset defect types to obtain defect identification element priorities; Based on the priority of the defect identification elements, the surface point cloud of the plastic mesh model is traversed and screened in sequence. If the surface point cloud distribution characteristics meet the point cloud distribution topology and the surface point cloud distribution density meets the point cloud mesh density threshold, abnormal defect data is located. The abnormal defect data includes defect type, defect coordinates, defect area and defect point cloud density deviation. The defect point cloud density deviation is the density difference between the intermediate value of the surface point cloud distribution density and the point cloud mesh density threshold. The defect degree analysis is performed based on the defect area and the defect point cloud density deviation to determine the defect intensity, and the abnormal defect information is obtained by combining the defect type and the defect coordinates.

8. The plastic product quality inspection system based on machine vision is characterized by: For implementing the steps of the method according to any one of claims 1 to 7, the system is communicatively connected with a three-dimensional scanner and an image sensor, and the system comprises: An information collection module, the information collection module is used to collect information of multiple parts of the plastic object to be inspected based on a three-dimensional scanner and an image sensor, and generate a point cloud data stream and an image data set; A contour information extraction module, the contour information extraction module is used to extract contour information from the image data set through a convergent contour recognition channel to determine contours of multiple identification components; A plastic mesh model building module, the plastic mesh model building module is used to respectively perform fitting filling of the point cloud data stream with the multiple identification component contours as constraints, generate multiple identification component point clouds that meet the expected fitting constraints, and build a plastic mesh model based on the fusion of the multiple identification component contours and the multiple identification component point clouds; A size deviation information positioning module, the size deviation information positioning module is used to compare and analyze the plastic mesh model with a standard mesh model in a standard model library to locate size deviation information, the size deviation information including a deviation value and a deviation coordinate; A preset defect identification element acquisition module, which is used to acquire preset defect identification elements, wherein the preset defect identification elements include point cloud distribution topology and point cloud grid density threshold, and correspond one to one with preset defect types; An abnormal defect information determination module, the abnormal defect information determination module is used to perform surface point cloud traversal screening and abnormal defect extraction on the plastic mesh model based on the preset defect identification elements, and determine abnormal defect information, wherein the abnormal defect information includes defect type, defect intensity and defect coordinates; A target quality inspection report generation module is used to generate a target quality inspection report based on the size deviation information and the abnormal defect information, and perform subsequent processing of the plastic object to be inspected according to the target quality inspection report.

Images