A quality visual inspection method and system for window gear injection mold

Abstract

The present invention relates to the technical field of visual inspection of gear injection molds, and specifically to a method and system for visual inspection of the quality of a vehicle window gear injection mold. The present invention selects a reference gear mold image through a gear mold visual analysis model, further extracts a first gear mold image feature and a second gear mold image feature for reference based on these image data and the gear mold visual analysis model, performs image feature analysis on the gear mold to be inspected according to the same model, and compares the first gear mold image feature and the second gear mold image feature to screen the degree of damage of the gear mold to be inspected. Through this method, the feature data of the gear mold to be inspected can be quantified based on the image data, so as to understand the degree of damage of different gear molds and facilitate subsequent maintenance of the gear mold.

Description

Technical Field

[0001] The invention relates to the technical field of visual inspection of gear injection molds, and in particular to a quality visual inspection method and system for a vehicle window gear injection mold. Background Art

[0002] The window gear is an important component of the window lifting system. In order to support the long-term use of the window lifting system, the durability of the window gear becomes the most important part. With the development of technology, today's window gears have been converted from metal gears to plastic gears. Compared with metal gears, plastic gears have better durability and do not require a lot of cost in gear lubrication. Therefore, plastic gears are widely used in window lifting systems in the current automotive field.

[0003] Plastic gears are important components of the window lift system, so the precision of plastic gears has an important impact on the normal use of the window lift system. The industrialized high-speed production of plastic gears is inseparable from the injection mold. Therefore, the quality of the injection mold determines the precision of the plastic gears. With the increase of use, the mold will be damaged, and the quality of the mold will definitely be affected. Therefore, the quality inspection of the plastic gear injection mold is a key step to establish the precision of the plastic gear.

[0004] Due to the different degrees of damage to gear molds, it is impossible to determine the impact of mold damage on the actual quality of plastic gears after production. Some damage may not require too high a cost to maintain, and some damage may even have very little impact on the actual plastic gear quality. However, the existing technology is unable to detect and determine the specific degree of damage to the gear mold.

[0005] Therefore, a quality visual inspection method and system for window gear injection mold are proposed. Summary of the invention

[0006] The purpose of the present invention is to provide a method and system for visual inspection of the quality of a vehicle window gear injection mold. A reference gear mold image is screened out through a gear mold visual analysis model, and a first gear mold image feature and a second gear mold image feature for reference are further extracted based on these image data and the gear mold visual analysis model. Image feature analysis is performed on the gear mold to be inspected according to the same model, and the first gear mold image feature and the second gear mold image feature are compared to screen the degree of damage of the gear mold to be inspected. This method can quantify the feature data of the gear mold to be inspected based on the image data, so as to understand the degree of damage of different gear molds and facilitate subsequent maintenance of the gear mold.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A quality visual inspection method for a window gear injection mold, comprising:

[0009] Acquire a historical window gear injection mold, establish a gear mold visual analysis model to screen the historical window gear injection mold to obtain a reference gear mold image;

[0010] Further, the reference gear mold image is obtained according to the mold image feature area screening model of the gear mold visual analysis model; the mold image feature area screening model includes a global mold image segmentation unit, a global mold image processing unit, a global mold image feature recognition unit and a global mold image feature weight output unit;

[0011] The global mold image segmentation unit performs image segmentation on the damaged gear mold image, and performs image feature marking on the segmented damaged gear mold image; the global mold image processing unit performs image preprocessing on the segmented damaged gear mold image;

[0012] The global mold image feature recognition unit recognizes the damaged gear mold image after the image preprocessing, and adds a damage classification mark to the image feature mark for identifying damage; the global mold image feature recognition unit includes 3 Transformer layers, 3 convolutional layers, 4 dilated convolutional layers, 4 residual convolutional layers, 2 dimensionality reduction activation layers, 1 fully connected layer and 1 classification layer; specifically:

[0013] The damaged gear mold image is input into the first layer of the Transformer layer, and the Transformer layer is connected to the first layer of the convolution layer; the first layer of the convolution layer is connected to the four layers of the dilated convolution layer and the first layer of the Transformer layer to identify the global image features of the damaged gear mold image;

[0014] The first convolution layer is connected to the second convolution layer; the second convolution layer is connected to the four residual convolution layers and the first Transformer layer to identify the local image key features of the damaged gear mold image; the global image features and the local image key features are first image feature fused through the first dimension reduction activation layer;

[0015] The second convolution layer is connected to the third convolution layer; the third convolution layer outputs the standard image key features; the output result of the fusion of the standard image key features and the first image features is subjected to the second image feature fusion through the dimensionality reduction activation layer of the first layer; the output result of the second image feature fusion is input into the full leveling layer and the classification layer to obtain the damage classification mark;

[0016] The global mold image feature weight output unit counts the damage classification mark of the image feature mark, adds the damage weight of the image feature mark according to the damage classification mark, and outputs the result;

[0017] Extracting a first gear mold image feature and a second gear mold image feature of the reference gear mold image according to the gear mold visual analysis model, and storing the first gear mold image feature and the second gear mold image feature; the reference gear mold image includes a normal gear mold image and a damaged gear mold image, the first gear mold image feature is obtained according to the normal gear mold image; the second gear mold image feature is obtained according to the damaged gear mold image;

[0018] Further, a normal gear mold image is obtained, and the global mold image segmentation unit and the global mold image processing unit of the mold image feature area screening model are used to perform image segmentation and image preprocessing on the normal gear mold image; the image segmentation size is consistent with the image segmentation size of the damaged gear mold image, and the image preprocessing method is consistent with the image preprocessing method of the damaged gear mold image; the image feature markers of the normal gear mold image after segmentation and the image feature markers of the damaged gear mold image after segmentation correspond one to one;

[0019] According to the damage weight of the image feature mark of the damaged gear mold image, the normal gear mold image corresponding to the image feature mark is obtained, the corresponding normal gear mold image is identified by the global mold image feature recognition unit of the mold image feature area screening model, and the first gear mold image feature is obtained for storage;

[0020] The image contour feature unit of the gear mold visual analysis model secondarily identifies the damaged gear mold image containing the image feature mark of the damage weight in the damaged gear mold image, and obtains the second gear mold image feature for storage; the image contour feature unit adopts the DeepLSD algorithm;

[0021] Extracting a third gear mold image feature of the vehicle window gear mold image to be detected according to the gear mold visual analysis model; comparing the first gear mold image feature with the third gear mold image feature, and if the first gear mold image feature and the third gear mold image feature are different; marking the corresponding vehicle window gear mold image; the third gear mold image feature is obtained based on the mold image feature area screening model of the gear mold visual analysis model, and the fourth gear mold image feature is obtained based on the image contour feature unit of the gear mold visual analysis model;

[0022] The fourth gear mold image feature of the marked vehicle window gear mold image is further extracted according to the gear mold visual analysis model; the gear mold visual analysis model calls the second gear mold image feature to compare the fourth gear mold image feature, and if the comparison similarity between the fourth gear mold image feature and the second gear mold image feature is less than the image feature threshold, the vehicle window gear mold image is marked as an injection mold to be maintained, and a reference gear mold image corresponding to the second gear mold image feature is output as a maintenance reference comparison image;

[0023] Furthermore, the following is calculated based on the image contour feature comparison unit of the gear mold visual analysis model:

[0024] ;

[0025] in, is the comparison similarity, is the point contour coordinate of the fourth gear mold image feature, is the point contour coordinate of the second gear mold image feature, is the point contour coordinate error distance, is the point contour coordinate distance threshold, is the total number of line contours, is the line profile coordinate of the i-th image feature of the fourth gear mold, is the line profile coordinate of the i-th feature of the second gear mold image, is the line profile coordinate error distance, is the line contour coordinate distance threshold, and Calculate weights for comparing similarities.

[0026] The present invention also provides a visual quality inspection system for a vehicle window gear injection mold, comprising:

[0027] A gear mold image screening module is used to obtain historical window gear injection molds, establish a gear mold visual analysis model, and screen the historical window gear injection molds to obtain reference gear mold images;

[0028] a gear mold feature recognition and storage module, which extracts a first gear mold image feature and a second gear mold image feature of the reference gear mold image according to the gear mold visual analysis model, and stores the first gear mold image feature and the second gear mold image feature;

[0029] The gear mold feature damaged area comparison module extracts the third gear mold image feature of the window gear mold image to be detected according to the gear mold visual analysis model; compares the first gear mold image feature with the third gear mold image feature, and if the first gear mold image feature and the third gear mold image feature are different; marks the corresponding window gear mold image;

[0030] The gear mold feature damage detail comparison module continues to extract the fourth gear mold image feature of the marked window gear mold image according to the gear mold visual analysis model; the gear mold visual analysis model calls the second gear mold image feature to compare the fourth gear mold image feature. If the comparison similarity between the fourth gear mold image feature and the second gear mold image feature is less than the image feature threshold, the window gear mold image is marked as an injection mold to be maintained, and a reference gear mold image corresponding to the second gear mold image feature is output as a maintenance reference comparison image.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] 1. The present invention obtains the damage weights of each image feature mark after the damaged gear mold image is segmented by establishing a mold image feature area screening model of the gear mold visual analysis model, and determines the severity of each damaged position of the damaged injection mold; this method can perform data quantification on the damage degree of the gear mold during use to determine and screen out the positions that are prone to damage, so as to reduce the amount of unnecessary image detection data, and at the same time provide more accurate quantitative area data for subsequent damage degree detection, providing a data basis for subsequent damage degree detection.

[0033] 2. The present invention screens the image segmentation position corresponding to the normal gear mold image through the damage weight of the image feature mark of the damaged gear mold image, and extracts the normal gear mold image features marked by the corresponding image data through the gear mold visual analysis model; this method can perform feature comparison based on the corresponding normal gear mold image to achieve rapid identification in the detection process, improve the detection efficiency of the degree of damage, and provide data basis for subsequent further identification of the degree of damage and maintenance of damaged molds that have a greater impact on gear precision.

[0034] 3. The image features of the damaged gear mold are further refined through the DeepLSD algorithm. The image to be detected, which has different features from the normal gear mold image, is refined by the DeepLSD algorithm and the feature coordinates are calculated together with the refined image features of the damaged gear mold. This method can quantify the damaged part of the image to be detected, so that the damage condition of the current gear mold can be obtained through the quantified similarity data, so as to screen out the gear molds with more serious damage and the gear molds with less damage, which is convenient for subsequent maintenance work and provides an accurate data basis for the maintenance work of damaged molds that have a greater impact on the precision of gears. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 is a flow chart of the method of the present invention;

[0036] Figure 2 A virtual device diagram of a mold image feature region screening model, an image contour feature unit and an image contour feature comparison unit of a gear mold visual analysis model of the present invention;

[0037] Figure 3 A network structure diagram of a global mold image feature recognition unit of the present invention;

[0038] Figure 4 The point contour coordinates and line contour coordinates diagram identified by DeepLSD of the present invention;

[0039] Figure 5 It is a system flow chart of the present invention. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] Plastic gears are the main components of the current window lifting system. The precision of plastic gears has always been the main factor in maintaining the window lifting system, and the precision of plastic gears is closely related to the gear injection mold. With the increase in the number of uses, the gear mold will also be damaged due to various reasons and need maintenance, but the degree of damage to the injection mold varies. Some damage has a low impact on the precision of the gear. If unified maintenance is performed, it is easy to increase the maintenance cost. However, the existing technology cannot accurately detect and judge the degree of damage to the mold. To this end, the present invention provides a quality visual inspection method and system for window gear injection molds. The technical solution is as follows:

[0042] Acquire a historical window gear injection mold, establish a gear mold visual analysis model to screen the historical window gear injection mold to obtain a reference gear mold image;

[0043] extracting a first gear mold image feature and a second gear mold image feature of the reference gear mold image according to the gear mold visual analysis model, and storing the first gear mold image feature and the second gear mold image feature;

[0044] Extracting a third gear mold image feature of the window gear mold image to be inspected according to the gear mold visual analysis model; comparing the first gear mold image feature with the third gear mold image feature, and if the first gear mold image feature and the third gear mold image feature are different; marking the corresponding window gear mold image;

[0045] The fourth gear mold image feature of the marked vehicle window gear mold image continues to be extracted according to the gear mold visual analysis model; the gear mold visual analysis model calls the second gear mold image feature to compare the fourth gear mold image feature; if the comparison similarity between the fourth gear mold image feature and the second gear mold image feature is less than the image feature threshold, the vehicle window gear mold image is marked as an injection mold to be maintained, and a reference gear mold image corresponding to the second gear mold image feature is output as a maintenance reference comparison image.

[0046] By establishing a gear mold visual analysis model, the image data of damaged molds in historical window gear injection molds are segmented and feature recognized, and the location information that is easily damaged during the production process is confirmed from a large amount of data. This method can quantify the damage data in the entire production process based on the existing data, so as to determine the location information of the damaged mold, facilitate subsequent data quantification and data feature comparison, and provide a relatively accurate damaged area location information basis for subsequent damage degree detection;

[0047] Based on the gear mold visual analysis model, the first gear mold image feature of the normal gear mold image is extracted, and at the same time, the feature is matched with the previous damaged position information, and further image feature extraction is performed based on the damaged area of ​​the damaged mold. Based on the above content, normal gear mold image data with more corresponding damaged areas and more accurate position areas can be provided. Through these data contents, the detection efficiency can be improved during the detection, and image comparison feature data can also be provided for subsequent damage degree detection, thereby improving the accuracy of damage degree data quantification and providing a more accurate data basis for gear mold damage degree detection.

[0048] The damage information of the image area of ​​the gear mold to be inspected is screened according to the image features of the normal gear mold, and the damage degree of the image damage features of the gear mold to be inspected is quantitatively calculated according to the damage features of the damaged gear mold. This method can quantify the data of the damage degree based on the gear mold to be inspected, and it is convenient to generate quantitative damage degree data from the image feature information, so as to automatically analyze the damage degree of the gear mold to be inspected and perform maintenance work on the damaged mold that has a greater impact on the gear precision.

[0049] Embodiment 1

[0050] For the purpose of specific explanation, the following contents are combined for elaboration:

[0051] Obtain a historical window gear injection mold, establish a gear mold visual analysis model to screen the historical window gear injection mold to obtain a reference gear mold image; the image data of all reference gear mold images are consistent in the shooting angle and resolution of the image data;

[0052] Further, the reference gear mold image is obtained according to the mold image feature area screening model of the gear mold visual analysis model; the mold image feature area screening model includes a global mold image segmentation unit, a global mold image processing unit, a global mold image feature recognition unit and a global mold image feature weight output unit; ... Figure 2 As shown, considering that in the actual production process of plastic gears, the gear mold has damage conditions in different positions, some positions are damaged more than others, and the impact on the actual gear production is also different. Therefore, by establishing a recognition model for damaged molds, through a large amount of historical image data of damaged molds, the damaged position information of the damaged molds can be quantified based on these image data, so as to facilitate the subsequent image data feature screening work, and provide an accurate quantitative data basis for the detection and maintenance of damaged molds that have a greater impact on the precision of gears;

[0053] The global mold image segmentation unit performs image segmentation on the damaged gear mold image, and performs image feature marking on the segmented damaged gear mold image; a unified image segmentation standard is adopted for all damaged gear mold images; the damaged gear mold images are all screened out based on expert experience and have a significant impact on the precision of subsequent plastic gear production;

[0054] Image feature marking is based on the image number and the number of image segmentations. For example, there are two images of broken gear molds, numbered A and B. The images are divided into four parts using cross segmentation. The image feature markings of image A are A00, A01, A10 and A11, and the image feature markings of image B are B00, B01, B10 and B11. The actual image feature marking method can be changed according to the actual situation.

[0055] The global mold image processing unit performs image preprocessing on the segmented damaged gear mold image; the image preprocessing includes image denoising, histogram augmentation and image edge sharpening using a Roberts operator;

[0056] The global mold image feature recognition unit recognizes the damaged gear mold image after the image preprocessing, and adds a damage classification mark to the image feature mark of the identified damage; the damage classification mark label is divided into 0 and 1, 0 represents that there is no damaged image that will not affect the gear precision in the image, and 1 represents a damaged image that affects the gear precision;

[0057] The global mold image feature recognition unit includes 3 Transformer layers, 3 convolutional layers, 4 dilated convolutional layers, 4 residual convolutional layers, 2 dimension reduction activation layers, 1 fully connected layer and 1 classification layer; specifically:

[0058] The damaged gear mold image is input into the first layer of the Transformer layer, and the Transformer layer is connected to the first layer of the convolution layer; the first layer of the convolution layer is connected to the four layers of the dilated convolution layer and the first layer of the Transformer layer to identify the global image features of the damaged gear mold image;

[0059] The first convolution layer is connected to the second convolution layer; the second convolution layer is connected to the four residual convolution layers and the first Transformer layer to identify the local image key features of the damaged gear mold image; the global image features and the local image key features are first image feature fused through the first dimension reduction activation layer;

[0060] The second convolution layer is connected to the third convolution layer; the third convolution layer outputs the standard image key features; the output result of the fusion of the standard image key features and the first image features is subjected to the second image feature fusion through the dimensionality reduction activation layer of the first layer; the output result of the second image feature fusion is input into the full leveling layer and the classification layer to obtain the damage classification mark;

[0061] The convolution layers are all 3*3 convolution layers, the dimension reduction activation layers are downsampling layers and relu layers, and the classification layers are Softmax layers; refer to Figure 3As shown, the Transformer layer corresponds to Figure 3 The transformer layer in the Transformer layer; through the attention mechanism of the Transformer layer, the key position information in the image can be accurately identified. At the same time, the dilated convolution layer can extract global features of the image data based on global information, extract local feature information in the image data based on the residual convolution layer, and perform feature fusion of the data through the dimension reduction activation layer. Through this pyramid-like structure, data can be extracted based on global image features, local image features and key image features to ensure the accuracy of data classification, and provide an accurate quantitative data basis for subsequent image data feature screening and the detection and maintenance of damaged molds that have a greater impact on gear precision;

[0062] The global mold image feature weight output unit counts the damage classification mark of the image feature mark, adds the damage weight of the image feature mark according to the damage classification mark, and outputs it; the damage weight is calculated as follows:

[0063] ;

[0064] in, is the damage weight, ln is the logarithmic function with e as the base, Num is the quantity statistics, A damage classification tag is provided for the model to identify damage anomalies; is the input segmented image;

[0065] Extracting the first gear mold image feature and the second gear mold image feature of the reference gear mold image according to the gear mold visual analysis model, and storing the first gear mold image feature and the second gear mold image feature; the reference gear mold image includes a normal gear mold image and a damaged gear mold image, the first gear mold image feature is obtained according to the normal gear mold image; the second gear mold image feature is obtained according to the damaged gear mold image; by using the normal gear mold image and the damaged gear mold image in combination, the existing gear mold image data can be utilized to the greatest extent, and the recognition errors existing in the model can be avoided by using the normal gear mold image, and the model recognition efficiency of the mold image can be accelerated, and the speed of mold detection can be improved, so as to provide an efficient data image basis for the subsequent detection and maintenance of damaged molds that have a greater impact on the precision of the gear;

[0066] Further, a normal gear mold image is obtained, and the global mold image segmentation unit and the global mold image processing unit of the mold image feature area screening model are used to perform image segmentation and image preprocessing on the normal gear mold image; the image segmentation size is consistent with the image segmentation size of the damaged gear mold image, and the image preprocessing method is consistent with the image preprocessing method of the damaged gear mold image; the image feature mark of the normal gear mold image after segmentation and the image feature mark of the damaged gear mold image after segmentation correspond to each other one by one; for example, the damaged gear mold image is segmented into 4 areas, namely X01, X02, X03 and X04, X is the damaged gear mold number, 01, 02, 03 and 04 are the segmentation area numbers, so the mark of the normal gear mold image is also Y01, Y02, Y03 and Y04, Y is the normal gear mold number, 01, 02, 03 and 04 are the segmentation area numbers; the corresponding numbers must satisfy the consistency of the segmentation area numbers, and the normal gear mold image must also adopt the same segmentation scheme, and the segmentation method and segmentation size adopted are the same;

[0067] According to the damage weight of the image feature mark of the damaged gear mold image, the normal gear mold image corresponding to the image feature mark is obtained, the corresponding normal gear mold image is identified by the global mold image feature recognition unit of the mold image feature area screening model, and the first gear mold image feature is obtained for storage;

[0068] The image contour feature unit of the gear mold visual analysis model secondly identifies the damaged gear mold image containing the image feature mark of the damage weight in the damaged gear mold image, obtains the second gear mold image feature for storage; the image contour feature unit adopts the DeepLSD algorithm; referring to Figure 2 As shown;

[0069] Based on the identified damaged gear mold image features, the image is labeled and the quantified damaged position information is mapped onto the normal gear mold image. This method can realize the rapid recognition and processing of the normal gear mold image based on the quantified position information, which is convenient for improving the efficiency of the subsequent detection process. At the same time, the secondary feature refinement recognition based on the damaged gear mold image features can also provide a data basis for mold comparison and recognition verification for subsequent detection work, and provide more refined damaged image data features for the subsequent screening of damaged molds that have a greater impact on gear precision. Inspection and maintenance work;

[0070] Extract the third gear mold image feature of the window gear mold image to be detected according to the gear mold visual analysis model; compare the first gear mold image feature with the third gear mold image feature, if the first gear mold image feature and the third gear mold image feature are not the same; mark the corresponding window gear mold image; the third gear mold image feature is obtained based on the mold image feature area screening model of the gear mold visual analysis model, and the fourth gear mold image feature is obtained based on the image contour feature unit of the gear mold visual analysis model; wherein the feature similarity between the first gear mold image feature and the third gear mold image feature is compared using image cosine similarity; the reuse of model parameters can avoid the recognition error caused by different model parameters to the greatest extent, and at the same time can improve the reuse degree of the model and increase its robustness, and can also reduce the cost of training and building a new model and adjusting model parameters; provide a more accurate model data basis for subsequent model recognition and screening of the detection and maintenance of gear precision molds;

[0071] The fourth gear mold image feature of the marked window gear mold image is further extracted according to the gear mold visual analysis model; the gear mold visual analysis model calls the second gear mold image feature to compare the fourth gear mold image feature; if the comparison similarity between the fourth gear mold image feature and the second gear mold image feature is less than the image feature threshold, the window gear mold image is marked as an injection mold to be maintained, and a reference gear mold image corresponding to the second gear mold image feature is output as a maintenance reference comparison image; the image feature threshold is set based on expert experience;

[0072] Further, refer to Figure 2 As shown, the calculation is performed based on the image contour feature comparison unit of the gear mold visual analysis model:

[0073] ;

[0074] in, is the comparison similarity, is the point contour coordinate of the fourth gear mold image feature, is the point contour coordinate of the second gear mold image feature, is the point contour coordinate error distance, is the point contour coordinate distance threshold, is the total number of line contours, is the line profile coordinate of the i-th image feature of the fourth gear mold, is the line profile coordinate of the i-th feature of the second gear mold image, is the line profile coordinate error distance, is the line contour coordinate distance threshold, and Calculate weights for comparing similarities; and The default settings are 0.5 and 0.5; the point contour coordinates and line contour coordinates refer to Figure 4 As shown in the figure; the line contour coordinates are calculated based on the coordinates of the corresponding pixels occupied by the line contour, and the point contour coordinates and line contour coordinates are calculated using the Euclidean distance for coordinate difference calculation; the point contour coordinates and line contour coordinates will calculate the similarity between the image contour features of the gear mold image to be detected and the image contour features of all stored damaged gear mold images, and the maximum contrast similarity is used as the contrast similarity result of the gear mold image to be detected;

[0075] Based on the identified damaged gear mold image, the damage data is quantified through the DeepLSD algorithm, and detailed data comparison is performed based on the identified point contour coordinates and line contour coordinates. Based on the quantified position recognition features, a more detailed damage data comparison can be performed on the damage features, and the damage features that have a significant impact on the precision of plastic gears can be screened out and extracted through quantitative data features, so as to facilitate subsequent mold inspection and maintenance work.

[0076] Embodiment 2

[0077] The present invention also provides a visual inspection system for the quality of a window gear injection mold, referring to Figure 5 As shown, including:

[0078] A gear mold image screening module is used to obtain historical window gear injection molds, establish a gear mold visual analysis model, and screen the historical window gear injection molds to obtain reference gear mold images;

[0079] In this embodiment, 6 groups of data were selected based on expert experience, and the recognition accuracy of the damaged gear mold image, which has a greater impact on the gear precision, was identified by the gear mold visual analysis model. The 6 groups of data were not trained, and the final recognition results are shown in Table 1:

[0080] Table 1 Accuracy of the gear mold visual analysis model in identifying damaged mold images with a greater impact on gear precision

[0081]

[0082] From Table 1, we can see that the recognition accuracy of the six groups of data is above 90%, and the recognition accuracy is relatively high;

[0083] a gear mold feature recognition and storage module, which extracts a first gear mold image feature and a second gear mold image feature of the reference gear mold image according to the gear mold visual analysis model, and stores the first gear mold image feature and the second gear mold image feature;

[0084] The gear mold feature damaged area comparison module extracts the third gear mold image feature of the window gear mold image to be detected according to the gear mold visual analysis model; compares the first gear mold image feature with the third gear mold image feature, and if the first gear mold image feature and the third gear mold image feature are different; marks the corresponding window gear mold image;

[0085] The gear mold feature damage detail comparison module continues to extract the fourth gear mold image feature of the marked window gear mold image according to the gear mold visual analysis model; the gear mold visual analysis model calls the second gear mold image feature to compare the fourth gear mold image feature. If the comparison similarity between the fourth gear mold image feature and the second gear mold image feature is less than the image feature threshold, the window gear mold image is marked as an injection mold to be maintained, and a reference gear mold image corresponding to the second gear mold image feature is output as a maintenance reference comparison image.

[0086] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A visual inspection method for the quality of a window gear injection mold, characterized in that: include: Obtain historical window gear injection molds, establish a gear mold visual analysis model to screen historical window gear injection molds to obtain reference gear mold images; Extracting first gear mold image features and second gear mold image features of the reference gear mold image according to the gear mold visual analysis model, and storing the first gear mold image features and the second gear mold image features; Extracting a third gear mold image feature of the window gear mold image to be inspected according to the gear mold visual analysis model; Compare the first gear mold image feature with the third gear mold image feature, and if the first gear mold image feature and the third gear mold image feature are different, mark the corresponding window gear mold image; The fourth gear mold image feature of the marked window gear mold image is further extracted according to the gear mold visual analysis model; the gear mold visual analysis model calls the second gear mold image feature to compare the fourth gear mold image feature. If the comparison similarity between the fourth gear mold image feature and the second gear mold image feature is less than the image feature threshold, the window gear mold image is marked as an injection mold to be maintained, and the reference gear mold image corresponding to the second gear mold image feature is output as a maintenance reference comparison image; the comparison similarity is calculated based on the image contour feature comparison unit of the gear mold visual analysis model, and the specific formula is: ； in, To compare similarity, is the point contour coordinate of the fourth gear mold image feature, is the point contour coordinate of the second gear mold image feature, is the point contour coordinate error distance, is the point contour coordinate distance threshold, is the total number of line contours, is the line profile coordinate of the i-th fourth gear mold image feature, is the line profile coordinate of the i-th second gear mold image feature, is the line profile coordinate error distance, is the line contour coordinate distance threshold, and Calculate weights for comparing similarities.

2. The method for visually inspecting the quality of a window gear injection mold according to claim 1, characterized in that: Establishing a gear mold visual analysis model to screen the historical window gear injection mold to obtain a reference gear mold image includes: The reference gear mold image is obtained according to the mold image feature area screening model of the gear mold visual analysis model; the mold image feature area screening model includes a global mold image segmentation unit, a global mold image processing unit, a global mold image feature recognition unit and a global mold image feature weight output unit; The global mold image segmentation unit performs image segmentation on the damaged gear mold image, and performs image feature marking on the segmented damaged gear mold image; the global mold image processing unit performs image preprocessing on the segmented damaged gear mold image; The global mold image feature recognition unit recognizes the damaged gear mold image after the image preprocessing, and adds a damaged classification mark to the image feature mark that identifies the damage; the global mold image feature weight output unit counts the damage classification mark of the image feature mark, adds the damage weight of the image feature mark according to the damage classification mark, and outputs it.

3. The quality visual inspection method of a window gear injection mold according to claim 2 is characterized in that: The global mold image feature recognition unit recognizes the damaged gear mold image after the image preprocessing, and adds a damage classification mark to the image feature mark for identifying damage, including: the global mold image feature recognition unit includes 3 Transformer layers, 3 convolution layers, 4 expansion convolution layers, 4 residual convolution layers, 2 dimensionality reduction activation layers, 1 fully connected layer and 1 classification layer, including: The damaged gear mold image is input into the first layer of the Transformer layer, and the Transformer layer is connected to the first layer of the convolution layer; the first layer of the convolution layer is connected to the four layers of the dilated convolution layer and the first layer of the Transformer layer to identify the global image features of the damaged gear mold image; The first convolution layer is connected to the second convolution layer; the second convolution layer is connected to the four residual convolution layers and the first Transformer layer to identify the local image key features of the damaged gear mold image; the global image features and the local image key features are first image feature fused through the first dimension reduction activation layer; The second convolution layer is connected to the third convolution layer; the third convolution layer outputs standard image key features; the output result of the fusion of the standard image key features and the first image features is subjected to a second image feature fusion through the dimensionality reduction activation layer of the first layer; the output result of the second image feature fusion is input into the full leveling layer and the classification layer to obtain the damage classification mark.

4. The method for visually inspecting the quality of a window gear injection mold according to claim 1, characterized in that: The reference gear mold image includes a normal gear mold image and a damaged gear mold image, and the first gear mold image feature is acquired based on the normal gear mold image; the second gear mold image feature is acquired based on the damaged gear mold image.

5. The method for visually inspecting the quality of a window gear injection mold according to claim 1, characterized in that: Extracting a first gear mold image feature and a second gear mold image feature of the reference gear mold image according to the gear mold visual analysis model, and storing the first gear mold image feature and the second gear mold image feature, including: A normal gear mold image is obtained, and the global mold image segmentation unit and the global mold image processing unit of the mold image feature area screening model are used to perform image segmentation and image preprocessing on the normal gear mold image; the image segmentation size is consistent with the image segmentation size of the damaged gear mold image, and the image preprocessing method is consistent with the image preprocessing method of the damaged gear mold image; the image feature marks of the normal gear mold image after segmentation and the image feature marks of the damaged gear mold image after segmentation correspond one to one; According to the damage weight of the image feature mark of the damaged gear mold image, the normal gear mold image corresponding to the image feature mark is obtained, the corresponding normal gear mold image is identified by the global mold image feature recognition unit of the mold image feature area screening model, and the first gear mold image feature is obtained for storage; The image contour feature unit of the gear mold visual analysis model secondary identifies the damaged gear mold image containing the image feature mark of the damage weight in the damaged gear mold image, obtains the second gear mold image feature for storage; the image contour feature unit adopts the DeepLSD algorithm.

6. The method for visually inspecting the quality of a window gear injection mold according to claim 1, characterized in that: The third gear mold image feature is acquired based on a mold image feature area screening model of the gear mold visual analysis model, and the fourth gear mold image feature is acquired based on an image contour feature unit of the gear mold visual analysis model.

7. A visual inspection system for the quality of a window gear injection mold, characterized in that: include: A gear mold image screening module is used to obtain historical window gear injection molds, establish a gear mold visual analysis model, and screen the historical window gear injection molds to obtain reference gear mold images; a gear mold feature recognition and storage module, which extracts a first gear mold image feature and a second gear mold image feature of the reference gear mold image according to the gear mold visual analysis model, and stores the first gear mold image feature and the second gear mold image feature; The gear mold feature damaged area comparison module extracts the third gear mold image feature of the window gear mold image to be detected according to the gear mold visual analysis model; compares the first gear mold image feature with the third gear mold image feature, and if the first gear mold image feature and the third gear mold image feature are different; marks the corresponding window gear mold image; a gear mold feature damage detail comparison module, which continues to extract a fourth gear mold image feature of the marked window gear mold image according to the gear mold visual analysis model; The gear mold visual analysis model calls the second gear mold image feature to compare with the fourth gear mold image feature. If the comparison similarity between the fourth gear mold image feature and the second gear mold image feature is less than the image feature threshold, the window gear mold image is marked as an injection mold to be maintained, and a reference gear mold image corresponding to the second gear mold image feature is output as a maintenance reference comparison image; the comparison similarity is calculated based on the image contour feature comparison unit of the gear mold visual analysis model, and the specific formula is: ； in, is the comparison similarity, is the point contour coordinate of the fourth gear mold image feature, is the point contour coordinate of the second gear mold image feature, is the point contour coordinate error distance, is the point contour coordinate distance threshold, is the total number of line contours, is the line profile coordinate of the i-th image feature of the fourth gear mold, is the line profile coordinate of the i-th feature of the second gear mold image, is the line profile coordinate error distance, is the line contour coordinate distance threshold, and Calculate weights for comparing similarities.

Images