Inverter housing gluing quality monitoring system and method based on visual inspection

Abstract

The present application relates to the field of intelligent detection, and provides an inverter housing gluing quality monitoring system and method based on visual detection, which uses computer vision-based image processing technology to perform ROI extraction on the inverter housing gluing status monitoring image, and then performs image cutting and gluing status detection on the extracted gluing area ROI image, and then performs local activity feature selection on the glue line segment state features obtained after each detection, so as to intelligently judge whether the gluing consistency meets the preset requirements based on the global fine-grained difference representation between the selected glue line segment state features. In this way, by analyzing the ROI image of the gluing area, the waste of computing resources can be greatly reduced, and it is also beneficial to improve the detection efficiency. At the same time, the subtle features of the glue line can be accurately identified, thereby achieving a more detailed gluing quality assessment.

Description

Technical Field

[0001] The present application relates to the field of intelligent detection, and more specifically, to a system and method for monitoring the gluing quality of an inverter housing based on visual detection. Background Art

[0002] As modern manufacturing industry continues to improve its requirements for product quality, especially in key areas such as new energy and automotive electronics, inverters, as core components of power conversion, must be strictly controlled in every link of their manufacturing process. The gluing process of the inverter housing is one of the key steps to ensure the sealing, waterproofness and durability of the equipment.

[0003] However, when processing the inverter housing gluing status monitoring image, traditional image detection technology often processes the entire image uniformly, lacking targeted analysis of the gluing area, resulting in a large amount of computing resources being wasted on background areas that are irrelevant to the gluing quality, which not only reduces the detection efficiency, but is also easily interfered by background noise, affecting the accuracy of gluing quality detection. In addition, when detecting the gluing status, traditional methods usually only judge whether there are obvious defects in the glue line from a macroscopic level, such as serious breakpoints or a large amount of glue overflow. For the subtle features of the glue line, such as small cracks and bubbles, microscopic width changes, and surface texture differences, traditional methods lack sufficient sensitivity and precision, making it difficult to achieve accurate detection.

[0004] Therefore, an optimized inverter housing glue coating quality monitoring solution is desired. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present application provides a system and method for monitoring the gluing quality of an inverter housing based on visual inspection.

[0006] According to one aspect of the present application, a method for monitoring the gluing quality of an inverter housing based on visual inspection is provided, which comprises: acquiring an inverter housing gluing status monitoring image acquired by an industrial camera; performing ROI extraction on the inverter housing gluing status monitoring image to obtain a gluing area ROI image; performing image cutting on the gluing area ROI image along the glue line to obtain a set of glue line segment ROI images; performing gluing status detection on each glue line segment ROI image in the set of glue line segment ROI images to obtain a set of glue line segment status feature coding maps; performing gluing on each glue line segment status feature coding map in the set of glue line segment status feature coding maps. The line segment state feature is sparsely processed to obtain a set of glue line segment state feature coding vectors, including: feature decoupling and feature flattening of the glue line segment state feature coding map to obtain a set of glue line segment state local feature vectors; local importance calculation and neighborhood activity feature selection are performed on the set of glue line segment state local feature vectors to obtain the glue line segment state feature coding vector; glue line segment state difference semantic measurement is performed on the set of glue line segment state feature coding vectors to obtain a global fine-grained difference coding representation vector of the glue line; based on the global fine-grained difference coding representation vector of the glue line, a quality monitoring result is obtained to indicate whether the glue coating consistency meets the preset requirements.

[0007] According to another aspect of the present application, a visual inspection-based inverter housing gluing quality monitoring system is provided, which includes: an inverter housing gluing status data acquisition module, which is used to obtain an inverter housing gluing status monitoring image acquired by an industrial camera; an inverter housing gluing area ROI extraction module, which is used to perform ROI extraction on the inverter housing gluing status monitoring image to obtain a gluing area ROI image; an inverter housing gluing area ROI image cutting module, which is used to perform image cutting on the gluing area ROI image along the glue line to obtain a set of glue line segment ROI images; an inverter housing gluing status detection module, which is used to perform gluing status detection on each glue line segment ROI image in the set of glue line segment ROI images to obtain a set of glue line segment status feature coding maps; a glue line segment status feature sparse coding module, which is used to encode each glue line segment status feature in the set of glue line segment status feature coding maps. The glue line segment state feature is sparsely processed on the graph to obtain a set of glue line segment state feature coding vectors, wherein the glue line segment state feature sparse coding module includes: a glue line segment state feature decoupling and flattening unit, which is used to perform feature decoupling and feature flattening on the glue line segment state feature coding graph to obtain a set of glue line segment state local feature vectors; a glue line segment state local feature selection unit, which is used to perform local importance calculation and neighborhood activity feature selection on the set of glue line segment state local feature vectors to obtain the glue line segment state feature coding vector; a glue line segment state feature difference measurement module, which is used to perform glue line segment state difference semantic measurement on the set of glue line segment state feature coding vectors to obtain a global fine-grained difference coding representation vector for glue lines; and a glue coating quality monitoring result generation module, which is used to obtain a quality monitoring result indicating whether the glue coating consistency meets the preset requirements based on the global fine-grained difference coding representation vector for glue lines.

[0008] The present application has significant technical effects due to the adoption of the above technical solutions: the inverter housing gluing quality monitoring system and method based on visual detection provided by the present application uses computer vision-based image processing technology to perform ROI extraction on the inverter housing gluing status monitoring image, and then performs image cutting and gluing status detection on the extracted gluing area ROI image, and then performs local activity feature selection on the glue line segment state features obtained after each detection, so as to intelligently judge whether the gluing consistency meets the preset requirements based on the global fine-grained difference representation between the selected glue line segment state features. In this way, by analyzing the ROI image of the gluing area, the waste of computing resources can be greatly reduced, and it is also beneficial to improve the detection efficiency. At the same time, the subtle features of the glue line can be accurately identified, thereby achieving a more detailed gluing quality assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0010] Figure 1 The present invention is a flowchart of a method for monitoring inverter housing glue coating quality based on visual inspection according to an embodiment of the present application.

[0011] Figure 2 Schematic diagram of data flow of a method for monitoring inverter housing glue coating quality based on visual inspection according to an embodiment of the present application.

[0012] Figure 3 This is a flow chart of step S5 of the inverter housing glue coating quality monitoring method based on visual inspection according to an embodiment of the present application.

[0013] Figure 4 This is a flow chart of step S52 of the inverter housing glue coating quality monitoring method based on visual inspection according to an embodiment of the present application.

[0014] Figure 5 The system block diagram of the inverter housing glue coating quality monitoring system based on visual detection according to an embodiment of the present application. DETAILED DESCRIPTION

[0015] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0016] As the requirements for product quality in modern manufacturing continue to increase, especially in key areas such as new energy and automotive electronics, inverters, as core components of power conversion, need to be strictly controlled in every link of their production process. The gluing process of the inverter housing is one of the key steps to ensure the sealing, waterproofness and durability of the equipment.

[0017] However, when monitoring the gluing status of the inverter housing, traditional image detection technology usually processes the entire image uniformly, lacking targeted analysis of the gluing area. This approach not only results in a large amount of computing resources being wasted on background areas that are irrelevant to the gluing quality, but also reduces the detection efficiency. It is also easily interfered by background noise, which in turn affects the accuracy of gluing quality detection. In addition, traditional detection methods usually only identify obvious defects of the glue line from a macroscopic level, such as severe breakpoints or significant glue overflow, but lack sufficient sensitivity and detection accuracy for subtle features of the glue line, such as tiny cracks, bubbles, subtle changes in width, and differences in surface texture, making it difficult to achieve high-precision detection.

[0018] In view of the above technical problems, the technical concept of the present application is to obtain the inverter housing gluing state monitoring image collected by the industrial camera, use the image recognition and analysis technology based on computer vision to extract the ROI of the inverter housing gluing state monitoring image, then perform image cutting and gluing state detection on the extracted ROI image of the gluing area, and then perform local activity feature selection on the glue line segment state features obtained after each detection, so as to intelligently judge whether the gluing consistency meets the preset requirements according to the global fine-grained difference representation between the selected glue line segment state features. The present application can accurately extract the gluing area (ROI) from the entire image, avoiding the invalid calculation of irrelevant background areas. This not only greatly reduces the waste of computing resources, but also speeds up the detection speed and improves the overall efficiency. In addition, it can accurately identify subtle features such as fine cracks, bubbles, microscopic width changes and surface texture differences. This greatly exceeds the ability of traditional methods to judge obvious defects only at the macro level, and provides a more comprehensive and detailed quality assessment.

[0019] Figure 1 The present invention is a flowchart of a method for monitoring inverter housing glue coating quality based on visual inspection according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of the inverter housing gluing quality monitoring method based on visual inspection according to an embodiment of the present application. Figure 1 and Figure 2As shown, according to the embodiment of the present application, the inverter housing gluing quality monitoring method based on visual detection includes: S1, acquiring the inverter housing gluing status monitoring image collected by an industrial camera; S2, performing ROI extraction on the inverter housing gluing status monitoring image to obtain a gluing area ROI image; S3, performing image cutting on the gluing area ROI image along the glue line to obtain a set of glue line segment ROI images; S4, performing glue status detection on each glue line segment ROI image in the set of glue line segment ROI images to obtain a set of glue line segment status feature coding maps; S5, performing glue line segment status feature sparseness on each glue line segment status feature coding map in the set of glue line segment status feature coding maps to obtain a set of glue line segment status feature coding vectors; S6, performing glue line segment status difference semantic measurement on the set of glue line segment status feature coding vectors to obtain a global fine-grained difference coding representation vector for the glue line; S7, based on the global fine-grained difference coding representation vector for the glue line, obtaining a quality monitoring result for indicating whether the glue consistency meets the preset requirements.

[0020] In step S1, the inverter housing gluing state monitoring image collected by the industrial camera is obtained. It should be understood that the inverter housing gluing state monitoring image contains information on the position and shape of the glue line, the width of the glue line, the surface texture of the glue line, etc. In addition, it will also include information on the appearance and surrounding environment of the inverter housing. Specifically, the glue area information in the inverter housing gluing state monitoring image, such as the position, shape, width, etc. of the glue line, is the basis for directly judging the consistency of the glue. By analyzing the image, it can be determined whether the glue is in the specified position, whether it meets the shape requirements, and whether the width is up to standard. The surface feature information of the glue, such as bubbles, cracks, and surface texture, further supplements the basis for judging the consistency of the glue. Even if the position and width of the glue line meet the requirements, but there are a large number of bubbles or obvious cracks, it also indicates that the glue process is unstable, and the quality monitoring results will still reflect poor consistency of the glue.

[0021] The following is a detailed description of a specific implementation process of "obtaining inverter housing glue status monitoring images collected by industrial cameras": First, selecting an industrial camera suitable for the application scenario is the first step in the entire image acquisition process. Considering the needs of inverter housing glue detection, high-resolution and high-speed cameras are usually required to capture clear and rapidly changing glue processes. Such cameras can provide sufficient pixel density to show the subtle features of the glue line, while having the ability to respond quickly to ensure that no important moments are missed. In addition to the camera itself, the choice of lens is equally important. The right lens can ensure the quality of the image, for example, the choice of focal length should take into account the working distance and field of view. In order to obtain more detailed image information, a macro lens may also be required to magnify the glue area to better identify defects such as fine cracks and bubbles. In addition, lighting conditions are also one of the important factors affecting image quality. Proper light source arrangement can reduce shadows and enhance contrast, making the glue area more obvious. In glue detection, LED ring lights or coaxial lights are often used. Both light sources can effectively illuminate the target area and provide a clear picture for subsequent image analysis.

[0022] After selecting the right hardware, the next step is to install it in a stable working environment. In order to ensure that the angle and distance of each shot are consistent, the industrial camera should be firmly fixed on a bracket and keep its position relative to the object to be measured (inverter housing) unchanged. Such a setting helps to eliminate errors caused by camera movement and ensure image consistency and reliability. At the same time, it is also very important to maintain a constant working environment. This includes controlling factors such as temperature and humidity to avoid imaging errors caused by environmental changes. For example, too high a temperature may cause the internal components of the camera to heat up, thereby affecting the image quality; while excessive humidity may cause the lens to fog or produce water vapor condensation, interfering with the imaging effect. Therefore, when designing a glue coating inspection workstation, it is often considered to be equipped with an air conditioning system or a dehumidifier to ensure the best working environment. In addition, dust prevention measures cannot be ignored. Dust not only affects the cleanliness of the camera lens, but may also be deposited on the surface of the inverter housing, interfering with the uniformity of the glue coating. To this end, air filtration equipment can be set up around the work area, and the work surface can be cleaned regularly to ensure a dust-free environment.

[0023] After completing the above preparations, the next step is to accurately adjust the parameters of the industrial camera to ensure that it can capture the ideal image. Exposure time is a key parameter that determines the amount of light received by each pixel. If the exposure time is too short, the image may appear dim and blurry; conversely, if the exposure time is too long, overexposure may occur and detail information may be lost. Therefore, it is necessary to adjust the exposure time according to the specific light intensity to find a balance. The gain value is also a parameter that needs attention. Proper adjustment of the gain can increase the image brightness, but be careful not to introduce too much noise. Because too high gain will cause the signal-to-noise ratio to decrease and affect the image quality. The frame rate refers to how many frames of images the camera can capture per second, which is particularly important for real-time detection on the production line. Determine the frame rate of the camera according to the speed of the production line, so as to keep up with the production rhythm without losing important information. The choice of trigger mode should not be ignored either. Synchronously triggering image acquisition through PLC (programmable logic controller) or other sensor signals can ensure that each picture corresponds to a specific product state and achieve accurate positioning.

[0024] Finally, in order to smoothly transmit the acquired images to the subsequent processing system and conduct efficient image analysis on this basis, the interface design must be done well. This not only involves the selection of physical connection methods, such as GigEVision, USB3 Vision, Camera Link and other interface types, but also involves software-level integration. In addition, in order to meet the real-time requirements on the production line, it is also necessary to optimize the image transmission path and ensure low-latency data exchange. This means not only ensuring that the network bandwidth is sufficient to support the rapid transmission of large amounts of data, but also ensuring that the image processing speed is fast enough and does not become a bottleneck in production. For example, image processing tasks can be accelerated through a distributed computing architecture, or GPU acceleration technology can be used to improve the efficiency of algorithm operation.

[0025] In step S2, ROI extraction is performed on the inverter housing gluing state monitoring image to obtain a ROI image of the gluing area. Specifically, in an embodiment of the present application, the step S2 includes: performing ROI extraction based on the U-Net network on the inverter housing gluing state monitoring image to obtain the ROI image of the gluing area. It should be understood that the inverter housing gluing state monitoring image contains a large amount of background information that is not related to the gluing quality detection, such as production equipment, work platform, etc. These background information will interfere with the subsequent feature extraction and analysis. Therefore, in order to focus the subsequent image processing and analysis on the gluing area, the present application performs ROI extraction based on the U-Net network on the inverter housing gluing state monitoring image to obtain the ROI image of the gluing area. Those of ordinary skill in the art should know that the U-Net network was originally designed for medical image segmentation, and performs well in tasks such as distinguishing between targets and backgrounds and extracting specific areas. The glued area of ​​the inverter housing can be seen as the target area that needs to be segmented in the image. The U-Net network can effectively learn the feature differences between the glued area and other parts of the inverter housing and the background, so as to accurately segment the glued area as the ROI. In this way, after removing irrelevant background information, the interference of background factors on the glue quality detection results can be effectively avoided. For example, the reflection of other objects in the background may be misjudged as a glue defect. After extracting the ROI image, only the glued area is focused on, which can more accurately detect the width, continuity, and whether there are bubbles of the glue line and other glue state characteristics, thereby improving the accuracy of glue quality detection.

[0026] The following is a detailed description of a specific implementation process of "U-Net network-based ROI extraction of the inverter housing gluing status monitoring image to obtain the gluing area ROI image": First, in order to achieve ROI extraction based on the U-Net network, a high-quality data set must be prepared for training and verifying the model. This data set should contain a large number of inverter housing gluing status monitoring images, and each image must be manually annotated to clearly indicate the location of the gluing area. These annotation information is crucial for supervised learning, as they guide the U-Net network to learn to distinguish the difference between the gluing area and other background elements. In addition, the data set should also cover as many scene changes as possible, such as different lighting conditions, viewing angles, and various possible types of gluing defects, so as to enhance the generalization ability of the model. In the process of preparing data, it is also necessary to pay attention to the balance of the data to avoid excessive representation of a certain type of sample and affect the learning effect of the model.

[0027] With the appropriate dataset, the next step is to build a U-Net network model. U-Net is a classic convolutional neural network architecture that was originally designed for biomedical image segmentation tasks. It gradually reduces the spatial size of the image through a series of downsampling layers while increasing the number of feature maps; it then restores the spatial resolution through an upsampling layer, and finally outputs a prediction mask that is consistent with the size of the input image. In this embodiment, the structural parameters of U-Net can be adjusted according to the actual needs of inverter housing glue coating detection, such as changing the convolution kernel size, the number of filters, or adding jump connections, etc., in order to better adapt to specific application scenarios. In addition, considering that there may be computing resource limitations in industrial environments, while ensuring accuracy, you can try to use a lightweight version of U-Net, such as UNet++ or Tiny-U-Net, to reduce model complexity and increase reasoning speed.

[0028] Next is the model training phase. In this process, the U-Net network is trained using the prepared data set so that it can automatically identify the glued area. During training, a loss function is defined to measure the gap between the model output and the true label, and the back-propagation algorithm is used to continuously update the network weights until convergence. Commonly used loss functions include cross entropy loss, Dice coefficient loss, and their combinations. In order to prevent overfitting, various strategies such as regularization terms, data enhancement techniques, and early stopping mechanisms can also be introduced. It is worth noting that since the glued area usually only occupies a small part of the entire image, there may be a problem of class imbalance. At this time, weighted loss or resampling methods can be used to compensate. As the number of training times increases, the U-Net network will gradually learn to accurately locate and extract the glued area from the complex background.

[0029] After the model training is completed, the next step is to apply it to the actual glue coating status monitoring image. When a new image is input into the trained U-Net network, the network will generate a binary prediction mask, where 1 represents the pixel point belonging to the glue coating area, and 0 represents the pixel point not belonging to the glue coating area. However, directly using the original prediction results may have some noise interference, such as blurred edges or sporadic isolated small areas. In order to solve these problems, post-processing methods such as morphological operations (such as expansion, corrosion) and connected domain analysis can be used to further optimize the results of ROI extraction. For example, the area threshold can be set to remove those connected domains that are too small and illogical, so as to ensure that the final ROI image of the glue coating area is clean and coherent.

[0030] Finally, in order to evaluate the effect of ROI extraction, a reasonable evaluation index system needs to be established. Common evaluation criteria include intersection over union (IoU), pixel accuracy, F1 score, etc. Quantitative analysis of a large number of test samples can not only intuitively reflect the performance level of the model, but also provide a reference for subsequent improvements. If it is found that certain types of gluing defects are difficult to correctly identify, more relevant cases can be collected and added to the training set to retrain the model until a satisfactory ROI extraction accuracy is achieved.

[0031] In step S3, the ROI image of the glue-coated area is cut along the glue line to obtain a set of ROI images of glue line segments. It should be understood that the glue line, as a key carrier of glue coating quality, may have different states in different parts. For example, the starting and ending positions of the glue line, the turning points, etc. are more prone to glue coating defects than the straight line segments. Based on this, the present application cuts the ROI image of the glue-coated area along the glue line to obtain a set of ROI images of glue line segments. In this way, it is possible to perform a refined analysis of each part of the glue line, and to understand in detail the glue coating state of different parts of the glue line, such as whether there are defects such as breakpoints, uneven thickness, bubbles, etc., to avoid ignoring local defects due to overall analysis.

[0032] In step S4, the glue state detection is performed on each glue line segment ROI image in the set of glue line segment ROI images to obtain a set of glue line segment state feature coding maps. Specifically, in an embodiment of the present application, the step S4 includes: using a glue state detector based on the Mobile-Former model to perform a glue state detection on each glue line segment ROI image in the set of glue line segment ROI images to obtain a set of glue line segment state feature coding maps. Accordingly, it is considered that glue coating defects may appear at different scales, such as fine cracks or larger glue overflow phenomena. Therefore, in order to pay attention to both the local details in the image and the global structure, the present application uses a glue state detector based on the Mobile-Former model to perform a glue state detection on each glue line segment ROI image in the set of glue line segment ROI images to effectively capture multi-scale features from local details to global structures, and obtain a set of glue line segment state feature coding maps. It should be understood that the Mobile-Former model combines the MobileNet lightweight convolutional network with the Transformer architecture. The MobileNet part uses operations such as deep separable convolution to efficiently extract local features while reducing the amount of calculation. The Transformer part, with its self-attention mechanism, can model global information to fully explore the complex relationship between these features, thereby more accurately representing the gluing status of the glue line segment. Compared with traditional convolutional neural networks, it can perform better in feature extraction and improve the accuracy of gluing status detection. This combination pays attention to local details and global structures at the same time, so that both tiny cracks and larger glue overflows can be accurately captured in its feature extraction process. This ensures that no important information is missed during the gluing status detection process, providing a comprehensive and reliable basis for subsequent accurate judgment of whether there are defects in the gluing and the type of defects.

[0033] In step S5, the glue line state feature sparseness is performed on each glue line state feature coding map in the set of glue line state feature coding maps to obtain a set of glue line state feature coding vectors. Specifically, Figure 3 FIG. 5 is a flow chart of step S5 of the inverter housing gluing quality monitoring method based on visual inspection according to an embodiment of the present application. Figure 3 As shown, the step S5 includes: S51, performing feature decoupling and feature flattening on the glue line segment state feature coding map to obtain a set of glue line segment state local feature vectors; S52, performing local importance calculation and neighborhood activity feature selection on the set of glue line segment state local feature vectors to obtain the glue line segment state feature coding vector.

[0034] It should be understood that the glue line segment state feature coding map contains rich information, but there may be a large number of redundant or highly correlated features, which may not be very critical for subsequent processing. Therefore, in order to avoid the overfitting problem caused by too many similar features and make the model more concise and efficient, the present application performs glue line segment state feature sparsification on each glue line segment state feature coding map in the set of glue line segment state feature coding maps to obtain a set of glue line segment state feature coding vectors. In particular, this processing method can identify redundant features in the glue line segment that are individually important but highly correlated with the neighborhood, and remove them to improve model efficiency. At the same time, the relationship between features and neighborhoods is comprehensively considered to accurately capture complex patterns such as overflow and cracks, thereby obtaining a set of features that retain key information and have lower dimensions.

[0035] Specifically, the feature encoding graph of the glue line state is firstly subjected to feature decoupling and feature flattening to obtain a set of local feature vectors of the glue line state. The above process can be expressed as: ;in, is the characteristic coding diagram of the glue line segment state, is a feature decoupling operation, and is the first, second, and third in the set of glue segment state feature encoding matrices and The glue line segment state feature encoding matrix, is the feature flattening operation, and are the first, second, and third local feature vectors in the set of glue segment states. and The local feature vector of the glue segment state.

[0036] It should be understood that in the glue line segment state feature coding map, different features may be intertwined and coupled together. For example, the width feature of the glue coating may be mixed with the texture feature of the glue coating surface in the coding map, which makes it difficult for the model to distinguish and utilize these features separately. Through feature decoupling, the glue line segment state feature coding map is decomposed into a set of glue line segment state feature coding matrices, which can separate these mixed features. Each local feature matrix focuses on a specific glue line segment local feature, making the features clearer and independent, and convenient for subsequent processing. After feature flattening, the set of glue line segment state feature coding matrices is converted into a set of glue line segment state local feature vectors, so that each glue line segment state local feature vector represents a different feature dimension, and these dimensions are interrelated and relatively independent, which can lay the foundation for subsequent feature selection.

[0037] Specifically, Figure 4FIG. 5 is a flow chart of step S52 of the inverter housing gluing quality monitoring method based on visual inspection according to an embodiment of the present application. Figure 4 As shown, the step S52 includes: S521, performing local importance calculation on each glue line segment state local feature vector in the set of glue line segment state local feature vectors to obtain a set of glue line segment state local feature importance score values; S522, based on the set of glue line segment state local feature importance score values, arranging the set of glue line segment state local feature vectors in descending order to obtain a descending sequence of glue line segment state local feature vectors; S523, performing neighborhood activity feature selection on the descending sequence of the glue line segment state local feature vectors to obtain a descending sequence of the selected glue line segment state local feature vectors; S524, cascading the descending sequence of the selected glue line segment state local feature vectors to obtain the glue line segment state feature encoding vector.

[0038] Next, the local importance of each glue line segment state local feature vector in the set of glue line segment state local feature vectors is calculated to obtain a set of glue line segment state local feature importance score values. The above process can be expressed as: ;in, is the first in the set of local feature vectors of the glue segment state The local feature vector of the glue segment state, and They are The corresponding weight matrix and bias vector, is matrix multiplication, is the weight score vector, is the set of local feature importance scores of the glue segment state, and They are the first, second, and third in the set of local feature importance scores of the glue line state. and The importance score of the local feature of each glue segment state.

[0039] It should be understood that in the set of local feature vectors of the glue line segment state, each local feature vector of the glue line segment state contributes differently to the judgment of whether the glue coating consistency meets the preset requirements. By calculating the importance score of the local feature of the glue line segment state, it is possible to quantify the importance of each local feature vector to the overall task (judging whether the glue coating consistency meets the preset requirements), and to clarify which features play a key role in judging the glue coating quality and which are relatively minor. In other words, the importance score of the local feature of the glue line segment state can enable the model to focus on key features, remove minor features that may interfere with the judgment, and then more accurately learn the intrinsic connection between glue coating consistency and features, thereby achieving more accurate glue coating consistency judgment.

[0040] Then, based on the set of importance scores of the local features of the glue line segment states, the set of local feature vectors of the glue line segment states is arranged in descending order to obtain a descending sequence of local feature vectors of the glue line segment states. The above process can be expressed as: ;in, Based on The set of local feature vectors of the glue line segment state is arranged in descending order, and They are the first, second, and third in the descending sequence of the local eigenvectors of the glue segment state. and The local feature vector of the glue segment state.

[0041] It should be understood that in the face of a large number of local feature vectors of glue line segment states, it is difficult to effectively select the most valuable features for judging glue consistency without a clear distinction in importance. In this application, based on the set of importance score values ​​of the local features of the glue line segment states, the set of local feature vectors of the glue line segment states is sorted in descending order, and the local features of the glue line segment states can be screened according to their importance, which can make the feature selection process more scientific and efficient, and avoid the waste of computing resources and degradation of model performance caused by blindly selecting features.

[0042] More specifically, in the embodiment of the present application, the step S523 includes: based on the neighborhood features of each glue line state local feature vector in the descending sequence of the glue line state local feature vector, calculating the feature neighborhood activity of each glue line state local feature vector to obtain a sequence of glue line state local feature neighborhood activity. This process can be expressed as: ;in, yes Middle The eigenvalues ​​at the positions, yes The number of eigenvalues ​​in , and They are and The corresponding weighted average eigenvalue of the glue segment state, yes The corresponding glue segment state local feature neighborhood activity.

[0043] Based on the sequence of neighborhood activity of the local feature of the glue line state, feature selection is performed on the descending sequence of the local feature vector of the glue line state to obtain the descending sequence of the selected local feature vector of the glue line state. This process can be expressed as: ;in, represents feature selection of the descending sequence of the local feature vectors of the glue segment state, and They are the first, second, and third in the descending sequence of the local feature vectors of the selected glue line segment state. and The local feature vector of the glue segment state after selection.

[0044] More specifically, in an embodiment of the present application, based on the sequence of the local feature neighborhood activity of the glue line segment state, feature selection is performed on the descending sequence of the local feature vectors of the glue line segment state to obtain the descending sequence of the local feature vectors of the glue line segment state after the selection, including: comparing the local feature neighborhood activity of the glue line segment state corresponding to each local feature vector of the glue line segment state in the descending sequence of the local feature vectors of the glue line segment state with a preset threshold to obtain the descending sequence of the local feature vectors of the glue line segment state after the selection; wherein, in response to the local feature neighborhood activity of the glue line segment state corresponding to the local feature vector of the glue line segment state being greater than or equal to the preset threshold, the local feature vector of the glue line segment state remains unchanged; in response to the local feature neighborhood activity of the glue line segment state corresponding to the local feature vector of the glue line segment state being less than the preset threshold, the local feature vector of the glue line segment state is set to zero. The above process can be expressed as: ;in, For Perform feature selection, is the descending sequence of the local eigenvectors of the glue segment state The local feature vector of the glue segment state, is the preset threshold, yes The corresponding glue line state local feature neighborhood activity, is the descending sequence of the local eigenvectors of the selected glue line segment state The local feature vector of the glue segment state after selection.

[0045] It should be understood that when judging the consistency of glue coating, the importance of the local feature vector of a single glue line state is crucial, but the relationship between features cannot be ignored. Analyzing only based on the importance of a single feature may ignore the synergy between features. For example, the glue line width feature is important alone, but it interacts with the glue distribution characteristics of the adjacent area and jointly determines the consistency of glue coating. By calculating the neighborhood activity of the local feature of the glue line state, the relationship between the local feature and its neighboring features can be considered, and the interactive information between features can be fully captured, so as to more accurately evaluate the glue coating state.

[0046] Accordingly, considering that when judging the consistency of gluing, selecting features only based on the importance score of a single feature may ignore the synergy between features. Each local feature vector does not exist in isolation, and it interacts with the surrounding neighborhood features. For example, in the monitoring of the gluing status of the inverter housing, the width feature of a certain point of the glue line may be more important alone, but it works together with the neighborhood features such as the glue line height and surface texture at the adjacent position to more accurately reflect the actual situation of gluing. The present application performs feature selection on the descending sequence of the local feature vectors of the glue line segment state based on the sequence of the neighborhood activity of the local features of the glue line segment state, thereby ensuring that the selected local features of the glue line segment state are not only important in themselves, but also can cooperate with the surrounding features, so as to better represent the complex glue state information in the original data.

[0047] Finally, the descending sequence of the selected glue line segment state local feature vectors is cascaded to obtain the glue line segment state feature encoding vector. The above process can be expressed as: ;in, and They are the first, second, and third in the descending sequence of the local feature vectors of the selected glue line segment state. and The local feature vector of the selected glue segment state, For cascade operation, is the feature encoding vector of the glue segment state.

[0048] It should be understood that each vector in the descending sequence of the local feature vectors of the glue line state after selection contains information that is important for judging the glue coating state. Through the cascade operation, these scattered local feature vectors can be connected in sequence to form a single glue line state feature encoding vector. This vector can comprehensively reflect the state information of the glue line segment in different aspects, avoid misjudgment caused by missing or one-sided information, and is conducive to improving the accuracy of judging the consistency of glue coating. It is also worth mentioning that the cascade of the descending sequence retains the order of importance of the features. In the previous feature selection process, the feature vectors have been sorted according to importance, and this order is continued during the cascade. This means that in the final glue line state feature encoding vector, the front part corresponds to the features that are more critical to judging the glue coating state. This order helps the model to give priority to the most important feature information when processing the vector, thereby improving the efficiency and accuracy of information processing.

[0049] In step S6, the set of the glue line segment state feature coding vectors is subjected to a glue line segment state difference semantic measurement to obtain a glue line global fine-grained difference coding representation vector. Specifically, in an embodiment of the present application, the step S6 includes: calculating the glue line segment state difference semantic measurement coefficient between any two glue line segment state feature coding vectors in the set of the glue line segment state feature coding vectors to obtain the glue line global fine-grained difference coding representation vector composed of a plurality of glue line segment state difference semantic measurement coefficients. It should be understood that the defects that may occur during the gluing process are often very subtle, such as slight width unevenness or discontinuity problems, and these details that are not easy to detect but may affect the quality of the final product. Based on this, in the technical solution of the present application, the glue line segment state difference semantic measurement coefficient between any two glue line segment state feature coding vectors in the set of the glue line segment state feature coding vectors is calculated to obtain a glue line global fine-grained difference coding representation vector composed of a plurality of glue line segment state feature coding vectors. That is to say, by calculating the difference semantic measurement coefficient between any two glue line segment state feature coding vectors, the subtle differences between different glue line segments can be deeply analyzed. For example, even if two glue line segments appear to meet the basic glue coating standards, the finer-grained features, such as the difference in micro-texture of the glue line surface, can be accurately captured by this measurement coefficient, thereby providing a basis for comprehensive evaluation of glue coating consistency. In particular, in a specific embodiment of the present application, the cosine similarity between two vectors can be calculated as a semantic measurement coefficient of the difference in glue line state between the two vectors.

[0050] In step S7, based on the global fine-grained difference coding representation vector of the glue line, a quality monitoring result is obtained to indicate whether the gluing consistency meets the preset requirements. Specifically, in an embodiment of the present application, the step S7 includes: inputting the global fine-grained difference coding representation vector of the glue line into a quality monitoring module based on a classifier to obtain the quality monitoring result, and the quality monitoring result is used to indicate whether the gluing consistency meets the preset requirements. That is, the global fine-grained difference coding representation vector of the glue line obtained by calculating the global and fine-grained semantic differences using the set of the glue line segment state feature coding vectors is classified and processed, thereby utilizing the classifier's powerful ability in processing data classification problems to accurately classify according to the feature pattern of the input data, and obtain a quality monitoring result of whether the gluing consistency meets the preset requirements. In this way, the production enterprise can timely discover the problems existing in the gluing process based on this result, and take corresponding improvement measures, such as adjusting the parameters of the gluing equipment, optimizing the gluing process, etc., so as to ensure the stability of product quality. In a specific embodiment of the present application, the global fine-grained difference encoding representation vector of the glue line is input into a quality monitoring module based on a classifier to obtain the quality monitoring result, and the quality monitoring result is used to indicate whether the glue coating consistency meets the preset requirements, including: using the fully connected layer of the classifier to fully connect the global fine-grained difference encoding representation vector of the glue line to obtain a global fine-grained difference fully connected encoding feature vector of the glue line; inputting the global fine-grained difference fully connected encoding feature vector of the glue line into the Softmax classification function of the classifier to obtain the probability value of the global fine-grained difference encoding representation vector of the glue line belonging to each classification label, and the classification labels include those used to indicate that the glue coating consistency meets the preset requirements and those used to indicate that the glue coating consistency does not meet the preset requirements; the classification label corresponding to the largest of the probability values ​​is determined as the quality monitoring result.

[0051] Here, since each glue line segment state feature coding vector in the set of glue line segment state feature coding vectors respectively represents the selective image semantic feature representation of the corresponding glue line segment ROI image based on the image semantic local associated neighborhood activity, when calculating the difference semantic measurement coefficient, the image semantic selective feature representation between local image semantic space domains will have a micro-macro representation deviation relative to the global image semantic space domain, thereby causing a dynamic deviation in the mapping of features to class target probabilities when the glue line global fine-grained difference coding representation vector is input into the classifier-based quality monitoring module, thereby reducing the accuracy of the obtained quality monitoring results.

[0052] Therefore, when the global fine-grained difference coding representation vector of the glue line is input into the quality monitoring module based on the classifier to obtain the quality monitoring result, the global fine-grained difference coding representation vector of the glue line is optimized. The optimization process includes: based on the feature value of the global fine-grained difference coding representation vector of the glue line, the feature value of the global fine-grained difference coding representation vector of the glue line is optimized. Distance and Distance to get glue line global fine-grained difference encoding - distance matrix and glue line global fine-grained difference encoding two distance matrices .

[0053] Determine the number of super-distributed eigenvalues ​​in the glue line global fine-grained difference coding representation vector whose difference from the eigenvalue mean is greater than the eigenvalue variance, and respectively calculate the reciprocal of the logarithm of the number of super-distributed eigenvalues ​​with base two and the exponential value of the reciprocal of the number of super-distributed eigenvalues ​​with base natural constants to obtain the first glue line global fine-grained difference coding full network interactive representation value The global fine-grained difference encoding of the second glue line represents the value of the whole network interaction : ;in, represents the glue line global fine-grained difference encoding representation vector, The glue line global fine-grained difference encoding represents the vector eigenvalues, Representation calculation The number of The glue line global fine-grained difference encoding represents the vector The number of eigenvalues ​​of the superdistribution in , and and is the glue line global fine-grained difference encoding representation vector The mean and variance of all eigenvalues ​​of .

[0054] Calculate the representation vector of the glue line global fine-grained difference encoding, the first glue line global fine-grained difference encoding hidden basic feature vector and the second glue line global fine-grained difference encoding hidden basic feature vector respectively with the inverse of the first glue line global fine-grained difference encoding full network interaction representation value and the inverse of the second glue line global fine-grained difference encoding full network interaction representation value as the index: ;in, It represents the characteristic vector of calculating the representation vector of the global fine-grained difference encoding of the glue line with the inverse of the first glue line global fine-grained difference encoding full network interaction representation value as the index, It represents the characteristic vector of calculating the glue line global fine-grained difference encoding representation vector with the reciprocal of the second glue line global fine-grained difference encoding full network interaction representation value as the index, represents the first glue line global fine-grained difference encoding hidden basis feature vector, Represents the second glue line global fine-grained difference encoding hidden basis feature vector.

[0055] The first glue line global fine-grained difference encoding hidden basic feature vector as a row vector is matrix multiplied with the glue line global fine-grained difference encoding-distance matrix to obtain a glue line global fine-grained difference encoding-distance query vector ,in, Represents matrix multiplication.

[0056] The glue line global fine-grained difference encoding two distance matrix is ​​matrix-multiplied with the second glue line global fine-grained difference encoding hidden basic feature vector as a column vector to obtain the glue line global fine-grained difference encoding two distance response vectors .

[0057] The weighted sum of the glue line global fine-grained difference coding one distance query vector and the glue line global fine-grained difference coding two distance response vectors is calculated to obtain an optimized glue line global fine-grained difference coding representation vector.

[0058] That is, in order to address the problem of insufficient overall active query encoding performance of the global fine-grained difference coding representation vector of the glue line under a fixed numerical sequence layout due to the long distance exceeding the local spacing limit, a multi-dimensional attribute basic form based on the self-distance matrix integration of the global fine-grained difference coding representation vector of the glue line is adopted to capture the complex architecture of its value domain and the whole network interaction, and the performance of the live sales temporal semantic query response coding vector is reconstructed by simulating the scaled multi-dimensional attribute hidden basis of the global fine-grained difference coding representation vector of the glue line, so as to realize the reconstruction of the active search performance connection of the global fine-grained difference coding representation vector of the glue line, and then reflect the predictable reconstruction of the actual series arrangement action of the global fine-grained difference coding representation vector of the glue line, and improve the accuracy of the quality monitoring result obtained by the quality monitoring module based on the classifier of the global fine-grained difference coding representation vector of the glue line.

[0059] In summary, the inverter housing gluing quality monitoring method based on visual inspection based on the embodiment of the present application is explained, which uses computer vision-based image processing technology to extract ROI from the inverter housing gluing status monitoring image, and then performs image cutting and gluing status detection on the extracted gluing area ROI image, and then performs local activity feature selection on the glue line segment state features obtained after each detection, so as to intelligently judge whether the gluing consistency meets the preset requirements based on the global fine-grained difference representation between the selected glue line segment state features. In this way, by analyzing the ROI image of the gluing area, the waste of computing resources can be greatly reduced, and it is also beneficial to improve the detection efficiency. At the same time, the subtle features of the glue line can be accurately identified, and a more detailed gluing quality assessment can be achieved.

[0060] Figure 5 FIG. 1 is a system block diagram of an inverter housing glue coating quality monitoring system based on visual detection according to an embodiment of the present application. Figure 5 As shown, according to the inverter housing gluing quality monitoring system 100 based on visual inspection according to the embodiment of the present application, it includes: an inverter housing gluing state data acquisition module 110, which is used to obtain the inverter housing gluing state monitoring image collected by an industrial camera; an inverter housing gluing area ROI extraction module 120, which is used to perform ROI extraction on the inverter housing gluing state monitoring image to obtain a gluing area ROI image; an inverter housing gluing area ROI image cutting module 130, which is used to perform image cutting on the gluing area ROI image along the glue line to obtain a set of glue line segment ROI images; an inverter housing gluing state detection module 140, which is used to extract the glue line segments in the set of glue line segment ROI images. The ROI image is subjected to glue coating state detection to obtain a set of glue line segment state feature coding maps; a glue line segment state feature sparse coding module 150 is used to perform glue line segment state feature sparseness on each glue line segment state feature coding map in the set of glue line segment state feature coding maps to obtain a set of glue line segment state feature coding vectors; a glue line segment state feature difference measurement module 160 is used to perform glue line segment state difference semantic measurement on the set of glue line segment state feature coding vectors to obtain a global fine-grained difference coding representation vector for glue lines; a glue coating quality monitoring result generation module 170 is used to obtain a quality monitoring result indicating whether the glue coating consistency meets preset requirements based on the global fine-grained difference coding representation vector for glue lines.

[0061] Here, those skilled in the art can understand that the specific functions and operations of the various units and modules in the inverter housing glue coating quality monitoring system 100 based on visual detection have been described in the above reference. Figures 1 to 4 The description of the inverter housing gluing quality monitoring method based on visual inspection has been introduced in detail, and therefore, its repeated description will be omitted.

[0062] In summary, the inverter housing gluing quality monitoring system 100 based on visual inspection according to the embodiment of the present application is explained, which uses computer vision-based image processing technology to perform ROI extraction on the inverter housing gluing status monitoring image, and then performs image segmentation and gluing status detection on the extracted gluing area ROI image, and then performs local activity feature selection on the glue line segment state features obtained after each detection, so as to intelligently judge whether the gluing consistency meets the preset requirements based on the global fine-grained difference representation between the selected glue line segment state features. In this way, by analyzing the ROI image of the gluing area, the waste of computing resources can be greatly reduced, and it is also beneficial to improve the detection efficiency. At the same time, the subtle features of the glue line can be accurately identified, thereby achieving a more detailed gluing quality assessment.

Claims

1. A method for monitoring the gluing quality of inverter housing based on visual inspection, characterized in that: include: Acquire the inverter housing glue coating status monitoring image collected by the industrial camera; Performing ROI extraction on the inverter housing gluing state monitoring image to obtain a gluing area ROI image; The ROI image of the glue coating area is cut along the glue line to obtain a set of glue line segment ROI images; the glue coating state of each glue line segment ROI image in the set of glue line segment ROI images is detected to obtain a set of glue line segment state feature coding maps; the glue line segment state feature of each glue line segment state feature coding map in the set of glue line segment state feature coding maps is sparsely processed to obtain a set of glue line segment state feature coding vectors, including: feature decoupling and feature flattening of the glue line segment state feature coding maps to obtain a set of glue line segment state local feature vectors; local importance calculation and neighborhood activity feature selection are performed on the set of glue line segment state local feature vectors to obtain the glue line segment state feature coding vector; glue line segment state difference semantic measurement is performed on the set of glue line segment state feature coding vectors to obtain a global fine-grained difference coding representation vector of the glue line; based on the global fine-grained difference coding representation vector of the glue line, a quality monitoring result is obtained to indicate whether the glue coating consistency meets the preset requirements.

2. The method for monitoring inverter housing gluing quality based on visual inspection according to claim 1 is characterized in that: Performing ROI extraction on the inverter housing gluing state monitoring image to obtain a gluing area ROI image, including: performing ROI extraction based on a U-Net network on the inverter housing gluing state monitoring image to obtain the gluing area ROI image.

3. The method for monitoring inverter housing gluing quality based on visual inspection according to claim 2 is characterized in that: Performing glue coating state detection on each glue line segment ROI image in the set of glue line segment ROI images to obtain a set of glue line segment state feature coding maps, including: using a glue coating state detector based on the Mobile-Former model to perform glue coating state detection on each glue line segment ROI image in the set of glue line segment ROI images to obtain a set of glue line segment state feature coding maps.

4. The method for monitoring inverter housing gluing quality based on visual inspection according to claim 3 is characterized in that: The set of local feature vectors of the glue line segment state is subjected to local importance calculation and feature selection of neighborhood activity to obtain the feature coding vector of the glue line segment state, including: performing local importance calculation on each local feature vector of the glue line segment state in the set of local feature vectors of the glue line segment state to obtain a set of local feature importance score values ​​of the glue line segment state; based on the set of local feature importance score values ​​of the glue line segment state, arranging the set of local feature vectors of the glue line segment state in descending order to obtain a descending sequence of local feature vectors of the glue line segment state; performing feature selection of neighborhood activity on the descending sequence of local feature vectors of the glue line segment state to obtain a descending sequence of local feature vectors of the glue line segment state after selection; and cascading the descending sequence of local feature vectors of the glue line segment state after selection to obtain the feature coding vector of the glue line segment state.

5. The method for monitoring inverter housing gluing quality based on visual inspection according to claim 4 is characterized in that: Performing feature selection of neighborhood activity on the descending sequence of the local feature vectors of the glue line segment state to obtain a descending sequence of the local feature vectors of the glue line segment state after selection, including: based on the neighborhood features of each local feature vector of the glue line segment state in the descending sequence of the local feature vectors of the glue line segment state, calculating the feature neighborhood activity of each local feature vector of the glue line segment state to obtain a sequence of local feature neighborhood activities of the glue line segment state; based on the sequence of local feature neighborhood activities of the glue line segment state, performing feature selection on the descending sequence of the local feature vectors of the glue line segment state to obtain a descending sequence of the local feature vectors of the glue line segment state after selection.

6. The method for monitoring inverter housing gluing quality based on visual inspection according to claim 5 is characterized in that: Based on the sequence of the local feature neighborhood activity of the glue line segment state, feature selection is performed on the descending sequence of the local feature vectors of the glue line segment state to obtain the descending sequence of the local feature vectors of the glue line segment state after selection, including: comparing the local feature neighborhood activity of the glue line segment state corresponding to each local feature vector of the glue line segment state in the descending sequence of the local feature vectors of the glue line segment state with a preset threshold to obtain the descending sequence of the local feature vectors of the glue line segment state after selection; wherein, in response to the local feature neighborhood activity of the glue line segment state corresponding to the local feature vector of the glue line segment state being greater than or equal to the preset threshold, the local feature vector of the glue line segment state remains unchanged; in response to the local feature neighborhood activity of the glue line segment state corresponding to the local feature vector of the glue line segment state being less than the preset threshold, the local feature vector of the glue line segment state is set to zero.

7. The method for monitoring inverter housing gluing quality based on visual inspection according to claim 6 is characterized in that: Performing glue line segment state difference semantic measurement on the set of glue line segment state feature coding vectors to obtain a glue line global fine-grained difference coding representation vector, including: calculating the glue line segment state difference semantic measurement coefficient between any two glue line segment state feature coding vectors in the set of glue line segment state feature coding vectors to obtain the glue line global fine-grained difference coding representation vector composed of multiple glue line segment state difference semantic measurement coefficients.

8. The method for monitoring inverter housing gluing quality based on visual inspection according to claim 7 is characterized in that: Based on the global fine-grained difference encoding representation vector of the glue line, a quality monitoring result is obtained to indicate whether the glue coating consistency meets the preset requirements, including: inputting the global fine-grained difference encoding representation vector of the glue line into a classifier-based quality monitoring module to obtain the quality monitoring result, and the quality monitoring result is used to indicate whether the glue coating consistency meets the preset requirements.

9. An inverter housing gluing quality monitoring system based on visual inspection, characterized in that: include: The inverter housing gluing state data acquisition module is used to obtain the inverter housing gluing state monitoring image collected by the industrial camera; The inverter housing glue coating area ROI extraction module is used to extract the ROI of the inverter housing glue coating state monitoring image to obtain the glue coating area ROI image; the inverter housing glue coating area ROI image cutting module is used to cut the glue coating area ROI image along the glue line to obtain a set of glue line segment ROI images; The inverter housing gluing state detection module is used to perform gluing state detection on each glue line segment ROI image in the set of glue line segment ROI images to obtain a set of glue line segment state feature coding images; A glue line segment state feature sparse coding module is used to perform glue line segment state feature sparseness on each glue line segment state feature coding map in the set of glue line segment state feature coding maps to obtain a set of glue line segment state feature coding vectors, wherein the glue line segment state feature sparse coding module includes: a glue line segment state feature decoupling and flattening unit, used to perform feature decoupling and feature flattening on the glue line segment state feature coding map to obtain a set of glue line segment state local feature vectors; a glue line segment state local feature selection unit, used to perform local importance calculation and neighborhood activity feature selection on the set of glue line segment state local feature vectors to obtain the glue line segment state feature coding vector; a glue line segment state feature difference measurement module, used to perform glue line segment state difference semantic measurement on the set of glue line segment state feature coding vectors to obtain a global fine-grained difference coding representation vector for glue lines; a glue coating quality monitoring result generation module, used to obtain a quality monitoring result indicating whether the glue coating consistency meets preset requirements based on the global fine-grained difference coding representation vector for glue lines.

10. The inverter housing glue coating quality monitoring system based on visual detection according to claim 9 is characterized in that: The glue coating quality monitoring result generation module is used to: input the glue line global fine-grained difference encoding representation vector into the classifier-based quality monitoring module to obtain the quality monitoring result, and the quality monitoring result is used to indicate whether the glue coating consistency meets the preset requirements.

Images