Surface defect detection method for outdoor furniture table-board processing

Abstract

The invention relates to the technical field of defect detection, and particularly discloses a surface defect detection method for outdoor furniture tabletop processing, which comprises the following steps of: capturing surface state images of an OCSB (Optical Common Strand Board) laminated glass tabletop plate under different illumination conditions by triggering a multi-light-source image acquisition device; the visual expression effect of the image is enhanced through image preprocessing and multi-illumination-condition table panel surface state image fusion processing, then background texture interference of the image after enhancement fusion is eliminated through a template matching algorithm, a suspected bubble area in the image is further positioned through an image segmentation algorithm, visual features of the suspected bubble area are extracted, and the visual expression effect of the image is improved. Therefore, a feature discrimination algorithm based on machine vision is constructed, and intelligent detection of bubble defects of the film-coated glass table panel is realized. According to the method, by fusing the image information of the laminated glass table panel under different illumination conditions, glare and shadow interference caused by the high reflection characteristic of OCSB laminated glass are effectively overcome, and then the detection reliability of bubble defects is improved.

Description

Technical Field

[0001] The present application relates to the field of defect detection technology, and more specifically, to a surface defect detection method for outdoor furniture tabletop processing. Background Art

[0002] As an integral part of modern living environments, outdoor furniture surfaces must be made of materials that offer weather resistance, UV resistance, and mechanical strength. OCSB (Optically Clear Structural Bonding) coated glass, with its high light transmittance, corrosion resistance, and structural stability, has become a popular choice for high-end outdoor furniture surfaces.

[0003] However, during the lamination process, fluctuations in ambient temperature and humidity or inaccurate process parameters can easily lead to micron-level bubble defects between the glass and the film layer. Such defects not only reduce the aesthetics of the product, but also cause problems such as film peeling and stress concentration, accelerate material aging during long-term outdoor use, and even pose a safety hazard. Traditional manual visual inspection has problems such as low efficiency (single-piece inspection takes about 2-3 minutes) and a high rate of missed inspections. It is also significantly affected by the experience and fatigue of the inspectors, making it difficult to meet the needs of large-scale production. With the continuous growth of the annual export volume of outdoor furniture, the development of high-precision and high-efficiency intelligent defect detection technology has become a key requirement for improving product qualification rates and reducing after-sales costs.

[0004] In the field of machine vision inspection, bubble defect identification typically relies on image processing techniques such as edge detection and morphological processing. However, due to the highly reflective surface of OCSB coated glass, its imaging is prone to glare or shadows, obscuring the bubble edges and internal texture. This makes it difficult for traditional image processing algorithms to accurately identify the characteristics of bubble defects in coated glass tabletops, which may lead to false or missed detections.

[0005] Therefore, an optimized surface defect detection method for outdoor furniture table top processing is expected. Summary of the Invention

[0006] In order to solve the above-mentioned technical problems, the present application is proposed. The embodiment of the present application provides a surface defect detection method for outdoor furniture tabletop processing, which captures the surface state images of OCSB coated glass tabletop panels under different lighting conditions by triggering a multi-light source image acquisition device, and enhances the visual expression effect of the image through image preprocessing and multi-lighting condition tabletop surface state image fusion processing, and then uses a template matching algorithm to eliminate the background texture interference of the enhanced fused image, and further locates the suspected bubble area in the image through an image segmentation algorithm, extracts its shape, texture, brightness distribution and other visual features, and thus constructs a feature discrimination algorithm based on machine vision to realize intelligent detection of bubble defects in coated glass tabletop panels. This method effectively overcomes the glare and shadow interference caused by the high reflective characteristics of OCSB coated glass by fusing the image information of coated glass tabletop panels under different lighting conditions, thereby improving the detection reliability of bubble defects.

[0007] According to one aspect of the present application, a surface defect detection method for outdoor furniture tabletop processing is provided, comprising:

[0008] Place the OCSB coated glass tabletop to be inspected at the inspection station;

[0009] triggering a camera and a light source to capture a sequence of images of the surface state of the OCSB coated glass desktop panel under different lighting conditions;

[0010] Performing image preprocessing on each desktop panel surface state image in the sequence of desktop panel surface state images and then performing image fusion to obtain a desktop panel surface state enhanced fused image;

[0011] subtracting background texture from the desktop surface state enhanced fused image based on template matching to obtain a desktop surface state enhanced fused image suppressed by background texture;

[0012] Extracting the suspected bubble area ROI image block from the background texture suppressed desktop surface state enhanced fusion image by using an image segmentation algorithm;

[0013] The visual features of the suspected bubble region ROI image block are extracted, and based on the visual features of the suspected bubble region ROI image block, it is determined whether the bubbles in the suspected bubble region ROI image block are real bubbles.

[0014] Compared with the prior art, the surface defect detection method for outdoor furniture tabletop processing provided in this application captures surface state images of OCSB coated glass tabletops under different lighting conditions by triggering a multi-light source image acquisition device. The method then enhances the visual expression of the image through image preprocessing and fusion processing of the surface state images of the tabletops under multiple lighting conditions. Furthermore, a template matching algorithm is used to eliminate background texture interference in the enhanced fused image. Furthermore, an image segmentation algorithm is used to locate suspected bubble areas in the image, extracting visual features such as their shape, texture, and brightness distribution. This method then constructs a feature discrimination algorithm based on machine vision to achieve intelligent detection of bubble defects in coated glass tabletops. By fusing image information of coated glass tabletops under different lighting conditions, this method effectively overcomes the glare and shadow interference caused by the high reflectivity of OCSB coated glass, thereby improving the reliability of bubble defect detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended 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 drawings, the same reference numerals generally represent the same components or steps.

[0016] Figure 1 This is a flow chart of a surface defect detection method for outdoor furniture tabletop processing according to an embodiment of the present application.

[0017] Figure 2 This is a data flow diagram of a surface defect detection method for outdoor furniture tabletop processing according to an embodiment of the present application.

[0018] Figure 3 This is a flowchart of sub-step S3 of the surface defect detection method for outdoor furniture table top processing according to an embodiment of the present application.

[0019] Figure 4 This is a flowchart of sub-step S32 of the surface defect detection method for outdoor furniture table top processing according to an embodiment of the present application.

[0020] Figure 5 This is a flowchart of sub-step S321 of the surface defect detection method for outdoor furniture table top processing according to an embodiment of the present application.

[0021] Figure 6 This is a flowchart of sub-step S3212 of the surface defect detection method for outdoor furniture table top processing according to an embodiment of the present application.

[0022] Figure 7This is a flowchart of sub-step S32122 of the surface defect detection method for outdoor furniture table top processing according to an embodiment of the present application. DETAILED DESCRIPTION

[0023] As used in this application and the claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not intended to refer to the singular but may include the plural. Generally speaking, the terms "comprises" and "include" only indicate the inclusion of the steps and elements specifically identified, and these steps and elements do not constitute an exclusive list. A method or apparatus may also include other steps or elements.

[0024] Although the present application makes various references to certain modules in the system according to embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are illustrative only, and different aspects of the system and method can use different modules.

[0025] Flowcharts are used in this application to illustrate the operations performed by the systems according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, the various steps may be processed in reverse order or simultaneously, as needed. Furthermore, other operations may be added to these processes, or one or more operations may be removed from these processes.

[0026] 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 herein.

[0027] It is worth noting that in this application, all actions to obtain data are carried out in compliance with the relevant data protection laws and policies of the country where they are located and with the authorization given by the owner of the corresponding device.

[0028] In response to the technical problems described in the above background technology, this application proposes a surface defect detection method for outdoor furniture tabletop processing. The method captures the surface state images of OCSB coated glass tabletop panels under different lighting conditions by triggering a multi-light source image acquisition device. The method then enhances the visual expression of the image through image preprocessing and fusion processing of the surface state images of the tabletop panels under multiple lighting conditions. Furthermore, the method utilizes a template matching algorithm to eliminate the background texture interference of the enhanced fused image, and further locates the suspected bubble area in the image through an image segmentation algorithm, extracting its visual features such as shape, texture, and brightness distribution, thereby constructing a feature discrimination algorithm based on machine vision to achieve intelligent detection of bubble defects in coated glass tabletop panels. By fusing the image information of coated glass tabletop panels under different lighting conditions, the method effectively overcomes the glare and shadow interference caused by the high reflectivity of OCSB coated glass, thereby improving the detection reliability of bubble defects.

[0029] Figure 1 This is a flow chart of a surface defect detection method for outdoor furniture tabletop processing according to an embodiment of the present application. Figure 2 Schematic diagram of data flow for the surface defect detection method for outdoor furniture table top processing according to the embodiment of the present application. Figure 1 and Figure 2 As shown, the surface defect detection method for outdoor furniture table top processing includes the following steps: S1, placing the OCSB coated glass table top to be inspected at a detection station; S2, triggering a camera and a light source to collect a sequence of table top surface state images of the OCSB coated glass table top to be inspected under different lighting conditions; S3, performing image preprocessing on each table top surface state image in the sequence of table top surface state images and then performing image fusion to obtain a table top surface state enhanced fusion image; S4, subtracting background texture from the table top surface state enhanced fusion image based on template matching to obtain a background texture suppressed table top surface state enhanced fusion image; S5, extracting a suspected bubble area ROI image block from the background texture suppressed table top surface state enhanced fusion image through an image segmentation algorithm; S6, extracting visual features of the suspected bubble area ROI image block, and judging whether the bubbles in the suspected bubble area ROI image block are real bubbles based on the visual features of the suspected bubble area ROI image block.

[0030] In the above-mentioned surface defect detection method for outdoor furniture tabletop processing, the step S1 is to place the OCSB coated glass tabletop to be inspected at the inspection station. It should be understood that the OCSB coated glass tabletop is a high-frequency load-bearing component used in outdoor furniture, and its surface bubble defects may lead to a decrease in structural strength or damage to the aesthetics. Manual inspection is greatly affected by light and fatigue. The present application uses machine vision technology to achieve standardized inspection, which helps to improve inspection efficiency and accuracy. Specifically, since OCSB coated glass is an optically sensitive material, its surface defect detection has extremely high requirements on the stability of the imaging environment. The inspection station needs to be equipped with a precise positioning fixture and an anti-vibration device to ensure that the relative position of the tabletop and the image acquisition system is fixed, to avoid optical path deviation or image distortion caused by mechanical vibration or tilt, and to isolate environmental stray light interference.

[0031] In the above-mentioned surface defect detection method for outdoor furniture tabletop processing, the step S2 triggers the camera and the light source to collect a sequence of surface state images of the OCSB coated glass tabletop to be detected under different lighting conditions. It should be understood that since the optical response of bubbles on the surface of the OCSB coated glass is directionally sensitive, a single light source cannot fully capture the defect characteristics. Therefore, the present application uses a multi-angle, multi-mode light source combination to stimulate the differentiated reflection / refraction characteristics of bubbles, and collects surface state images of the OCSB coated glass tabletop to be detected under different lighting conditions. The optical behavior of bubbles under different lighting conditions can be used to effectively enhance the visualization of bubble defects and provide high-quality image data for subsequent defect identification.

[0032] In the above-mentioned surface defect detection method for outdoor furniture table top processing, the step S3 is to perform image preprocessing on each desktop board surface state image in the sequence of desktop board surface state images and then perform image fusion to obtain a desktop board surface state enhanced fusion image. Figure 3 FIG. 1 is a flow chart of sub-step S3 of the surface defect detection method for outdoor furniture table top processing according to an embodiment of the present application. Figure 3 As shown, the step S3 includes the steps of: S31, performing image distortion correction, noise filtering and illumination normalization on each desktop panel surface state image in the sequence of desktop panel surface state images to obtain a sequence of preprocessed desktop panel surface state images; S32, performing image fusion on the sequence of preprocessed desktop panel surface state images to obtain the desktop panel surface state enhanced fused image.

[0033] Specifically, the step S31 performs image distortion correction, noise filtering and illumination normalization on each desktop board surface state image in the sequence of desktop board surface state images to obtain a sequence of pre-processed desktop board surface state images. It should be understood that, considering the lens distortion, environmental noise and uneven illumination problems that may exist in the image acquisition process, directly performing image processing on the original desktop board surface state image often has poor results. Therefore, the present application further performs image distortion correction, noise filtering and illumination normalization on each desktop board surface state image to improve image quality and ensure the accuracy and robustness of subsequent image processing algorithms. Specifically, the image distortion correction is performed by using a high-precision checkerboard calibration plate to calibrate camera parameters through a geometric transformation method, and calculate the lens distortion coefficient (including radial distortion coefficients k1, k2 and tangential distortion coefficients p1, p2); a distortion correction mapping table is established based on the calibration results, and a bilinear interpolation algorithm is used to perform reverse mapping correction on the original desktop board surface state image. For edge regions (>80% radius from the center), additional B-spline-based local refinement correction is applied to eliminate residual distortion, thereby avoiding feature blurring caused by misalignment during the subsequent image fusion process. Image noise in desktop surface images primarily originates from the readout noise of the CMOS sensor and environmental electromagnetic interference. This noise manifests as randomly distributed salt and pepper noise and Gaussian noise, which mask the weak edge signals of micron-scale bubbles and cause loss of detail in the subsequent fused image. To address this, a combination of median filtering and non-local mean filtering can be used for noise removal. The median filtering algorithm effectively suppresses salt and pepper noise, while the non-local mean filtering algorithm effectively removes Gaussian noise while preserving image edge and detail information. By combining median filtering and non-local mean filtering, image quality can be significantly improved, providing clearer and more accurate image data for subsequent bubble defect detection. In addition, in a multi-light source system, the brightness and color temperature of different light sources vary significantly (such as the brightness of annular white light is 5000 lux, the brightness of infrared light is 2000 lux, and the color temperature ranges from 3000K to 8500K), resulting in extremely different grayscale distributions of the same defect in different images. For example, the grayscale value of a bubble in a side-lit image is 50-80, and it may be 120-150 in a polarized light image. Direct fusion will produce artifacts due to brightness mismatch, and the grayscale difference of this background texture may be misjudged as a defect during subsequent template matching. To this end, the present application adopts an illumination normalization method, using the whiteboard image under each light source as the brightness reference, calculating the grayscale map of each image, comparing it with the grayscale value of the corresponding whiteboard image, and linearly stretching it to unify the grayscale range, thereby effectively reducing the impact of different lighting conditions on the grayscale distribution of the image, ensuring that image information under different lighting conditions can be seamlessly integrated, and avoiding artifacts caused by brightness mismatch.

[0034] Specifically, step S32 performs image fusion on the sequence of pre-processed desktop surface state images to obtain the desktop surface state enhanced fused image. It should be understood that in the actual application of machine vision inspection technology, the high reflective properties of the OCSB coated glass surface become one of the core factors that make it difficult to detect. The transparent coating on its surface has both protective and decorative functions, but the smooth film layer forms a double reflective interface with the glass substrate, which easily produces specular glare and structural shadows during imaging. Specifically, when the light source is incident at a low angle, the coated surface will reflect strong light like a mirror, forming a bright spot in the image. This type of glare will directly mask the subtle grayscale changes at the edge of the bubble - for example, the refraction halo at the interface between the bubble and the glass may be completely covered by the glare, making the defect outline unrecognizable. For bubbles with greater depth or surface undulations, the high reflective properties will amplify the light and shadow contrast. The defect edge may produce overexposed or underexposed areas due to light refraction, forming irregular shadows that interfere with the capture of the internal texture of the bubble (such as the spiderweb-like texture at the interface between the air and the coating). The above problems make it difficult for images captured under a single lighting condition to accurately present the geometric features (such as diameter, depth) and optical features (such as edge blur, internal reflection pattern) of bubbles. It may even misjudge normal textures as defects or miss key details of real defects. By fusing images captured under multiple lighting conditions, the present application can integrate multi-dimensional information of defects under different spectra, angles and intensities of illumination. For example, a strong light environment can highlight the shadow changes caused by surface undulations, a weak light environment can suppress reflective interference to highlight internal impurities, and polarized light can eliminate the influence of surface mirror reflections. In other words, by fusing images under different lighting conditions, not only can the contrast between the defect area and the background be enhanced, but also the information loss caused by a single light source can be reduced by complementing the image features under different lighting conditions, thereby more comprehensively exposing the morphological details of defects such as bubbles. Among them, Figure 4 FIG. 1 is a flow chart of sub-step S32 of the surface defect detection method for outdoor furniture table top processing according to an embodiment of the present application. Figure 4 As shown, the step S32 includes the steps of: S321, extracting a selected subset of the preprocessed desktop panel surface state images from the sequence of the preprocessed desktop panel surface state images; S322, performing weighted averaging on the selected subset of the preprocessed desktop panel surface state images to obtain the desktop panel surface state enhanced fusion image.

[0035] More specifically, the step S321 extracts a selected subset of the pre-processed desktop board surface state images from the sequence of pre-processed desktop board surface state images. In particular, considering that the sequence of pre-processed desktop board surface state images may contain some redundant information and low-quality frames, such as repeated images under similar lighting angles or low-quality images caused by motion blur, over-exposure / under-exposure, direct full-quantity fusion not only increases the computational complexity, but also introduces noise interference. Therefore, the present application further performs image screening on the sequence of pre-processed desktop board surface state images to extract a subset of images with high information entropy, significant features and complementarity, so as to optimize the subsequent image fusion efficiency and quality. Among them, Figure 5 FIG. 1 is a flow chart of sub-step S321 of the surface defect detection method for outdoor furniture table top processing according to an embodiment of the present application. Figure 5 As shown, the step S321 includes the steps of: S3211, extracting the visual features of each preprocessed desktop panel surface state image in the sequence of the preprocessed desktop panel surface state images to obtain a set of desktop panel surface state visual feature coding vectors; S3212, performing feature selection based on visual feature quality on the set of desktop panel surface state visual feature coding vectors to obtain a selected set of the desktop panel surface state visual feature coding vectors; S3213, based on the selected set of the desktop panel surface state visual feature coding vectors, screening a selected subset of the preprocessed desktop panel surface state images from the sequence of the preprocessed desktop panel surface state images.

[0036] In a specific example of the present application, the step S3211 includes: performing surface state visual feature extraction based on the MobileNe model on each preprocessed desktop panel surface state image in the sequence of the preprocessed desktop panel surface state images to obtain a set of desktop panel surface state visual feature encoding vectors. Specifically, in order to accurately identify the unique contribution and value of each preprocessed desktop panel surface state image in the multi-source image fusion process, it is necessary to accurately extract the visual features of the image, including the shape, size, distribution of bubbles, and optical effects such as surface reflection and shadow, so as to provide an effective basis for subsequent image screening. To this end, the present application adopts a lightweight MobileNe model to process each preprocessed desktop panel surface state image. With its efficient convolutional neural network structure, the MobileNe model can significantly reduce the computational complexity while maintaining high recognition accuracy, thereby achieving rapid feature extraction. In this application, the MobileNe model can effectively capture the edge, texture and shape features in the image by performing a depth-separable convolution operation on the input preprocessed desktop surface state image, encode each image into a highly representative visual feature coding vector, and generate a set of desktop surface state visual feature coding vectors as a comprehensive description of key information in the image such as bubble morphology, surface reflection, shadow, etc.

[0037] In a specific example of the present application, the step S3212 performs feature selection based on visual feature quality on the set of desktop board surface state visual feature coding vectors to obtain a selected set of desktop board surface state visual feature coding vectors. Specifically, since traditional feature selection methods (such as variance threshold and mutual information screening) only focus on the statistical characteristics of a single sample and ignore the distribution correlation between image groups, it is easy to cause redundant or incomplete coverage of the selected subset features. In this regard, the present application proposes a feature selection method based on group distribution learning, which evaluates the uniqueness and information value of each desktop board surface state visual feature coding vector by analyzing the information gain of each desktop board surface state visual feature coding vector relative to the semantic distribution of the entire image data set, thereby screening out a selected set of images with high information entropy and strong complementarity. Among them, Figure 6 FIG. 1 is a flow chart of sub-step S3212 of the surface defect detection method for outdoor furniture table top processing according to an embodiment of the present application. Figure 6As shown, the step S3212 includes the steps of: S32121, extracting the i-th desktop panel surface state visual feature coding vector from the set of desktop panel surface state visual feature coding vectors as the individual desktop panel surface state visual feature coding vector; S32122, determining whether to delete the i-th individual desktop panel surface state visual feature coding vector based on the semantic increment of the individual desktop panel surface state visual feature coding vector relative to the set of desktop panel surface state visual feature coding vectors.

[0038] Figure 7 FIG. 1 is a flow chart of sub-step S32122 of the surface defect detection method for outdoor furniture table top processing according to an embodiment of the present application. Figure 7 As shown, the step S32122 includes the steps of: S321221, performing feature modulation based on group distribution guidance on the individual desktop panel surface state visual feature coding vector to obtain an individual modulated desktop panel surface state visual feature coding vector; S321222, performing information gain measurement on the individual modulated desktop panel surface state visual feature coding vector and the set of the desktop panel surface state visual feature coding vector to determine whether to delete the i-th individual desktop panel surface state visual feature coding vector.

[0039] In a specific example of the present application, step S321221 includes: first, calculating a group distribution semantic map of the set of desktop board surface state visual feature encoding vectors to obtain a desktop board surface state visual feature group distribution semantic map, which is expressed as:

[0040] X={x1,x2,...,x i ,...,x n}

[0041]

[0042] Among them, X represents the set of visual feature encoding vectors of the desktop surface state, x1, x2, x i 、x j and x n denote the first, second, i-th, j-th and n-th desktop surface state visual feature encoding vectors in the set of desktop surface state visual feature encoding vectors, n is the number of desktop surface state visual feature encoding vectors, R(·,·) is the feature association metric function, [·;·] denotes feature cascade, W r and b represent the weight parameter matrix and bias term, sigmoid is the activation function, r i,j Represents x i and x j The semantic association interaction encoding vector of the desktop surface state between them, T is ri,j The characteristic scale value of h is the vector r i,j The index of the feature value at each position in , M represents the semantic map of the group distribution of visual features of the desktop surface state.

[0043] That is, by constructing a semantic map of the group distribution of the desktop board surface state visual features, capturing the global feature relationships such as the eigenvalue distribution, feature co-occurrence frequency, and feature space topological structure of the desktop board surface state visual feature encoding vector, the desktop board surface state visual feature encoding vector is placed in a broader context for examination, and the potential high-order feature correlation and implicit semantic space are excavated to provide group context prior knowledge for subsequent feature modulation, so as to more accurately process the desktop board surface state visual feature encoding vector.

[0044] Then, the individual desktop panel surface state visual feature encoding vector is mapped to the feature space of the desktop panel surface state visual feature population distribution semantic map to obtain the individual modulated desktop panel surface state visual feature encoding vector, which is expressed as:

[0045]

[0046] Among them, v i Represents x i The corresponding visual feature encoding vector of the surface state of the individual modulated desktop board.

[0047] That is, by mapping the individual desktop board surface state visual feature encoding vector to the feature space of the desktop board surface state visual feature group distribution semantic map, the fusion of individual features and group contextual information is realized, and the global structural information and association information such as feature value distribution, co-occurrence frequency and topological structure contained in the desktop board surface state visual feature group distribution semantic map are used to modulate and contextualize the individual features, so that while retaining the original feature information, more group distribution semantics are encoded, forming an individual modulated desktop board surface state visual feature encoding vector that more comprehensively expresses the feature semantic connotation, providing a more discriminative feature representation for subsequent feature selection.

[0048] In a specific example of the present application, step S321222 includes: first, based on the difference adjustment of the individual modulated desktop board surface state visual feature encoding vector relative to the individual modulated desktop board surface state visual feature encoding vector, performing global universal optimization on the desktop board surface state visual feature group distribution semantic map to obtain an optimized desktop board surface state visual feature group distribution semantic map, which is expressed as:

[0049]

[0050] in,(·) Trepresents the transpose of a vector, represents vector multiplication, (·) ⊙-1 It represents the inverse of each eigenvalue in the calculation vector, and M' represents the semantic map of the group distribution of visual features of the optimized desktop surface state.

[0051] Specifically, by analyzing the differences between the mapped modulated individual desktop board surface state visual feature encoding vectors and the original individual desktop board surface state visual feature encoding vectors, the global structure of the desktop board surface state visual feature group distribution semantic map is adjusted, correcting the unevenness of local correlation defects in global energy transfer, so that the desktop board surface state visual feature group distribution semantic map more accurately reflects the overall distribution pattern and universal correlation of feature vectors. The resulting optimized desktop board surface state visual feature group distribution semantic map improves the uneven global semantic energy distribution problem caused by local correlation defects in the desktop board surface state visual feature group distribution semantic map, and enhances its accurate expression of the overall correlation of the individual modulated desktop board surface state visual feature encoding vectors, making the difference significance measurement of individual features in the group context more reliable.

[0052] Secondly, based on the optimized desktop board surface state visual feature group distribution semantic map, the feature distribution of the individual modulated desktop board surface state visual feature coding vector is optimized to obtain the optimized individual modulated desktop board surface state visual feature coding vector, which is expressed as:

[0053] x' i =ln(v i )⊙[ln(x i ) ⊙-1 ]

[0054]

[0055] Where ln(·) represents the logarithmic function with the natural constant e as the base, ⊙ represents the dot product, and x' i Represents x i The corresponding optimized individual desktop surface state visual feature encoding vector, v' i Indicates v i The corresponding optimized individual modulated desktop board surface state visual feature encoding vector.

[0056] That is, the global feature association laws and universal distribution patterns contained in the optimized semantic map of the group distribution of desktop board surface state visual features are used to adjust the distribution of eigenvalues ​​of each dimension of the individual modulated feature coding vector, so that the energy distribution and spatial topological relationship of the individual modulated desktop board surface state visual features in the group semantic background are more consistent with the statistical laws of the overall data, eliminating the influence of residual local association defects, and forming an optimized individual modulated desktop board surface state visual feature coding vector that is more in line with the global feature structure, significantly improving the expression accuracy and consistency of the individual modulated desktop board surface state visual features in the global background, reducing the feature discrimination interference caused by local noise or association defects, and thus improving the global modeling ability and anti-interference performance of the model.

[0057] Then, the individual desktop board surface state visual feature semantic increment operator of the optimized individual modulated desktop board surface state visual feature coding vector relative to the set of desktop board surface state visual feature coding vectors is calculated, which is expressed as follows:

[0058]

[0059] Where L represents v i The vector scale, v i,k Indicates v i The eigenvalue of the kth position in the equation, λ represents v i The corresponding semantic incremental intermediate parameter, arctan(·) represents the inverse tangent function, Indicates v i The corresponding semantic increment operator of the visual features of the individual desktop surface state.

[0060] That is, by calculating the semantic incremental operator of the visual feature of the individual desktop board surface state, the contribution of the optimized individual modulated desktop board surface state visual feature coding vector to the semantic expression ability of the group feature set is quantified, and the impact of removing the optimized individual modulated desktop board surface state visual feature coding vector on the overall feature discrimination information is measured, providing a quantifiable scientific basis for subsequent threshold-based feature deletion decisions.

[0061] Finally, based on the comparison between the semantic increment operator of the individual desktop panel surface state visual feature and a preset threshold, it is determined whether to delete the i-th individual desktop panel surface state visual feature encoding vector, which is expressed as:

[0062]

[0063] Where θ represents the preset threshold and mask(·) represents the mask operation.

[0064] It should be understandable that the larger the semantic incremental operator, the higher the unique information contribution and feature discrimination value of the optimized individual modulated desktop panel surface state visual feature encoding vector relative to the group feature set. By comparing the individual desktop panel surface state visual feature semantic incremental operator with the preset threshold and taking the preset threshold as the decision boundary, important features that play a key role in bubble defect detection and non-important features with insufficient contribution are distinguished, thereby achieving a balance between the number of features and the performance of the detection model, ensuring that the retained features can not only fully express the semantic information required for defect discrimination, but also eliminate irrelevant or redundant features, thereby improving detection efficiency and accuracy.

[0065] In a specific example of the present application, step S3213 filters a selected subset of the pre-processed desktop panel surface state images from the sequence of pre-processed desktop panel surface state images based on the selected set of desktop panel surface state visual feature encoding vectors. That is, based on the retained desktop panel surface state visual feature encoding vectors, the corresponding pre-processed desktop panel surface state images are traced back to construct a selected subset of images containing the most information and complementarity, thereby ensuring that subsequent image fusion processing can focus on the most representative image frames, thereby reducing the consumption of computing resources and improving the accuracy and robustness of the fusion results.

[0066] More specifically, in step S322, weighted averaging is performed on the selected subset of the pre-processed desktop panel surface state images to obtain the desktop panel surface state enhanced fusion image. That is, by performing weighted averaging processing on the selected subset of the screened images, the effective information in each pre-processed desktop panel surface state image is integrated to generate an enhanced fusion image that comprehensively presents the desktop panel surface state. Here, the weight distribution of the selected subset of the pre-processed desktop panel surface state image can be determined based on the semantic incremental operator of each desktop panel surface state visual feature encoding vector in the above-mentioned feature selection process. In this way, by reasonably allocating weights, the interference of noise and redundant information is effectively suppressed, so that the fused image can more clearly show the true state of the desktop panel surface.

[0067] In the above-mentioned surface defect detection method for outdoor furniture tabletop processing, the step S4 subtracts the background texture from the tabletop surface state enhanced fusion image based on template matching to obtain a background texture suppressed tabletop surface state enhanced fusion image. It should be understood that since OCSB coated glass often has an imitation wood grain / stone grain decorative layer, its high-frequency texture is easily misjudged as a defect. Therefore, the present application further separates the background and foreground through template matching technology, which can suppress texture interference in a targeted manner. Specifically, first, images of defect-free samples under different lighting conditions are collected, and a standard background template is generated through registration and averaging processing; then, the phase correlation method is used to calculate the translation amount of the tabletop surface state enhanced fusion image and the template, and sub-pixel alignment is achieved through affine transformation; then, the aligned standard background template is subtracted from the tabletop surface state enhanced fusion image to obtain a tabletop surface state image with the background texture removed. Through this process, the interference of the decorative layer texture on defect detection is effectively reduced, making defects such as bubbles more prominent in the image, thereby improving the accuracy and reliability of detection.

[0068] In the above-mentioned surface defect detection method for outdoor furniture tabletop processing, the step S5 extracts the suspected bubble area ROI image block from the background texture suppressed desktop board surface state enhanced fusion image through an image segmentation algorithm. In a specific example of the present application, the step S5 includes: inputting the background texture suppressed desktop board surface state enhanced fusion image into an image segmentation module based on the Mask R-CNN model to obtain the suspected bubble area ROI image block. Specifically, the present application utilizes a semantic segmentation network based on deep learning, namely the Mask R-CNN model, to perform fine segmentation on the desktop board surface state enhanced fusion image after background texture suppression. The Mask R-CNN model has a powerful instance segmentation capability and can accurately distinguish different objects and their contours in the image, especially tiny defects such as bubbles. It learns the subtle differences between bubbles and background by training a large number of annotated image samples, so that it can automatically identify the area in the background texture suppressed desktop board surface state enhanced fusion image that may be related to bubble defects, and segment the suspected bubble area in the image from the complex background to form the ROI image block of the suspected bubble area. Here, the suspected bubble region ROI image block focuses on the portion of the image that is most likely to contain bubble defects, which significantly narrows the scope of subsequent analysis and improves detection efficiency.

[0069] In the above-mentioned surface defect detection method for outdoor furniture tabletop processing, the step S6 extracts the visual features of the suspected bubble area ROI image block, and judges whether the bubbles in the suspected bubble area ROI image block are real bubbles based on the visual features of the suspected bubble area ROI image block. Specifically, for the extracted suspected bubble area ROI image block, the MobileNe model or other applicable deep learning models can be reused to perform more refined visual feature extraction on it, and the outline, color, texture of the bubble and its position and size in the image and other visual features are mined. Subsequently, a classification algorithm, such as a support vector machine (SVM) or a deep learning classifier is used to make classification decisions on the extracted image block visual features to judge whether the suspected bubbles in the suspected bubble area ROI image block are real defective bubbles, thereby achieving accurate identification of bubble defects.

[0070] In summary, a surface defect detection method for outdoor furniture tabletop processing based on the embodiment of the present application is explained, which captures the surface state images of the OCSB coated glass tabletop under different lighting conditions by triggering a multi-light source image acquisition device, and enhances the visual expression of the image through image preprocessing and multi-lighting condition tabletop surface state image fusion processing. Then, a template matching algorithm is used to eliminate the background texture interference of the enhanced fused image, and an image segmentation algorithm is further used to locate the suspected bubble area in the image, and extract its shape, texture, brightness distribution and other visual features, thereby constructing a feature discrimination algorithm based on machine vision to achieve intelligent detection of bubble defects in coated glass tabletops. This method effectively overcomes the glare and shadow interference caused by the high reflective characteristics of OCSB coated glass by fusing the image information of coated glass tabletops under different lighting conditions, thereby improving the detection reliability of bubble defects.

[0071] In another embodiment of the present application, a production process flow of an OCSB coated glass desktop panel is also provided.

[0072] 1. Base material: 5mm tempered ultra-clear glass.

[0073] (1) Raw material preparation

[0074] Selection of raw materials: The main raw materials are quartz sand, soda ash, feldspar, limestone, etc. These raw materials are mixed in a certain proportion to ensure the chemical stability and physical properties of the glass.

[0075] Adding additives: In order to improve certain properties of glass, such as strength and transparency, appropriate amounts of additives such as clarifiers and colorants are added.

[0076] (2) Glass melting

[0077] Furnace Melting: The mixed raw materials are placed in a high-temperature furnace and heated to approximately 1500°C to melt and form molten glass. During the melting process, the material needs to be continuously stirred and clarified to remove bubbles and impurities, making the molten glass uniform and transparent.

[0078] Molding and Cooling: The molten glass goes through a molding process, such as float glass and calendering, to produce flat glass products. The glass then passes through a cooling device to gradually cool to room temperature, resulting in a sheet of ordinary glass.

[0079] (3) Pretreatment

[0080] Cutting: Based on the product size and shape requirements, use cutting equipment to cut the ordinary glass raw sheet into the required size and shape. When cutting, pay attention to dimensional accuracy and cutting quality to avoid defects such as chipping and cracking.

[0081] Edge grinding: Grinding the edges of the cut glass to remove sharp corners and burrs, making the edges smooth and flat. Edge grinding not only improves the safety of the glass, but also enhances its strength.

[0082] Cleaning and drying: Place the edge-grinded glass into the cleaning equipment, use clean water or detergent to clean the dust, oil and impurities on the surface, and then use the drying equipment to dry the glass to ensure that the glass surface is clean and dry, ready for subsequent use.

[0083] (4) Silk screen processing

[0084] Ink adjustment: According to the printing requirements of black, select suitable glass ink, such as solvent-based, water-based, UV curing and other types of inks, and add appropriate amounts of diluents and additives to adjust the ink's viscosity, drying speed, color and other properties.

[0085] Printing: The dried glass is placed on the printing table at the feed port of the glass screen printer. During the feeding process, an automatic detection device can be installed to perform preliminary corrections to the glass's position and posture. Once the glass is accurately positioned, the screen printing plate automatically descends and adheres to the glass surface. The ink scraper then applies pressure to the screen and moves, squeezing the ink through the mesh of the screen onto the glass surface, forming the desired screen color. During the printing process, the equipment automatically controls parameters such as printing pressure, scraping speed, and scraping angle to ensure consistent print quality.

[0086] Inspection: After printing, the glass is transported to the inspection area, where a visual inspection system or other inspection equipment automatically checks the print quality to see if the print meets the required clarity, color accuracy, adhesion, and other indicators. Any unqualified products are automatically marked or separated.

[0087] Drying: The qualified glass is transferred to the drying area for hot air drying, which allows the ink to dry quickly and solidify on the glass surface, improving the ink's adhesion and durability.

[0088] (5) Tempered treatment

[0089] Heating: The pre-treated glass is fed into a tempering furnace and heated to a temperature close to the softening point of the glass, generally around 600-700°C. During the heating process, the heating rate and temperature uniformity must be controlled to ensure that all parts of the glass are heated evenly to avoid stress concentration.

[0090] Cooling: After heating, the glass is quickly moved into a cooling device for rapid cooling. Common cooling methods include air cooling and liquid cooling. Air cooling uses a high-pressure fan to blow cold air onto the glass surface, rapidly cooling it; liquid cooling involves immersing the glass in a coolant. The cooling rate has a significant impact on the properties of tempered glass. Generally, a cooling rate of over 100°C / s is required to form a uniform compressive stress layer on the glass surface, improving its strength and impact resistance.

[0091] Annealing: After cooling, the glass may have some internal stress. To eliminate this stress and improve the stability and optical properties of the glass, annealing is required. The annealing temperature is generally around 550-600°C, and the annealing time depends on factors such as the thickness and size of the glass.

[0092] (6) Inspection of packaging

[0093] Inspection: A comprehensive quality inspection of tempered glass is conducted, including appearance inspection, dimensional measurement, and mechanical property testing. Appearance inspection primarily checks for surface defects such as scratches, bubbles, pitting, and cracks. Dimensional measurement verifies that the length, width, and diagonal dimensions of the glass meet requirements. Mechanical property testing assesses the strength and impact resistance of the glass through impact and bend tests.

[0094] Packaging: Qualified tempered glass will be packaged in wooden frames according to the size and shape of the glass to prevent it from being damaged during transportation and storage.

[0095] 2. Surface layer: PMMA (polymethyl methacrylate) color film

[0096] (1) Raw material preparation

[0097] Selection of main raw materials: PMMA (polymethyl methacrylate), a high molecular polymer suitable for outdoor use, is selected as the base resin. PMMA resin has good weather resistance, mechanical strength and chemical stability.

[0098] Additives: An appropriate amount of pigment and masterbatch is added to provide the desired color. Antioxidants, light stabilizers, and UV absorbers are also needed to enhance the film's aging resistance in outdoor environments. Additionally, depending on the film's specific performance requirements, other additives such as plasticizers, lubricants, and flame retardants may also be required.

[0099] Raw material mixing: Various raw materials are put into a high-speed mixer in a certain proportion and fully mixed to ensure that pigments, additives, etc. are evenly dispersed in the resin. The mixing time is generally about 10-30 minutes, and the mixing temperature is controlled according to the properties of the raw materials and the formulation requirements.

[0100] (2) Extrusion molding

[0101] Extruder preheating: Before production, heat the extruder to the set temperature, generally between 150℃-250℃.

[0102] Raw material melt extrusion: The mixed raw materials are added into the hopper of the extruder. The raw materials are gradually pushed forward under the rotation of the screw and melted into a uniform melt under the action of high temperature and shear force.

[0103] Filter impurities: The molten material is filtered through a filter to remove possible impurities and incompletely melted particles to ensure the quality of the film.

[0104] Extrusion film forming: The filtered melt is extruded through the die of the extruder to form a continuous film-like melt. The shape and size of the die determine the width and thickness of the film.

[0105] (3) Cooling and shaping

[0106] Chilling roller cooling: The extruded film melt immediately enters the cooling roller for cooling and shaping. The cooling roller usually adopts water cooling or air cooling to quickly reduce the temperature of the film to around room temperature, fix the molecular structure of the film, and form a stable film morphology.

[0107] Adjust cooling parameters: According to factors such as film thickness and production speed, adjust parameters such as the temperature, speed and flow rate of the cooling roller to ensure the cooling effect and quality uniformity of the film.

[0108] (4) Stretch orientation

[0109] Longitudinal stretching: After cooling and shaping, the film enters the longitudinal stretching machine. Under certain temperature and speed conditions, the film is stretched longitudinally to orient the molecular chains of the film in the longitudinal direction, thereby improving the longitudinal strength and tensile properties of the film.

[0110] Transverse stretching: The film after longitudinal stretching enters the transverse stretching machine for transverse stretching, so that the molecular chains of the film are in the transverse direction.

[0111] Orientation further improves the mechanical properties and dimensional stability of the film. The stretching ratio is generally determined by the final use and performance requirements of the film, usually between 2-5 times.

[0112] (5) Surface treatment

[0113] Corona treatment: In order to increase the surface tension of the film and enhance the adhesion of ink, glue, etc. on the film surface, it is usually necessary to

[0114] Corona treatment is to cause the molecules on the film surface to undergo oxidation and polarization reactions under the action of a high-voltage electric field, thereby increasing the surface roughness and polarity.

[0115] Coating treatment: According to the needs, a functional coating such as anti-scratch coating, waterproof coating, anti-static coating, etc. is applied on the surface of the film.

[0116] To improve the performance and added value of the film. The coating method can be a variety of ways such as coating, printing, and vapor deposition.

[0117] (6) Printing and lamination

[0118] Printing: Printing patterns on the film is done using gravure printing. Before printing, a printing plate roller or printing file needs to be made according to the design requirements, and then the ink is evenly transferred to the film surface to form a clear and bright pattern.

[0119] Lamination: For multi-layer outdoor color films, such as those with barrier and thermal insulation properties, films with different functions need to be laminated. Lamination methods can be dry lamination, wet lamination, or extrusion lamination, where two or more layers of film are tightly bonded together using glue or hot melt to form a composite film.

[0120] (7) Cutting and packaging

[0121] Slitting: The large roll of film is cut into the required width and length on a slitting machine. When slitting, attention should be paid to the sharpness of the cutter and the slitting speed to ensure a neat and smooth cut.

[0122] Packaging: The slit film is packaged in various forms, including rolls, bags, and boxes, depending on the film's intended use and customer requirements. During packaging, care must be taken to prevent moisture, sunlight, and contamination to ensure the film's quality is not affected during storage and transportation.

[0123] 3. Combination of base material and surface layer (using flat lamination process)

[0124] (1) Preparation before production

[0125] Material preparation: Prepare the produced base glass, surface material PMMA color film and PUR hot melt adhesive.

[0126] Adjust equipment parameters: adjust the hot pressing temperature and dust collector height of the film laminating machine according to the film laminating process, and adjust the height between the upper and lower pressing rollers according to the thickness of the veneer panel.

[0127] Surface cleaning: First, remove dust from the glass substrate surface, then wipe it with industrial alcohol to remove oil and dirt. After both steps are completed, let it dry naturally. Ensure that the surface of the board is flat, clean, dry, and free of oil, dust, and other impurities.

[0128] Apply affinity agent to the substrate: After the substrate board is cleaned of dust and oil and dried, the affinity agent is applied to the surface of the substrate board. Currently, this is done manually, spraying the affinity agent evenly on the board surface. After spraying, wipe the entire surface again with a clean towel to ensure that the affinity agent is evenly applied to the substrate board. After application, let it dry naturally for 45-60 minutes. The ideal ambient temperature is 20℃-25℃ and the humidity is 50%-60%.

[0129] Installing PMMA film and adjusting the distance: Install the PMMA film roll onto the film unwinding machine, adjust the brake and the distance on both sides to ensure that the material can be unfolded smoothly and neatly during transportation, and pour the water-based PUR hot melt adhesive into the glue coating machine.

[0130] Loading, plasma corona treatment and gluing

[0131] Sheet conveying: Place the substrate sheet to be laminated onto the conveyor belt, start the conveyor belt, and convey the sheet forward at a constant speed. During the conveying process, the sheet can be aligned and positioned by the edge-aligning conveyor device to ensure the sheet's edges are neat.

[0132] Plasma corona treatment:

[0133] Starting the plasma equipment: When the plasma corona treatment equipment is turned on, the power supply begins applying a high-voltage electric field to the electrodes, causing corona discharge in the gas surrounding the electrodes, forming a plasma region. The substrate is slowly fed into the plasma corona equipment via a conveyor belt. The plasma then causes physical and chemical changes on the material surface, introducing polar groups, increasing surface energy, and improving properties such as hydrophilicity and adhesion.

[0134] Preheating: After the substrate has been treated with plasma corona, it passes through the first drying tunnel to preheat the substrate before gluing for flat lamination production to improve the fluidity and bonding effect of the glue. The preheating temperature is generally adjusted according to factors such as the material and thickness of the substrate and the properties of the glue, and is usually between 60℃ and 70℃.

[0135] Glue Injection: Pour water-based PUR hot melt adhesive into the glue applicator. Adjust the speed and gap of the glue rollers, or the flow rate of the glue pump to ensure that the glue is applied evenly and in the right amount. Excessive glue may cause glue overflow and weak adhesion, while too little glue may affect the bond strength.

[0136] (3) Veneers

[0137] Heating: The conveyor belt continues to transport the material through the second drying tunnel, where a circulating fan continuously circulates the heated air within the drying tunnel, achieving uniform heating. This method has high heating efficiency and good temperature uniformity, allowing the glue or coating to dry and cure quickly on the surface of the board, typically between 60°C and 70°C.

[0138] Lamination: When the coated substrate reaches the lamination area, upper and lower pressing rollers squeeze and laminate the substrate and PMMA film. Ensure uniform and moderate pressure, typically between 0.3MPa and 0.8MPa, to ensure a tight fit between the substrate and PMMA film, avoiding air bubbles, wrinkles, and other quality issues.

[0139] Cooling and shaping: After lamination, the sheets and decorative materials need to be cooled to allow the water-based PUR hot melt adhesive to quickly solidify and shape. Natural cooling is usually used, and the cooling time is generally between 5s and 30s.

[0140] Health care: The flat-laid desktop panel needs to be placed in a room with a temperature of 20℃-25℃ and a humidity of 50%-60% for 48 hours to allow the water-based PUR adhesive to completely cure and ensure the bonding strength between the substrate board and the PMMA film.

[0141] (4) Trimming

[0142] Trimming: After 48 hours of curing, the finished tabletop enters the trimming area. This manual trimming removes excess PMMA film from the edges, leaving them neat and beautiful. During trimming, attention must be paid to the sharpness and cutting accuracy of the tool to avoid burrs, gaps, and other issues.

[0143] (5) Quality inspection and packaging

[0144] Quality inspection: The finished desktop panels are inspected for quality. The main inspection items include the bonding strength between the substrate and the PMMA film, surface flatness, the presence of bubbles, wrinkles, degumming and other defects, and whether the dimensional accuracy meets the requirements.

[0145] Packaging: Products that have passed inspection are packaged and categorized by size, shape, and specifications. Separated by plastic film, they are then secured together with wooden frames and placed on pallets. Care must be taken to protect the product surface during packaging to prevent damage during transportation and storage.

[0146] The basic principles of the present invention have been described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, and effects mentioned in the present invention are merely illustrative and non-limiting, and should not be construed as necessarily possessed by each embodiment of the present invention. Furthermore, the specific details of the above embodiments are provided for illustrative purposes and to facilitate understanding, and are not intended to be limiting. These details do not necessarily limit the present invention to being implemented using these specific details.

[0147] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, please refer to the relevant description of other embodiments. In the several embodiments provided by the present invention, it should be understood that the disclosed system and method can be implemented in other ways. For example, the system embodiment described above is only schematic. For example, the unit division is only a logical function division, and there may be other division methods in actual implementation. The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0148] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be encompassed therein. Any reference to a figure in a claim should not be construed as limiting the claim to which it relates.

[0149] In addition, it is obvious that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units stated in the system claims can also be implemented by one unit through software or hardware.

[0150] Finally, it should be noted that the above description has been provided for purposes of illustration and description. Furthermore, the above embodiments are intended only to illustrate the technical solutions of the present invention and are not intended to be limiting. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art will appreciate that the technical solutions of the present invention may be modified or replaced with equivalents without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A surface defect detection method for outdoor furniture tabletop processing, characterized in that: include: Place the OCSB coated glass tabletop to be inspected at the inspection station; triggering a camera and a light source to capture a sequence of images of the surface state of the OCSB coated glass desktop panel under different lighting conditions; Performing image preprocessing on each desktop panel surface state image in the sequence of desktop panel surface state images and then performing image fusion to obtain a desktop panel surface state enhanced fused image; subtracting background texture from the desktop surface state enhanced fused image based on template matching to obtain a desktop surface state enhanced fused image suppressed by background texture; Extracting the suspected bubble area ROI image block from the background texture suppressed desktop surface state enhanced fusion image by using an image segmentation algorithm; The visual features of the suspected bubble region ROI image block are extracted, and based on the visual features of the suspected bubble region ROI image block, it is determined whether the bubbles in the suspected bubble region ROI image block are real bubbles.

2. The surface defect detection method for outdoor furniture table top processing according to claim 1, characterized in that: Performing image preprocessing on each desktop panel surface state image in the sequence of desktop panel surface state images and then performing image fusion to obtain a desktop panel surface state enhanced fused image, including: performing image distortion correction, noise filtering, and illumination normalization on each desktop panel surface state image in the sequence of desktop panel surface state images to obtain a sequence of pre-processed desktop panel surface state images; Image fusion is performed on the sequence of pre-processed desktop panel surface state images to obtain the desktop panel surface state enhanced fused image.

3. The surface defect detection method for outdoor furniture table top processing according to claim 2, characterized in that: Performing image fusion on the sequence of pre-processed desktop panel surface state images to obtain the desktop panel surface state enhanced fused image, comprising: extracting a selected subset of pre-processed desktop board surface state images from the sequence of pre-processed desktop board surface state images; A weighted average is performed on a selected subset of the pre-processed desktop panel surface state images to obtain the desktop panel surface state enhanced fused image.

4. The surface defect detection method for outdoor furniture table top processing according to claim 3, characterized in that: Extracting a selected subset of pre-processed desktop panel surface state images from the sequence of pre-processed desktop panel surface state images comprises: Extracting visual features of each pre-processed desktop panel surface state image in the sequence of pre-processed desktop panel surface state images to obtain a set of desktop panel surface state visual feature encoding vectors; Performing feature selection based on visual feature quality on the set of desktop panel surface state visual feature coding vectors to obtain a selected set of desktop panel surface state visual feature coding vectors; Based on the selected set of desktop panel surface state visual feature encoding vectors, a selected subset of the pre-processed desktop panel surface state images is filtered from the sequence of pre-processed desktop panel surface state images.

5. The surface defect detection method for outdoor furniture table top processing according to claim 4, characterized in that: Extracting visual features of each pre-processed desktop panel surface state image in the sequence of pre-processed desktop panel surface state images to obtain a set of desktop panel surface state visual feature encoding vectors, including: A surface state visual feature extraction based on the MobileNe model is performed on each preprocessed desktop panel surface state image in the sequence of the preprocessed desktop panel surface state images to obtain a set of desktop panel surface state visual feature coding vectors.

6. The surface defect detection method for outdoor furniture table top processing according to claim 5, characterized in that: Performing feature selection based on visual feature quality on the set of desktop panel surface state visual feature encoding vectors to obtain a selected set of desktop panel surface state visual feature encoding vectors, comprising: Extracting the i-th desktop board surface state visual feature coding vector from the set of desktop board surface state visual feature coding vectors as the individual desktop board surface state visual feature coding vector; Based on the semantic increment of the individual desktop panel surface state visual feature coding vector relative to the set of desktop panel surface state visual feature coding vectors, it is determined whether to delete the i-th individual desktop panel surface state visual feature coding vector.

7. The surface defect detection method for outdoor furniture table top processing according to claim 6, characterized in that: Determining whether to delete the i-th individual desktop panel surface state visual feature coding vector based on a semantic increment of the individual desktop panel surface state visual feature coding vector relative to the set of desktop panel surface state visual feature coding vectors includes: Performing a population distribution-guided feature modulation on the individual desktop panel surface state visual feature coding vector to obtain an individual modulated desktop panel surface state visual feature coding vector; An information gain metric is performed on the individual modulated desktop board surface state visual feature coding vector and the set of desktop board surface state visual feature coding vectors to determine whether to delete the i-th individual desktop board surface state visual feature coding vector.

8. The surface defect detection method for outdoor furniture table top processing according to claim 7, characterized in that: Performing a population distribution-guided feature modulation on the individual desktop panel surface state visual feature coding vector to obtain an individual modulated desktop panel surface state visual feature coding vector, comprising: Calculating a group distribution semantic map of the set of desktop panel surface state visual feature encoding vectors to obtain a desktop panel surface state visual feature group distribution semantic map; The individual desktop board surface state visual feature encoding vector is mapped to the feature space of the desktop board surface state visual feature group distribution semantic map to obtain the individual modulated desktop board surface state visual feature encoding vector.

9. The surface defect detection method for outdoor furniture table top processing according to claim 8, characterized in that: Performing information gain measurement on the individual modulated desktop panel surface state visual feature coding vector and the set of desktop panel surface state visual feature coding vectors to determine whether to delete the i-th individual desktop panel surface state visual feature coding vector, comprising: Based on the difference adjustment of the individual modulated desktop board surface state visual feature coding vector relative to the individual modulated desktop board surface state visual feature coding vector, performing global universal optimization on the desktop board surface state visual feature group distribution semantic map to obtain an optimized desktop board surface state visual feature group distribution semantic map; Based on the optimized desktop board surface state visual feature group distribution semantic map, feature distribution optimization is performed on the individual modulated desktop board surface state visual feature coding vector to obtain an optimized individual modulated desktop board surface state visual feature coding vector; Calculating an individual desktop board surface state visual feature semantic increment operator of the optimized individual modulated desktop board surface state visual feature encoding vector relative to the set of desktop board surface state visual feature encoding vectors; Based on the comparison between the semantic increment operator of the individual desktop panel surface state visual feature and a preset threshold, it is determined whether to delete the i-th individual desktop panel surface state visual feature encoding vector.

10. The surface defect detection method for outdoor furniture table top processing according to claim 9, characterized in that: Extracting a suspected bubble region ROI image block from the background texture suppressed desktop surface state enhanced fusion image using an image segmentation algorithm, including: The background texture suppressed desktop surface state enhanced fusion image is input into an image segmentation module based on the Mask R-CNN model to obtain the suspected bubble area ROI image block.

Images