Coating method of special sealing primer for wet basal plane

Abstract

The invention discloses a coating method of a special sealing primer for a wet base plane, which comprises the following steps of: acquiring a brush coating state detection image through a camera, and respectively performing HOG (Histogram of Oriented Gradient) feature extraction and depth feature extraction on the brush coating state detection image by adopting an image recognition and processing technology of computer vision; and automatically obtaining the first brushing quality detection result based on the cross-modal fusion feature between the HOG feature of the brushing state detection image and the shallow semantic multi-scale joint feature. Thus, through an automatic computer vision technology, the influence caused by individual differences of different operators is eliminated, the consistency of detection results is ensured, tiny or hidden defects in the coating can be accurately identified, the problems of non-uniformity, bubbles, peeling and the like can be solved, the possibility of human errors is reduced, and the production efficiency is improved. Therefore, the automation level of brushing quality detection is improved.

Description

Technical Field

[0001] The present application relates to the field of intelligent coating, and more specifically, to a coating method of a special sealing primer for a wet base surface. Background Art

[0002] When applying the sealer on a damp surface, problems such as unevenness, bubbles or peeling may occur, which not only affect the appearance quality of the coating, but also reduce the protective effect of the coating layer, thereby affecting the durability and service life of the coating layer. Therefore, brushing quality inspection becomes particularly important, which can help to timely discover and solve problems in the coating process, ensure the integrity and functionality of the coating, and thus improve the durability of the entire structure.

[0003] However, the traditional manual inspection method relies on the operator's experience and visual judgment, which is not only time-consuming and laborious, but also difficult to ensure the consistency of the inspection results due to individual differences between different operators. In addition, manual inspection is often affected by environmental factors (such as lighting conditions, observation angle, etc.), and cannot accurately identify small or hidden defects, further reducing the reliability and accuracy of inspection.

[0004] Therefore, an optimized coating method for a special sealer for a wet substrate is desired. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a coating method for a special sealing primer for a wet base surface, which collects a brushing state detection image through a camera, and uses computer vision image recognition and processing technology to perform HOG feature extraction and depth feature extraction on the brushing state detection image, respectively, so as to automatically obtain the first brushing quality detection result based on the cross-modal fusion features between the HOG features of the brushing state detection image and the shallow semantic multi-scale joint features. In this way, through automated computer vision technology, the influence of individual differences among different operators is eliminated, the consistency of the detection results is ensured, and small or hidden defects in the coating, including unevenness, bubbles, and peeling, can be accurately identified, reducing the possibility of human errors, thereby improving the automation level of brushing quality detection.

[0006] According to one aspect of the present application, a coating method of a special sealing primer for a wet base surface is provided, comprising:

[0007] Apply the first coat of special sealer for wet base surface on the surface to be coated;

[0008] Performing a brushing quality test on the first coat of the special sealing primer for the wet base surface to obtain a first coat of brushing quality test result;

[0009] In response to the first coating quality inspection result being that the quality meets the requirements, applying a second coat of a special sealing primer for a damp base surface;

[0010] The first coating quality test of the first wet base surface special sealing primer is carried out to obtain the first coating quality test result, including:

[0011] Acquire a brushing state detection image collected by a camera;

[0012] Extracting the HOG features of the brushing state detection image to obtain a brushing state HOG feature vector;

[0013] Inputting the painting state detection image into a painting state multi-scale feature extractor to obtain a painting state shallow feature map and a painting state semantic feature map;

[0014] Performing attention-based salient joint perception processing on the painting state shallow feature map and the painting state semantic feature map to obtain a painting state multi-scale joint salient perception feature map;

[0015] Inputting the painting state HOG feature vector and the painting state multi-scale joint salient perception feature map into a gated multimodal fusion coding unit of a Transformer-like architecture to obtain a HOG feature-guided painting state multi-scale modulation salient perception feature map as a HOG feature-guided painting state multi-scale modulation salient perception feature;

[0016] Based on the HOG feature-guided multi-scale modulation salient perception features of the painting state, the first painting quality detection result is obtained.

[0017] Compared with the prior art, the present application provides a coating method for a special sealing primer for a wet base surface, which collects brushing state detection images through a camera, and uses computer vision image recognition and processing technology to perform HOG feature extraction and depth feature extraction on the brushing state detection images, respectively, so as to automatically obtain the first brushing quality detection result based on the cross-modal fusion features between the HOG features of the brushing state detection images and the shallow semantic multi-scale joint features. In this way, through automated computer vision technology, the influence of individual differences between different operators is eliminated, the consistency of the detection results is ensured, and small or hidden defects in the coating, including unevenness, bubbles, and peeling, can be accurately identified, reducing the possibility of human error, thereby improving the automation level of brushing quality detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] 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.

[0019] Figure 1 It is a flow chart of a coating method of a special sealing primer for a wet base surface according to an embodiment of the present application;

[0020] Figure 2 A data flow diagram of a coating method of a special sealing primer for a wet base surface according to an embodiment of the present application;

[0021] Figure 3 This is a flow chart of sub-step S2 of the method for applying a sealing primer specifically for a damp substrate according to an embodiment of the present application. DETAILED DESCRIPTION

[0022] 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.

[0023] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "an" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0024] Although the present application makes various references to certain modules in the system according to the 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 only illustrative, and different aspects of the system and method can use different modules.

[0025] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps may be processed in reverse order or simultaneously as required. Meanwhile, other operations may also be added to these processes, or a certain step or several steps of 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 here.

[0027] Traditional manual inspection methods rely on the operator's experience and visual judgment. This method is not only time-consuming and laborious, but also difficult to ensure the consistency of the inspection results due to individual differences between different operators. In addition, manual inspection is often affected by environmental factors (such as lighting conditions, observation angle, etc.), and cannot accurately identify small or hidden defects, further reducing the reliability and accuracy of the inspection. Therefore, an optimized coating method for a special sealer for wet substrates is desired.

[0028] In the technical solution of the present application, a coating method of a special sealing primer for a wet base surface is proposed. Figure 1 The present invention is a flow chart of a coating method of a sealing primer specially used for a wet base surface according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of a coating method for a special sealing primer for a wet substrate according to an embodiment of the present application. Figure 1 and Figure 2 As shown, the coating method of the special sealing primer for wet substrates according to the embodiment of the present application includes the steps of: S1, applying a first coat of the special sealing primer for wet substrates on the coated surface; S2, performing a brushing quality inspection on the first coat of the special sealing primer for wet substrates to obtain a first brushing quality inspection result; S3, in response to the first brushing quality inspection result being that the quality meets the requirements, applying a second coat of the special sealing primer for wet substrates.

[0029] In particular, the S1, applies the first layer of special sealing primer for wet base surface on the coated surface. It is worth mentioning that the preparation process of the special sealing primer for wet base surface is: heat the epoxy resin to 50℃-60℃; add plasticizer to the heated epoxy resin and stir evenly to obtain a first mixed material; add filler during stirring of the first mixed material to obtain a second mixed material, wherein the filler is talcum powder; add auxiliary agent during stirring of the second mixed material to obtain a third mixed material, wherein the auxiliary agent is any one or a combination of several of defoaming agent, leveling agent and lubricant; add solvent to the third mixed material to obtain a fourth mixed material, wherein the solvent is ethanol or acetone; add curing agent to the fourth mixed material and stir evenly to obtain a special sealing primer for water-bearing construction surface, wherein the curing agent is a polyamide curing agent or an amine curing agent

[0030] In particular, the step S2 is to perform a brushing quality test on the first wet base surface special sealing primer to obtain a first brushing quality test result. In particular, in a specific example of the present application, Figure 3 As shown, the S2 includes: S21, acquiring a brushing state detection image collected by a camera; S22, extracting the HOG features of the brushing state detection image to obtain a brushing state HOG feature vector; S23, inputting the brushing state detection image into a brushing state multi-scale feature extractor to obtain a brushing state shallow feature map and a brushing state semantic feature map; S24, performing attention-based significant joint perception processing on the brushing state shallow feature map and the brushing state semantic feature map to obtain a brushing state multi-scale joint significant perception feature map; S25, inputting the brushing state HOG feature vector and the brushing state multi-scale joint significant perception feature map into a gated multimodal fusion encoding unit of a Transformer-like architecture to obtain a HOG feature-guided brushing state multi-scale modulation significant perception feature map as a HOG feature-guided brushing state multi-scale modulation significant perception feature; S26, obtaining the first brushing quality detection result based on the HOG feature-guided brushing state multi-scale modulation significant perception feature.

[0031] Specifically, the S21 and S22 obtain a brushing state detection image captured by a camera; and extract the HOG features of the brushing state detection image to obtain a brushing state HOG feature vector. Considering that the brushing state detection image is captured by a camera, there is interference from lighting conditions, and the brushing state detection image contains edge and shape feature information about the brushing, and the uniformity and quality of the brushing are usually related to the clarity of the edge and the regularity of the shape. Based on this, in the technical solution of the present application, the HOG features of the brushing state detection image are extracted to obtain a brushing state HOG feature vector, so that features can be stably extracted under different lighting conditions, so as to more clearly understand and describe the local shape, and provide data support for the subsequent evaluation of the brushing quality.

[0032] Specifically, in S23, the brushing state detection image is input into the brushing state multi-scale feature extractor to obtain a brushing state shallow feature map and a brushing state semantic feature map. Considering that subtle changes in the brushing process may be manifested at different scales, such as shallow textures and structures, and deep high-level complex semantic feature information. The pyramid network (such as the feature pyramid network FPN) can extract features at multiple scales, which is very important for processing brushing states of different sizes and details. Based on this, in the technical solution of the present application, the brushing state detection image is input into a brushing state multi-scale feature extractor based on a pyramid network to capture and extract features of different levels about the brushing state, and obtain a brushing state shallow feature map and a brushing state semantic feature map. In particular, the brushing state multi-scale feature extractor based on a pyramid network here is essentially a pyramid network model.

[0033] Specifically, in S24, the shallow feature map of the brushing state and the semantic feature map of the brushing state are subjected to attention-based significant joint perception processing to obtain a multi-scale joint significant perception feature map of the brushing state. Considering that the shallow feature map of the brushing state and the semantic feature map of the brushing state respectively represent the feature information of different scales of the brushing state, and there is a mutual interaction relationship between the features of the two. Therefore, in order to enhance the overall fusion feature representation of the brushing state based on the mutual interaction relationship between the two, in the technical solution of the present application, the shallow feature map of the brushing state and the semantic feature map of the brushing state are input into a significant joint perception module based on the attention mechanism to obtain a multi-scale joint significant perception feature map of the brushing state. In detail, first, the spatial and channel feature information of the feature is maintained by reshaping the shallow feature map of the brushing state and the semantic feature map of the brushing state to obtain a shape reshaping matrix of the shallow feature of the brushing state and a shape reshaping matrix of the semantic feature of the brushing state. Then, the dependency relationship between the two sets of shape reshaping matrices is calculated by the feature channel-by-channel interaction perception module to obtain a shallow-semantic dependency matrix of the brushing state and a semantic-shallow dependency matrix of the brushing state. That is, the feature channel-by-channel interaction perception module reveals the importance of different features in brush state detection by analyzing the interrelated dependencies in the two channels. Then, the two dependency matrices are pruned by the random dropout module. In particular, the random dropout is a regularization technique that can effectively prevent model overfitting by introducing randomness. Pruning effectively removes unimportant or redundant features, thereby helping the model focus on important features and enhancing the influence of these features in the final decision. Subsequently, the pruned brush state shallow-semantic dependency matrix and the brush state semantic feature shape reshaping matrix, the pruned brush state semantic-shallow dependency matrix and the brush state shallow feature shape reshaping matrix are matrix multiplied to strengthen the key feature information in the original brush state to obtain the optimized shallow and semantic features. Finally, the optimized features are reshaped to restore them to the form of feature maps, and the weighted sum between the optimized feature maps is obtained to obtain the brush state multi-scale joint significant perception feature map that finally integrates the key features of the brush state shallow and semantic interaction.

[0034] In an embodiment of the present application, the brushing state shallow feature map and the brushing state semantic feature map are subjected to attention-based significant joint perception processing to obtain a brushing state multi-scale joint significant perception feature map. Considering that the brushing state shallow feature map and the brushing state semantic feature map respectively represent the feature information of the brushing state at different scales, and there is a mutual interaction relationship between the features of the two, including: reshaping the feature shape of the brushing state shallow feature map and the brushing state semantic feature map to obtain a brushing state shallow feature shape reshaping matrix and a brushing state semantic feature shape reshaping matrix; inputting the brushing state shallow feature shape reshaping matrix and the brushing state semantic feature shape reshaping matrix into the feature channel-by-channel interactive perception module to obtain a brushing state shallow-semantic dependency matrix and a brushing state semantic-shallow dependency matrix; inputting the brushing state shallow-semantic dependency matrix and the brushing state semantic-shallow dependency matrix into the random inactivation module to obtain a pruned brushing state shallow-semantic dependency matrix. The invention relates to a method for obtaining a multi-scale joint salient perceptual feature map of the brushing state by multiplying the pruned brushing state semantic-shallow dependency matrix with the brushing state semantic feature shape reshaping matrix; multiplying the pruned brushing state semantic-shallow dependency matrix with the brushing state shallow feature shape reshaping matrix to obtain a dependency optimized brushing state shallow feature matrix; reshaping the dependency optimized brushing state semantic feature matrix and the dependency optimized brushing state shallow feature matrix to obtain an optimized brushing state shallow feature map and an optimized brushing state semantic feature map; and calculating the weighted sum of the optimized brushing state shallow feature map and the optimized brushing state semantic feature map to obtain the multi-scale joint salient perceptual feature map of the brushing state.

[0035] Among them, the process of inputting the brushing state shallow feature shape reshaping matrix and the brushing state semantic feature shape reshaping matrix into the feature channel-by-channel interactive perception module to obtain the brushing state shallow-semantic dependency matrix and the brushing state semantic-shallow dependency matrix includes: calculating the matrix product between the transposed matrix of the brushing state shallow feature shape reshaping matrix and the brushing state semantic feature shape reshaping matrix to obtain a first correlation feature matrix; calculating the positional division of the scales of the first correlation feature matrix and the first correlation feature matrix to obtain a first correlation feature modulation matrix; calculating the matrix product between the transposed matrix of the brushing state semantic feature shape reshaping matrix and the brushing state shallow feature shape reshaping matrix to obtain a second correlation feature matrix; calculating the positional division of the scales of the second correlation feature matrix and the second correlation feature matrix to obtain a second correlation feature modulation matrix; inputting the first correlation feature modulation matrix and the second correlation feature modulation matrix into the Softmax function respectively to obtain the brushing state shallow-semantic dependency matrix and the brushing state semantic-shallow dependency matrix.

[0036] In summary, in the above embodiment, the shallow feature map of the brushing state and the semantic feature map of the brushing state are processed by salient joint perception based on attention to obtain a multi-scale joint salient perception feature map of the brushing state. Considering that the shallow feature map of the brushing state and the semantic feature map of the brushing state respectively represent the feature information of the brushing state at different scales, and there is an interactive relationship between the features of the two, including: inputting the shallow feature map of the brushing state and the semantic feature map of the brushing state into a salient joint perception module based on the attention mechanism, and processing them with the following salient joint perception formula to obtain the multi-scale joint salient perception feature map of the brushing state; wherein the salient joint perception formula is:

[0037] M a =Reshape(F a )

[0038] M b =Reshape(F b )

[0039]

[0040] W ab = Dropout(M ab )

[0041] W ba = Dropout(M ba )

[0042]

[0043] Fa '=Reshape(M a ')

[0044] F b '=Reshape(M b ')

[0045] F c =αF a '+βF b '

[0046] Among them, F a and F b are the shallow feature map of the brushing state and the semantic feature map of the brushing state, reshape(·) is the shape reshaping operation, M a and M b are respectively the shallow feature shape reshaping matrix of the brushing state and the semantic feature shape reshaping matrix of the brushing state, is the matrix product, M a T and M b T is the transposed matrix of the shallow feature shape reshaping matrix of the brushing state and the semantic feature shape reshaping matrix of the brushing state, S1 and S2 are the scales of the first related feature matrix and the second related feature matrix, Softmax(·) is the Softmax function, M ab and M ba is the brush state shallow-semantic dependency matrix and the brush state semantic-shallow dependency matrix, Dropout(·) is the pruning process, W ab and W ba are the pruned brush state shallow-semantic dependency matrix and the pruned brush state semantic-shallow dependency matrix, respectively. b ' and M a 'respectively the dependency relationship optimized brushing state semantic feature matrix and the dependency relationship optimized brushing state shallow feature matrix, F a ' and F b ' are the shallow feature map of the optimized brushing state and the semantic feature map of the optimized brushing state, α and β are weight hyperparameters, F c It is a multi-scale joint salient perceptual feature map of the brushing state.

[0047] Specifically, in S25, the brushing state HOG feature vector and the brushing state multi-scale joint salient perceptual feature map are input into a gated multimodal fusion coding unit of a Transformer-like architecture to obtain a HOG feature-guided brushing state multi-scale modulation salient perceptual feature map as a HOG feature-guided brushing state multi-scale modulation salient perceptual feature. It should be understood that in the task of brushing quality detection, the brushing state HOG feature vector and the brushing state multi-scale joint salient perceptual feature map represent two different modal information. Specifically, the brushing state HOG feature vector captures the edge and texture information in the brushed image, while the brushing state multi-scale joint salient perceptual feature map contains higher-level semantic information and saliency information. The features of these two modalities can be effectively fused to enhance cross-modal understanding capabilities, thereby forming a more comprehensive brushing state representation. In the technical solution of the present application, the HOG feature vector of the brushing state and the multi-scale joint salient perception feature map of the brushing state are input into the gated multimodal fusion encoding unit of the Transformer-like architecture to obtain the HOG feature-guided brushing state multi-scale modulation salient perception feature map. In particular, the gated multimodal fusion encoding unit of the Transformer-like architecture realizes the effective fusion of different modal features by imitating the self-attention mechanism of the Transformer structure and combining the gated unit. Specifically, first, the multi-scale joint salient perception feature map of the brushing state is layer-normalized to separate the differences between different channels in the feature map. Then, the normalized features are point-convolutionally encoded and spatially convolutionally encoded to capture and extract the spatial feature information of the multi-scale features, and the multi-scale joint salient perception spatial deepening feature map of the brushing state is obtained, and the features are copied to retain the spatial information. Then, the copied features and spatial features are reshaped to obtain the backup brushing state multi-scale joint salient perception spatial deepening feature matrix and the brushing state multi-scale joint salient perception spatial deepening feature matrix and serve as the imitation query feature matrix and the imitation key feature matrix. Afterwards, the autocorrelation matrix between the simulated query feature matrix and the simulated key feature matrix is ​​calculated to capture the brush state correlation between the two. Furthermore, the attention gating unit based on the GELU function is used to dynamically adjust the autocorrelation weight of each eigenvalue in the autocorrelation matrix to achieve adaptive selection and enhancement of features to obtain a local autocorrelation matrix. Then, the brush state HOG feature vector and the local autocorrelation matrix are matrix multiplied to realize the correlation relationship between cross-modalities to obtain a fusion conditional weight vector. Finally, the fusion conditional weight vector is used as the guided conditional feature to weight the brush state multi-scale joint salient perception feature map to obtain the HOG feature-guided brush state multi-scale modulation salient perception feature map.

[0048] In an embodiment of the present application, the brushing state HOG feature vector and the brushing state multi-scale joint significant perception feature map are input into a gated multimodal fusion coding unit of a Transformer-like architecture to obtain a HOG feature-guided brushing state multi-scale modulation significant perception feature map as a HOG feature-guided brushing state multi-scale modulation significant perception feature, including: performing layer normalization and convolution coding on the brushing state multi-scale joint significant perception feature map to obtain a brushing state multi-scale joint significant perception space deepening feature map; copying the brushing state multi-scale joint significant perception space deepening feature map to obtain a backup brushing state multi-scale joint significant perception space deepening feature map; and reshaping the backup brushing state multi-scale joint significant perception space deepening feature map and the brushing state multi-scale joint significant perception space deepening feature map respectively to obtain a backup brushing state multi-scale joint significant perception space deepening feature matrix as a simulated query feature matrix and a brushing state multi-scale joint significant perception space deepening feature matrix. The state multi-scale joint salient perception spatial deepening feature matrix is ​​used as the imitation key feature matrix; the product between the imitation query feature matrix and the transposed matrix of the imitation key feature matrix is ​​calculated to obtain the brush state multi-scale joint salient perception local autocorrelation library matrix; the brush state multi-scale joint salient perception local autocorrelation library matrix is ​​input into the attention gating unit based on the GELU function to obtain the brush state multi-scale joint salient perception gated local autocorrelation library weight matrix; the brush state HOG feature vector is used as the value feature vector, and the product between the brush state multi-scale joint salient perception gated local autocorrelation library weight matrix and the brush state HOG feature vector is calculated to obtain the brush state cross-modal interaction fusion conditional weight vector; the brush state cross-modal interaction fusion conditional weight vector is used as the weight vector, and the brush state multi-scale joint salient perception feature map is weighted channel by channel to obtain the HOG feature-guided brush state multi-scale modulation salient perception feature map.

[0049] Among them, the process of performing layer normalization and convolution coding processing on the multi-scale joint salient perceptual feature map of the brushing state to obtain the multi-scale joint salient perceptual spatial deepening feature map of the brushing state includes: performing layer normalization processing on the multi-scale joint salient perceptual feature map of the brushing state to obtain a layer normalized multi-scale joint salient perceptual feature map of the brushing state; performing point convolution coding and spatial convolution coding on the layer normalized multi-scale joint salient perceptual feature map of the brushing state to obtain the multi-scale joint salient perceptual spatial deepening feature map of the brushing state.

[0050] In summary, in the above embodiment, the brushing state HOG feature vector and the brushing state multi-scale joint significant perception feature map are input into the gated multimodal fusion coding unit of the Transformer-like architecture to obtain the HOG feature-guided brushing state multi-scale modulation significant perception feature map as the HOG feature-guided brushing state multi-scale modulation significant perception feature, including: inputting the brushing state HOG feature vector and the brushing state multi-scale joint significant perception feature map into the gated multimodal fusion coding unit of the Transformer-like architecture, and processing them with the following multimodal fusion formula to obtain the HOG feature-guided brushing state multi-scale modulation significant perception feature map; wherein, the cross-domain joint coding formula is:

[0051] F1=Conv 3×3 (Conv 1×1 (LN(F c )))

[0052] F2=Copy(F1)

[0053]

[0054] Among them, F c is the multi-scale joint salient perceptual feature map of the brushing state, LN(·) is the layer normalization operation, Conv 1×1 (·) is the point convolutional coding, Conv 3×3 (·) is a convolution operation with a convolution kernel of 3×3, F1 is the multi-scale joint significant perception space deepening feature map of the brushing state, Copy(·) is a copy operation, F2 is the backup brushing state multi-scale joint significant perception space deepening feature map, reshape(·) is a shape reshaping operation, (·) T is the transpose of the feature matrix, is matrix multiplication, Gelu(·) is the GELU function, M1 is the weight matrix of the multi-scale joint significant perception gated local autocorrelation library of the brushing state, V f is the HOG feature vector of the brushing state, V a is the cross-modal interaction fusion condition weight vector of the brushing state, F c ' is the HOG feature-guided multi-scale modulation salient perceptual feature map of the brushing state.

[0055] Specifically, the S26 obtains the first brushing quality detection result based on the HOG feature-guided multi-scale modulation significant perception feature of the brushing state. In a specific example of the present application, the HOG feature-guided multi-scale modulation significant perception feature map of the brushing state is input into a quality inspection module based on a classifier to obtain the first brushing quality detection result. That is, the HOG feature-guided multi-scale modulation significant perception feature of the brushing state obtained by guiding the fusion of the brushing state HOG feature vector and the brushing state multi-scale joint significant perception feature map is used to perform classification processing, so as to automatically obtain the first brushing quality detection result. In this way, through automated computer vision technology, the influence of individual differences among different operators is eliminated, the consistency of the detection results is ensured, and small or hidden defects in the coating, including unevenness, bubbles, peeling and other problems, can be accurately identified, reducing the possibility of human errors, thereby improving the automation level of brushing quality detection.

[0056] Preferably, inputting the HOG feature-guided multi-scale modulation significant perceptual feature map of the painting state into a classifier-based quality inspection module to obtain the first painting quality inspection result comprises:

[0057] Expanding the HOG feature-guided brushing state multi-scale modulation salient perceptual feature map into a HOG feature-guided brushing state multi-scale modulation salient perceptual feature vector;

[0058] Subtract the zero norm of the HOG feature-guided brushing state multi-scale modulation significant perception feature vector from the length of the HOG feature-guided brushing state multi-scale modulation significant perception feature vector to obtain the HOG feature-guided brushing state multi-scale modulation significant perception isolated representation value;

[0059] Calculate the logarithm of the square of the HOG feature-guided brush-painted state multi-scale modulation significant perceived isolated representation value and the sum of the HOG feature-guided brush-painted state multi-scale modulation significant perceived isolated representation value with base 2 to obtain the HOG feature-guided brush-painted state multi-scale modulation significant perceived information order value;

[0060] Calculate each eigenvalue of the HOG feature-guided brush state multi-scale modulation significant perception eigenvector as the base, the difference between the HOG feature-guided brush state multi-scale modulation significant perception isolated representation value and one as the exponential power function, and perform point multiplication with the HOG feature-guided brush state multi-scale modulation significant perception information order value to obtain the HOG feature-guided brush state multi-scale modulation significant perception leading vector;

[0061] After performing a dot multiplication on the HOG feature-guided brushing state multi-scale modulation significant perception feature vector and the HOG feature-guided brushing state multi-scale modulation significant perception isolated representation value minus one, the dot multiplication is performed on the reciprocal of the HOG feature-guided brushing state multi-scale modulation significant perception isolated representation value to obtain the HOG feature-guided brushing state multi-scale modulation significant perception field constraint vector;

[0062] Calculate an exponential function with a natural constant as the base and each eigenvalue of the HOG feature-guided brushing state multi-scale modulation significant perception field constraint vector as the exponent to obtain the HOG feature-guided brushing state multi-scale modulation significant perception field bias vector;

[0063] Performing dot addition of the HOG feature-guided brushing state multi-scale modulation salient perception leading vector and the HOG feature-guided brushing state multi-scale modulation salient perception field bias vector to obtain an optimized HOG feature-guided brushing state multi-scale modulation salient perception feature vector; and

[0064] The optimized HOG feature-guided painting state multi-scale modulation significant perception feature vector is input into a classifier-based quality inspection module to obtain the first painting quality inspection result.

[0065] Specifically, the optimization process of the HOG feature-guided brushing state multi-scale modulation salient perceptual feature vector, for example, denoted as V, is expressed as:

[0066]

[0067] n=L-||V||0

[0068] Where V is the HOG feature-guided multi-scale modulation salient perceptual feature vector of the brushing state, V ⊙(n-1) To calculate each eigenvalue of the multi-scale modulation significant perceptual feature vector guided by the HOG feature as the base, the difference between the multi-scale modulation significant perceptual isolated representation value guided by the HOG feature and one is an exponential power function, n is the multi-scale modulation significant perceptual isolated representation value guided by the HOG feature, L is the length of the multi-scale modulation significant perceptual feature vector guided by the HOG feature, ||V||0 represents the 0 norm of the multi-scale modulation significant perceptual feature vector guided by the HOG feature, log represents a logarithm operation with base 2, exp() represents an exponential operation, ⊙ represents a dot product, represents point addition, V' is the optimized HOG feature-guided brushing state multi-scale modulation significant perceptual feature vector.

[0069] Here, the brushing state HOG feature vector and the brushing state multi-scale joint salient perception feature map respectively represent the HOG features of the brushing state detection image and the global saliency perception enhancement features of the multi-scale and multi-depth image semantic features based on the image semantic space distributed attention. Therefore, when performing heterogeneous multimodal semantic context association cross-domain encoding on it, the cross-domain association encoding semantic shift caused by the difference in feature modal expression patterns is taken into account, and it is expected to further improve the heterogeneous semantic aggregation regression comprehensibility of the HOG feature-guided brushing state multi-scale modulated salient perception feature map, thereby improving the accuracy of the first brushing quality inspection result obtained by inputting it into the classifier-based quality inspection module.

[0070] Therefore, the high-dimensional feature manifold of the HOG feature-guided multi-scale modulation significantly perceived feature vector of the brush state is represented by the eigenvalues ​​of the feature set as the vector field of the aggregation dimension, and the overlapping values ​​of the vector field of the HOG feature-guided multi-scale modulation significantly perceived feature vector at the isolated zero position are used as the order information to fix the local position of the eigenvalues ​​of its feature set, and a bias for the reversibility of the feature regression distribution field of the HOG feature-guided multi-scale modulation significantly perceived feature vector of the brush state is added as a reward to achieve the mapping target tracking of the regression distribution of the HOG feature-guided multi-scale modulation significantly perceived feature vector of the brush state for the eigenvalue position, so that the feature set of the HOG feature-guided multi-scale modulation significantly perceived feature vector of the brush state perceives the mapping migration to the aggregate distribution, thereby improving the accuracy of the first brushing quality detection result obtained by inputting the HOG feature-guided multi-scale modulation significantly perceived feature map into the classifier-based quality inspection module. In this way, through automated computer vision technology, the impact of individual differences among operators is eliminated, the consistency of the inspection results is ensured, and small or hidden defects in the coating, including unevenness, bubbles, and peeling, can be accurately identified, reducing the possibility of human error, thereby improving the level of automation of brush coating quality inspection.

[0071] In particular, in step S3, in response to the first coating quality inspection result being that the quality meets the requirements, a second coating of a special sealer for a wet substrate is applied. In this way, minor or hidden defects in the coating, including unevenness, bubbles, and peeling, can be avoided, the possibility of human error is reduced, and the automation level of coating quality inspection is improved.

[0072] In summary, the coating method of the special sealing primer for wet base surface according to the embodiment of the present application is explained, which collects the brushing state detection image through the camera, and uses the image recognition and processing technology of computer vision to extract HOG features and depth and shallowness features of the brushing state detection image respectively, so as to automatically obtain the first brushing quality detection result based on the cross-modal fusion features between the HOG features of the brushing state detection image and the shallow semantic multi-scale joint features. In this way, through the automated computer vision technology, the influence of individual differences between different operators is eliminated, the consistency of the detection results is ensured, and the small or hidden defects in the coating, including unevenness, bubbles and peeling, can be accurately identified, reducing the possibility of human error, thereby improving the automation level of brushing quality detection.

[0073] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A coating method for a special sealing primer for a wet base surface, characterized in that: include: Apply the first coat of special sealer for wet base surface on the surface to be coated; Performing a brushing quality test on the first coat of the special sealing primer for the wet base surface to obtain a first coat of brushing quality test result; In response to the first coating quality inspection result being that the quality meets the requirements, applying a second coat of a special sealing primer for a damp base surface; The first coating quality test of the first wet base surface special sealing primer is carried out to obtain the first coating quality test result, including: Acquire a brushing state detection image collected by a camera; Extracting the HOG features of the brushing state detection image to obtain a brushing state HOG feature vector; Inputting the painting state detection image into a painting state multi-scale feature extractor to obtain a painting state shallow feature map and a painting state semantic feature map; Performing attention-based salient joint perception processing on the painting state shallow feature map and the painting state semantic feature map to obtain a painting state multi-scale joint salient perception feature map; Inputting the painting state HOG feature vector and the painting state multi-scale joint salient perception feature map into a gated multimodal fusion coding unit of a Transformer-like architecture to obtain a HOG feature-guided painting state multi-scale modulation salient perception feature map as a HOG feature-guided painting state multi-scale modulation salient perception feature; Based on the HOG feature-guided multi-scale modulation salient perception features of the painting state, the first painting quality detection result is obtained.

2. The coating method of the special sealing primer for wet base surface according to claim 1, characterized in that: The brushing state detection image is input into a brushing state multi-scale feature extractor to obtain a brushing state shallow feature map and a brushing state semantic feature map, including: the brushing state detection image is input into a brushing state multi-scale feature extractor based on a pyramid network to obtain the brushing state shallow feature map and the brushing state semantic feature map.

3. The coating method of the special sealing primer for wet base surface according to claim 2, characterized in that: The shallow feature map of the brushing state and the semantic feature map of the brushing state are processed by attention-based salient joint perception to obtain a multi-scale joint salient perception feature map of the brushing state, including: inputting the shallow feature map of the brushing state and the semantic feature map of the brushing state into a salient joint perception module based on an attention mechanism to obtain the multi-scale joint salient perception feature map of the brushing state.

4. The coating method of the special sealing primer for wet base surface according to claim 3, characterized in that: Inputting the shallow feature map of the brushing state and the semantic feature map of the brushing state into a salient joint perception module based on an attention mechanism to obtain the multi-scale joint salient perception feature map of the brushing state, including: Performing feature shape reshaping on the painting state shallow feature map and the painting state semantic feature map to obtain a painting state shallow feature shape reshaping matrix and a painting state semantic feature shape reshaping matrix; Input the brushing state shallow feature shape reshaping matrix and the brushing state semantic feature shape reshaping matrix into a feature channel-by-channel interactive perception module to obtain a brushing state shallow-semantic dependency matrix and a brushing state semantic-shallow dependency matrix; Input the brush state shallow-semantic dependency matrix and the brush state semantic-shallow dependency matrix into a random dropout module to obtain a pruned brush state shallow-semantic dependency matrix and a pruned brush state semantic-shallow dependency matrix; Performing matrix multiplication of the pruned brush state shallow-semantic dependency matrix and the brush state semantic feature shape reshaping matrix to obtain a dependency optimized brush state semantic feature matrix; Performing matrix multiplication of the pruned brush state semantic-shallow dependency matrix and the brush state shallow feature shape reshaping matrix to obtain a dependency optimized brush state shallow feature matrix; Reshaping the dependency relationship optimized brushing state semantic feature matrix and the dependency relationship optimized brushing state shallow feature matrix to obtain an optimized brushing state shallow feature map and an optimized brushing state semantic feature map; A weighted sum of the optimized brushing state shallow feature map and the optimized brushing state semantic feature map is calculated to obtain the brushing state multi-scale joint salient perceptual feature map.

5. The coating method of the special sealing primer for wet base surface according to claim 4, characterized in that: Inputting the brushing state shallow feature shape reshaping matrix and the brushing state semantic feature shape reshaping matrix into a feature channel-by-channel interactive perception module to obtain a brushing state shallow-semantic dependency matrix and a brushing state semantic-shallow dependency matrix, including: Calculating the matrix product between the transposed matrix of the brushing state shallow feature shape reshaping matrix and the brushing state semantic feature shape reshaping matrix to obtain a first related feature matrix; calculating a position-wise division of the first correlation feature matrix and the scale of the first correlation feature matrix to obtain a first correlation feature modulation matrix; Calculating the matrix product between the transposed matrix of the brushing state semantic feature shape reshaping matrix and the brushing state shallow feature shape reshaping matrix to obtain a second correlation feature matrix; calculating a position-wise division of the scale of the second correlation feature matrix and the second correlation feature matrix to obtain a second correlation feature modulation matrix; The first correlation feature modulation matrix and the second correlation feature modulation matrix are respectively input into the Softmax function to obtain the brushing state shallow-semantic dependency matrix and the brushing state semantic-shallow dependency matrix.

6. The coating method of the special sealing primer for wet base surface according to claim 5, characterized in that: Inputting the brushing state HOG feature vector and the brushing state multi-scale joint significant perception feature map into a gated multimodal fusion encoding unit of a Transformer-like architecture to obtain a HOG feature-guided brushing state multi-scale modulation significant perception feature map, including: Performing layer normalization and convolution coding processing on the multi-scale joint salient perception feature map of the brushing state to obtain a multi-scale joint salient perception spatial deepening feature map of the brushing state; Copying the painting state multi-scale joint salient perception spatial deepening feature map to obtain a backup painting state multi-scale joint salient perception spatial deepening feature map; Reshape the backup painting state multi-scale joint salient perception space deepening feature map and the painting state multi-scale joint salient perception space deepening feature map respectively to obtain a backup painting state multi-scale joint salient perception space deepening feature matrix as a simulated query feature matrix and a painting state multi-scale joint salient perception space deepening feature matrix as a simulated key feature matrix; Calculating the product of the pseudo-query feature matrix and the transposed matrix of the pseudo-key feature matrix to obtain a brushing state multi-scale joint significant perception local autocorrelation library matrix; Inputting the brush state multi-scale joint significant perception local autocorrelation library matrix into the attention gating unit based on the GELU function to obtain the brush state multi-scale joint significant perception gated local autocorrelation library weight matrix; Taking the painting state HOG feature vector as the value feature vector, calculating the product between the painting state multi-scale joint saliency perception gated local autocorrelation library weight matrix and the painting state HOG feature vector to obtain the painting state cross-modal interaction fusion condition weight vector; The cross-modal interactive fusion condition weight vector of the brushing state is used as the weight vector, and the multi-scale joint salient perceptual feature map of the brushing state is weighted channel by channel to obtain the HOG feature-guided multi-scale modulation salient perceptual feature map of the brushing state.

7. The coating method of the special sealing primer for wet base surface according to claim 6, characterized in that: The multi-scale joint salient perception feature map of the brushing state is subjected to layer normalization and convolution coding processing to obtain a multi-scale joint salient perception spatial deepening feature map of the brushing state, including: Performing layer normalization processing on the multi-scale joint salient perceptual feature map of the painting state to obtain a layer normalized multi-scale joint salient perceptual feature map of the painting state; Point convolution encoding and spatial convolution encoding are performed on the layer-normalized multi-scale joint salient perceptual feature map of the brushing state to obtain the multi-scale joint salient perceptual spatial deepening feature map of the brushing state.

8. The coating method of the special sealing primer for wet base surface according to claim 7, characterized in that: Based on the HOG feature-guided multi-scale modulation significant perceptual features of the painting state, the first painting quality detection result is obtained, including: inputting the HOG feature-guided multi-scale modulation significant perceptual feature map of the painting state into a classifier-based quality inspection module to obtain the first painting quality detection result.

Images