Wood-plastic plate surface flatness detection device and method

Abstract

The invention discloses a device and a method for detecting the surface flatness of a wood-plastic plate, and the method comprises the steps: carrying out the image analysis of a surface image of the wood-plastic plate through employing a computer vision technology based on deep learning, removing the image noise through gray conversion and Gaussian filtering, extracting stable image texture distribution feature representation, and meanwhile, carrying out the detection of the surface flatness of the wood-plastic plate. The embedded representation of the statistical parameters of the image texture features is used as auxiliary information, and the multi-scale texture information fusion representation of the image texture distribution features and the texture statistical features is obtained through cross-domain joint coding, so that intelligent detection of the flatness of the wood-plastic plate is carried out. Through the mode, the anti-interference capability of detection can be effectively improved, and the recognition of illumination variation and small defects on the surface of the plate is enhanced, so that the detection precision and reliability are improved, and the requirements of modern industrial production are met.

Description

Technical Field

[0001] The present invention relates to the field of intelligent detection, and more specifically, to a device and method for detecting the surface flatness of a wood-plastic board. Background Art

[0002] As a new type of environmentally friendly material, wood plastic board is widely used in the fields of construction, furniture, etc. The flatness of its surface is an important indicator to measure the quality of the product, which directly affects the appearance and use effect of the product.

[0003] Traditional methods for detecting the surface flatness of wood-plastic panels mainly rely on manual visual observation or simple mechanical measurement, which is not only inefficient but also easily affected by subjective factors, and cannot meet the needs of modern industrial production for high-precision and high-efficiency detection.

[0004] At present, although there are attempts to use computer vision technology for flatness detection in the existing technology, in actual applications, due to the possible reflection phenomenon on the surface of wood-plastic boards, the existing detection methods are usually not adaptable enough to changes in lighting, which affects the accuracy and reliability of the detection.

[0005] Therefore, an optimized method for detecting the surface flatness of wood-plastic panels is expected. Summary of the invention

[0006] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a device and method for detecting the surface flatness of a wood-plastic board, which uses computer vision technology based on deep learning to perform image analysis on the surface image of the wood-plastic board, removes image noise through grayscale conversion and Gaussian filtering, and extracts a stable image texture distribution feature representation. At the same time, the embedded representation of the statistical parameters of the image texture features is used as auxiliary information, and the multi-scale texture information fusion representation of the image texture distribution features and texture statistical features is obtained through cross-domain joint encoding, so as to perform intelligent detection of the flatness of the wood-plastic board. In this way, the anti-interference ability of the detection can be effectively improved, and the recognition of illumination changes and minor defects on the surface of the board can be enhanced, thereby improving the accuracy and reliability of the detection and meeting the needs of modern industrial production.

[0007] According to one aspect of the present application, a device for detecting the surface flatness of a wood-plastic board is provided, comprising:

[0008] A wood-plastic board surface image acquisition module, used to acquire the surface image of the wood-plastic board collected by a camera;

[0009] An image denoising module, used for denoising the surface image of the wood-plastic board to obtain a denoised surface grayscale image of the wood-plastic board;

[0010] An image texture feature extraction module is used to extract the texture statistical features and image texture distribution features of the denoised surface grayscale image of the wood-plastic board to obtain a texture statistical feature embedding representation vector and a wood-plastic board surface texture feature map;

[0011] A cross-domain joint encoding module is used to perform cross-domain feature-assisted joint encoding on the texture statistical feature embedding representation vector and the surface texture feature map of the wood-plastic board to obtain a texture multi-scale fusion representation feature map assisted by texture statistical features;

[0012] The flatness detection module is used to determine whether the flatness of the wood-plastic board meets the requirements based on the texture multi-scale fusion representation feature map assisted by the texture statistical features.

[0013] According to another aspect of the present application, a method for detecting the surface flatness of a wood-plastic board is provided, comprising:

[0014] Acquire the surface image of the wood-plastic board collected by the camera;

[0015] De-noising the surface image of the wood-plastic board to obtain a de-noised surface grayscale image of the wood-plastic board;

[0016] Respectively extracting the texture statistical features and image texture distribution features of the denoised surface grayscale image of the wood-plastic board to obtain a texture statistical feature embedding representation vector and a wood-plastic board surface texture feature map;

[0017] Performing cross-domain feature-assisted joint encoding on the texture statistical feature embedding representation vector and the surface texture feature map of the wood-plastic board to obtain a texture multi-scale fusion representation feature map assisted by texture statistical features;

[0018] Based on the texture multi-scale fusion representation feature map assisted by the texture statistical features, it is determined whether the flatness of the wood-plastic board meets the requirements.

[0019] Compared with the prior art, the present application provides a device and method for detecting the surface flatness of a wood-plastic board. It uses computer vision technology based on deep learning to analyze the surface image of the wood-plastic board, removes image noise through grayscale conversion and Gaussian filtering, extracts a stable image texture distribution feature representation, and uses the embedded representation of image texture feature statistical parameters as auxiliary information. Through cross-domain joint encoding, a multi-scale texture information fusion representation of image texture distribution features and texture statistical features is obtained to perform intelligent detection of the flatness of the wood-plastic board. In this way, the anti-interference ability of the detection can be effectively improved, and the recognition of light changes and small defects on the surface of the board can be enhanced, thereby improving the accuracy and reliability of the detection and meeting the needs of modern industrial production. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] 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.

[0021] Figure 1 is a block diagram of a device for detecting surface flatness of a wood-plastic board according to an embodiment of the present application;

[0022] Figure 2 This is a data flow diagram of a device for detecting surface flatness of a wood-plastic board according to an embodiment of the present application;

[0023] Figure 3 It is a block diagram of an image texture feature extraction module in a wood-plastic board surface flatness detection device according to an embodiment of the present application;

[0024] Figure 4 The figure is a flow chart of a method for detecting the surface flatness of a wood-plastic board according to an embodiment of the present application. DETAILED DESCRIPTION

[0025] 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.

[0026] 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.

[0027] 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.

[0028] 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.

[0029] 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.

[0030] Traditional methods for detecting the surface flatness of wood-plastic boards mainly rely on manual visual observation or simple mechanical measurement, which is not only inefficient but also easily affected by subjective factors, and cannot meet the needs of modern industrial production for high-precision and high-efficiency detection. At present, although there are attempts to use computer vision technology for flatness detection in the prior art, in actual applications, due to the possible reflection phenomenon on the surface of wood-plastic boards, the existing detection methods are usually not adaptable enough to changes in light, which affects the accuracy and reliability of the detection. Therefore, an optimized method for detecting the surface flatness of wood-plastic boards is expected.

[0031] In the technical solution of the present application, a device for detecting the surface flatness of a wood-plastic board is proposed. Figure 1 4 is a block diagram of a device for detecting surface flatness of a wood-plastic board according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of the surface flatness detection device of the wood-plastic board according to the embodiment of the present application. Figure 1 and Figure 2 As shown, according to an embodiment of the present application, a device 300 for detecting the surface flatness of a wood-plastic board comprises: a wood-plastic board surface image acquisition module 310, for acquiring a wood-plastic board surface image collected by a camera; an image denoising module 320, for denoising the wood-plastic board surface image to obtain a denoised wood-plastic board surface grayscale image; an image texture feature extraction module 330, for respectively extracting texture statistical features and image texture distribution features of the denoised wood-plastic board surface grayscale image to obtain a texture statistical feature embedding representation vector and a wood-plastic board surface texture feature map; a cross-domain joint encoding module 340, for performing cross-domain feature-assisted joint encoding on the texture statistical feature embedding representation vector and the wood-plastic board surface texture feature map to obtain a texture statistical feature-assisted texture multi-scale fusion representation feature map; and a flatness detection module 350, for determining whether the flatness of the wood-plastic board meets the requirements based on the texture statistical feature-assisted texture multi-scale fusion representation feature map.

[0032] In particular, the WPC board surface image acquisition module 310 and the image denoising module 320 are used to acquire the WPC board surface image collected by the camera; and perform denoising on the WPC board surface image to obtain the denoised WPC board surface grayscale image. In a specific example of the present application, the WPC board surface image is first grayscale processed to obtain the WPC board surface grayscale image. It should be understood that since there may be reflection on the surface of the WPC board, the judgment of the flatness of the WPC board is affected. Therefore, in order to improve the stability and accuracy of the subsequent image feature analysis, the present application further performs grayscale processing on the WPC board surface image to remove the interference of color on the image analysis, so that the color change caused by the reflection becomes less obvious in the grayscale image, and the contrast and clarity of the image texture features (such as wood grain, scratches, unevenness) are more prominent, so as to facilitate the subsequent image texture feature extraction and flatness detection. At the same time, the WPC board surface grayscale image is denoised using a Gaussian filter to obtain the denoised WPC board surface grayscale image. Considering that during the acquisition and transmission of the surface image of the wood-plastic board, due to the influence of various factors (such as sensor noise, circuit noise, environmental interference, etc.), noise is often introduced into the image, affecting the image quality. In order to further eliminate the noise in the image, the present application further performs Gaussian filtering on the surface grayscale image of the wood-plastic board to obtain the denoised surface grayscale image of the wood-plastic board. Here, through the smoothing effect of the Gaussian filter, the high-frequency noise in the image can be effectively removed, while the edge and texture details of the image are maintained, thereby reducing the analysis error caused by noise. In addition, since the Gaussian filter has the ability to smooth the image, it can also reduce the local changes in brightness in the image, which helps to reduce the brightness unevenness or highlights caused by reflections, and provide higher quality input for subsequent feature extraction.

[0033] In particular, the image texture feature extraction module 330 is used to extract the texture statistical features and image texture distribution features of the denoised wood-plastic board surface grayscale image to obtain a texture statistical feature embedding representation vector and a wood-plastic board surface texture feature map. In particular, in a specific example of the present application, Figure 3 As shown, the image texture feature extraction module 330 includes: a texture statistical feature extraction unit 331, which is used to generate the texture statistical feature embedding representation vector based on the texture feature parameters of the denoised wood-plastic board surface grayscale image; an image texture distribution feature extraction unit 332, which is used to extract the texture distribution features of the denoised wood-plastic board surface grayscale image using a deep learning model to obtain the surface texture feature map of the wood-plastic board.

[0034] Specifically, the texture statistical feature extraction unit 331 is used to generate the texture statistical feature embedding representation vector based on the texture feature parameters of the grayscale image of the wood-plastic board surface after denoising. In a specific example of the present application, a set of texture feature parameters is first extracted from the grayscale image of the wood-plastic board surface after denoising using a grayscale co-occurrence matrix, and the set of texture feature parameters includes angular second-order moment, contrast, correlation, variance, inverse difference moment, mean sum, variance sum, sum entropy, entropy, variance of difference, and difference entropy; due to the slight unevenness of the surface of the wood-plastic board, such as scratches, bumps, etc., it is usually manifested as a slight change in local texture in the grayscale image. Therefore, in the technical solution of the present application, the angular second-order moment, contrast, correlation, variance, inverse difference moment, mean sum, variance sum, sum entropy, entropy, variance of difference, and difference entropy of the grayscale image of the wood-plastic board surface after denoising are further extracted using a grayscale co-occurrence matrix to obtain a set of texture feature parameters. In this way, the grayscale dependency between neighboring pixels in the grayscale image of the surface of the wood-plastic board after denoising can be effectively revealed, and the local statistical characteristics of the surface texture of the wood-plastic board, such as the intensity, direction and frequency of the texture, can be fully described, so as to capture the tiny uneven information on the surface of the wood-plastic board. Further, the texture statistical feature embedding matrix is ​​used to process the set of texture feature parameters to obtain the texture statistical feature embedding representation vector. Here, in order to integrate multi-dimensional texture feature parameter information to comprehensively characterize the global texture statistical characteristics of the surface of the wood-plastic board, the present application uses a texture statistical feature embedding matrix to embed the set of texture feature parameters, map each texture feature parameter to the same feature space, and capture the mutual correlation between the texture feature parameters at the same time, improve the discrimination of the feature representation, so as to form a more discriminative texture statistical feature embedding representation vector, which is conducive to the subsequent accurate judgment of the surface flatness of the wood-plastic board.

[0035] Specifically, the image texture distribution feature extraction unit 332 is used to extract texture distribution features from the denoised wood-plastic board surface grayscale image using a deep learning model to obtain the wood-plastic board surface texture feature map. In a specific example of the present application, the denoised wood-plastic board surface grayscale image is input into a texture feature extractor based on a hole convolutional neural network model to obtain the wood-plastic board surface texture feature map. Considering that texture feature parameters can only describe the local texture statistical characteristics of the denoised wood-plastic board surface grayscale image, but cannot capture the global distribution information of the image texture, the texture feature extractor based on the hole convolutional neural network model is used to extract features from the denoised wood-plastic board surface grayscale image to obtain a richer texture feature representation. Among them, the hole convolution can effectively expand the receptive field by introducing a hole structure in the convolution kernel, so that the network can capture the long-range texture dependency and global texture distribution features in the image at a lower computational complexity, thereby obtaining the wood-plastic board surface texture feature map, which more intuitively reflects the texture distribution of the wood-plastic board surface.

[0036] It is worth mentioning that in other specific examples of the present application, the texture statistical features and image texture distribution features of the denoised wood-plastic board surface grayscale image can be respectively extracted by other methods to obtain a texture statistical feature embedding representation vector and a wood-plastic board surface texture feature map, for example: input the denoised wood-plastic board surface grayscale image; extract texture statistical features, such as mean, variance, skewness and kurtosis, from the denoised wood-plastic board surface grayscale image; extract image texture distribution features, such as grayscale co-occurrence matrix, local binary pattern and Ga The invention discloses a method for obtaining a surface texture feature of a wood-plastic board by embedding the extracted texture statistical features into a low-dimensional vector as a statistical representation of the surface texture of the wood-plastic board; organizing the extracted image texture distribution features into a feature map as a spatial representation of the surface texture of the wood-plastic board; organizing the extracted texture statistical features (such as mean, variance, skewness and kurtosis) into a vector to obtain the texture statistical feature embedding representation vector; organizing the extracted image texture distribution features (such as gray level co-occurrence matrix, local binary pattern and Gabor filter response) into a feature map to obtain the surface texture feature map of the wood-plastic board.

[0037] In particular, the cross-domain joint encoding module 340 is used to perform cross-domain feature-assisted joint encoding on the texture statistical feature embedding representation vector and the surface texture feature map of the wood-plastic board to obtain a texture multi-scale fusion representation feature map assisted by texture statistical features. In a specific example of the present application, the texture statistical feature embedding representation vector and the surface texture feature map of the wood-plastic board are input into a cross-domain joint encoder based on an auxiliary modality to obtain a texture multi-scale fusion representation feature map assisted by texture statistical features. It should be understood that in order to comprehensively utilize the global texture distribution information and local texture statistical characteristics of the surface image of the wood-plastic board to improve the accuracy and robustness of the flatness detection of the wood-plastic board, the present application introduces a cross-domain joint encoder based on an auxiliary modality to jointly encode the texture statistical feature embedding representation vector and the surface texture feature map of the wood-plastic board. Among them, the heterogeneous Transformer cross-domain joint encoder introduces an attention mechanism in the encoding process, so that the model can pay attention to the correlation between texture statistical features and texture distribution features, realize the deep fusion of the two, and enhance the expression ability of feature representation. Specifically, in the encoding process, the texture statistical feature embedding representation vector and the wood-plastic board surface texture feature map are firstly linearly transformed and feature-shaped reshaped to adapt to the input requirements of the Transformer structure, and the local texture statistical features are used as queries, the global texture distribution features are used as keys and values, and the local texture statistical features are used as auxiliary information to guide the cross-domain union of the global texture distribution features and the local texture statistical features, thereby generating a texture multi-scale fusion representation feature map assisted by texture statistical features. In this way, the texture information of the wood-plastic board surface can be more comprehensively understood, the sensitivity of the model to local micro-defects and global texture changes can be enhanced, and the accuracy and robustness of detection can be improved.

[0038] In an embodiment of the present application, the texture statistical feature embedding representation vector and the surface texture feature map of the wood-plastic board are input into a cross-domain joint encoder based on an auxiliary modality to obtain the texture statistical feature-assisted texture multi-scale fusion representation feature map, including: performing a linear transformation and a feature shape reshaping on the texture statistical feature embedding representation vector to obtain a texture statistical feature auxiliary modality modulation matrix; performing a first linear transformation and a second linear transformation on the surface texture feature map of the wood-plastic board to obtain a first linear transformation surface texture feature map of the wood-plastic board and a second linear transformation surface texture feature map of the wood-plastic board; performing feature shape reshaping on the first linear transformation surface texture feature map of the wood-plastic board and the second linear transformation surface texture feature map of the wood-plastic board to obtain a first linear transformation surface texture feature map of the wood-plastic board. The invention relates to a method for obtaining an auxiliary modal attention query result matrix, an auxiliary modal attention query result matrix, and a second linear transformation wood-plastic board surface texture feature matrix; using the first linear transformation wood-plastic board surface texture feature matrix and the second linear transformation wood-plastic board surface texture feature matrix as a key matrix and a value matrix, and using the texture statistical feature auxiliary modal modulation matrix as a query matrix, inputting the query matrix, the key matrix and the value matrix into a Transformer structure to obtain an auxiliary modal attention query result matrix; performing shape reshaping and point convolution encoding on the auxiliary modal attention query result matrix to obtain an auxiliary modal attention query result feature map; calculating the position point multiplication of the auxiliary modal attention query result feature map and the wood-plastic board surface texture feature map to obtain a texture multi-scale fusion representation feature map assisted by the texture statistical feature.

[0039] In summary, in the above embodiment, the texture statistical feature embedding representation vector and the surface texture feature map of the wood-plastic board are cross-domain feature-assisted joint encoding to obtain a texture multi-scale fusion representation feature map assisted by texture statistical features, including: processing the surface texture feature map of the wood-plastic board and the texture statistical feature embedding representation vector using the following cross-domain joint encoding formula to obtain the texture multi-scale fusion representation feature map assisted by texture statistical features, wherein the cross-domain joint encoding formula is:

[0040] M s = reshape(σ(W s V s +B s ))

[0041] F1=σ(W1F i +B1)

[0042] F2=σ(W2F i +B2)

[0043] M1=reshape(F1)

[0044] M2=reshape(F2)

[0045]

[0046] F w =Conv 1×1 [reshape(M t )]

[0047] F s =F w ⊙F i

[0048] Among them, V s Denotes the texture statistical feature embedding representation vector, F i represents the surface texture feature map of the wood-plastic board, reshape(·) represents the feature shape reshaping, W s , W1 and W2 represent different linear transformation weight matrices, B s , B1 and B2 represent different bias matrices, σ represents the Sigmoid activation function, F1 represents the first linear transformation wood-plastic board surface texture feature map, F2 represents the second linear transformation wood-plastic board surface texture feature map, M s represents the texture statistical feature auxiliary modal modulation matrix, M1 represents the first linear transformation wood-plastic board surface texture feature matrix, M2 represents the second linear transformation wood-plastic board surface texture feature matrix, (·) T represents the transpose of a matrix, represents matrix multiplication, softmax represents the normalized exponential function, M t Represents the auxiliary modality attention query result matrix, Conv 1×1 represents point convolution, F w represents the auxiliary modality attention query result feature map, ⊙ represents the position point multiplication, F s It represents a texture multi-scale fusion representation feature map assisted by the texture statistical features.

[0049] In particular, the flatness detection module 350 is used to determine whether the flatness of the wood-plastic board meets the requirements based on the texture multi-scale fusion representation feature map assisted by the texture statistical feature. In a specific example of the present application, the texture multi-scale fusion representation feature map assisted by the texture statistical feature is input into a defect detector based on a classifier to obtain a defect detection result, and the defect detection result is used to indicate whether there is a defect; then, in response to the defect detection result being the presence of a defect, the flatness detection result is determined to be that the flatness does not meet the requirements. That is, the classifier model is used as a defect detector to analyze the texture multi-scale fusion representation feature map assisted by the texture statistical feature to determine whether there is a defect on the surface of the wood-plastic board. In an embodiment of the present application, the classifier is trained to learn the mapping relationship between the surface feature representation of the board samples with different flatness and the corresponding labels, so that in the test phase, when the classifier receives the texture multi-scale fusion representation feature map assisted by the texture statistical feature, it can classify and map it according to the learned mapping relationship to output the defect detection result of the wood-plastic board. If the defect detection results show that there are defects on the surface of the wood-plastic board, it means that its flatness does not meet the requirements and corresponding quality control measures need to be taken, such as marking it as a substandard product or repairing it. On the contrary, if the defect detection results show that there are no obvious defects on the surface of the wood-plastic board, it can be considered that its flatness meets the production standards and it can enter the next process or be stored and sold as a qualified product.

[0050] In an embodiment of the present application, considering that the texture statistical feature embedding representation vector and the surface texture feature map of the wood-plastic board respectively express the surface texture features of the wood-plastic board determined based on the void convolution coding of the grayscale image of the surface of the wood-plastic board after denoising and the surface texture statistical features of the wood-plastic board determined based on the grayscale co-occurrence matrix, when the heterogeneous Transformer cross-domain joint encoding based on modality assistance is performed on it, the texture multi-scale fusion representation feature map assisted by the texture statistical feature will also cause insufficient feature joint coding coverage due to the structured difference in cross-modal feature distribution, thereby resulting in outlier class reasoning mapping deviation, affecting the accuracy of the defect detection result obtained by the defect detector based on the classifier using the texture multi-scale fusion representation feature map assisted by the texture statistical feature.

[0051] In a preferred example, inputting the texture multi-scale fusion representation feature map assisted by the texture statistical feature into a classifier-based defect detector to obtain a defect detection result includes:

[0052] Calculate the sum of the absolute values ​​of each eigenvalue of the texture multi-scale fusion representation feature map assisted by the texture statistical feature to obtain a first texture multi-scale fusion and modulation value assisted by the texture statistical feature, and calculate the square root of the sum of the squares of each eigenvalue of the texture multi-scale fusion representation feature map assisted by the texture statistical feature to obtain a second texture multi-scale fusion and modulation value assisted by the texture statistical feature;

[0053] After performing dot subtraction on the texture multi-scale fusion representation feature map assisted by texture statistical features and the texture multi-scale fusion and modulation value assisted by the second texture statistical features, dot multiplication is performed with the number of eigenvalues ​​of the texture multi-scale fusion representation feature map assisted by texture statistical features and the reciprocal of the texture multi-scale fusion and modulation value assisted by the first texture statistical features, and the reciprocal of each eigenvalue is taken to obtain the texture multi-scale fusion phase conversion feature map assisted by the first texture statistical features;

[0054] After performing dot subtraction on the texture multi-scale fusion representation feature map assisted by texture statistical features and the texture multi-scale fusion and modulation value assisted by the first texture statistical features, perform dot multiplication with the square root of the number of eigenvalues ​​of the texture multi-scale fusion representation feature map assisted by texture statistical features and the reciprocal of the texture multi-scale fusion and modulation value assisted by the second texture statistical features, and take the reciprocal of each eigenvalue to obtain the texture multi-scale fusion phase conversion feature map assisted by the second texture statistical features;

[0055] Subtract the dot product feature map of the second texture statistical feature-assisted texture multi-scale fusion phase conversion feature map and the weighted hyperparameter from the first texture statistical feature-assisted texture multi-scale fusion phase conversion feature map to obtain an optimized texture statistical feature-assisted texture multi-scale fusion representation feature map;

[0056] The texture multi-scale fusion representation feature map assisted by the optimized texture statistical features is input into a classifier-based defect detector to obtain a defect detection result.

[0057] Here, the optimized representation of the texture multi-scale fusion representation feature map F assisted by the texture statistical feature is:

[0058]

[0059] f i ∈F∈R W×H×C

[0060] n=W×H×C

[0061] Wherein, F is the texture multi-scale fusion representation feature map assisted by the texture statistical feature, R represents a real number set, W, H, and C represent the width, length, and height of the texture multi-scale fusion representation feature map assisted by the texture statistical feature, respectively, and fi represents the i-th eigenvalue of the texture multi-scale fusion representation feature map assisted by the texture statistical feature, n represents the number of eigenvalues ​​of the texture multi-scale fusion representation feature map assisted by the texture statistical feature, α represents the texture multi-scale fusion and modulation value assisted by the first texture statistical feature, β represents the texture multi-scale fusion and modulation value assisted by the second texture statistical feature, ⊙ represents the point multiplication by position, Indicates positional subtraction, (·) ⊙-1 Represents the inverse of each eigenvalue of the calculated feature map, F1 represents the texture multi-scale fusion phase transition feature map assisted by the first texture statistical feature, F2 represents the texture multi-scale fusion phase transition feature map assisted by the second texture statistical feature, ω represents the weighted hyperparameter, and F′ represents the texture multi-scale fusion representation feature map assisted by the optimized texture statistical feature.

[0062] Therefore, in the preferred example, the difference of the feature value of the texture multi-scale fusion representation feature map assisted by texture statistical features relative to the overall feature set of the texture multi-scale fusion representation feature map assisted by texture statistical features and the modulation representation is used as the semantic change intensity information, and the phase-like conversion corresponding to the position-based intensity modulation is performed through different and modulation representation forms, so that the aggregation enhancement of the semantic change phase perception can improve the axial aggregation receptive field along the feature aggregation direction by performing a spatial translation operation based on alternating stacking under the set scale balance of the texture multi-scale fusion representation feature map assisted by texture statistical features, thereby improving the perception effect of the aggregated semantics of the texture multi-scale fusion representation feature map assisted by texture statistical features on the semantic changes of details, so as to improve the accuracy of the defect detection result obtained by the optimized texture multi-scale fusion representation feature map assisted by texture statistical features through a classifier-based defect detector by avoiding the counterfactual reasoning mapping of the outlier feature distribution of the texture multi-scale fusion representation feature map assisted by texture statistical features to the class regression probability.

[0063] As described above, the surface flatness detection device 300 of the wood-plastic board according to the embodiment of the present application can be implemented in various wireless terminals, such as a server with a surface flatness detection algorithm of the wood-plastic board. In a possible implementation, the surface flatness detection device 300 of the wood-plastic board according to the embodiment of the present application can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the surface flatness detection device 300 of the wood-plastic board can be a software module in the operating system of the wireless terminal, or can be an application developed for the wireless terminal; of course, the surface flatness detection device 300 of the wood-plastic board can also be one of the many hardware modules of the wireless terminal.

[0064] Alternatively, in another example, the wood-plastic board surface flatness detection device 300 and the wireless terminal may also be separate devices, and the wood-plastic board surface flatness detection device 300 may be connected to the wireless terminal via a wired and / or wireless network, and transmit interactive information in accordance with an agreed data format.

[0065] Furthermore, a method for detecting the surface flatness of a wood-plastic board is also provided.

[0066] Figure 4 Flow chart of the method for detecting the surface flatness of a wood-plastic board according to an embodiment of the present application. Figure 4 As shown, according to the embodiment of the present application, the surface flatness detection method of the wood-plastic board includes: S1, acquiring a surface image of the wood-plastic board collected by a camera; S2, denoising the surface image of the wood-plastic board to obtain a denoised surface grayscale image of the wood-plastic board; S3, respectively extracting the texture statistical features and image texture distribution features of the denoised surface grayscale image of the wood-plastic board to obtain a texture statistical feature embedding representation vector and a wood-plastic board surface texture feature map; S4, cross-domain feature-assisted joint encoding of the texture statistical feature embedding representation vector and the wood-plastic board surface texture feature map to obtain a texture statistical feature-assisted texture multi-scale fusion representation feature map; S5, based on the texture statistical feature-assisted texture multi-scale fusion representation feature map, determining whether the flatness of the wood-plastic board meets the requirements.

[0067] In summary, according to the embodiment of the present application, the surface flatness detection method of the wood-plastic board is explained, which uses computer vision technology based on deep learning to analyze the surface image of the wood-plastic board, removes image noise through grayscale conversion and Gaussian filtering, and extracts a stable image texture distribution feature representation. At the same time, the embedded representation of the image texture feature statistical parameters is used as auxiliary information, and the multi-scale texture information fusion representation of the image texture distribution feature and the texture statistical feature is obtained through cross-domain joint encoding, so as to perform intelligent detection of the flatness of the wood-plastic board. In this way, the anti-interference ability of the detection can be effectively improved, and the recognition of light changes and small defects on the surface of the board can be enhanced, thereby improving the accuracy and reliability of the detection and meeting the needs of modern industrial production.

[0068] 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 device for detecting the surface flatness of a wood-plastic board, characterized in that: include: A wood-plastic board surface image acquisition module, used to acquire the surface image of the wood-plastic board collected by a camera; An image denoising module, used for denoising the surface image of the wood-plastic board to obtain a denoised surface grayscale image of the wood-plastic board; An image texture feature extraction module is used to extract the texture statistical features and image texture distribution features of the denoised surface grayscale image of the wood-plastic board to obtain a texture statistical feature embedding representation vector and a wood-plastic board surface texture feature map; A cross-domain joint encoding module is used to perform cross-domain feature-assisted joint encoding on the texture statistical feature embedding representation vector and the surface texture feature map of the wood-plastic board to obtain a texture multi-scale fusion representation feature map assisted by texture statistical features; The flatness detection module is used to determine whether the flatness of the wood-plastic board meets the requirements based on the texture multi-scale fusion representation feature map assisted by the texture statistical features.

2. The surface flatness detection device of the wood-plastic board according to claim 1 is characterized in that: The image denoising module comprises: An image grayscale unit, used for grayscale processing the surface image of the wood-plastic board to obtain a grayscale image of the surface of the wood-plastic board; The Gaussian filtering denoising unit is used to perform image denoising on the surface grayscale image of the wood-plastic board using a Gaussian filter to obtain the denoised surface grayscale image of the wood-plastic board.

3. The surface flatness detection device of the wood-plastic board according to claim 2 is characterized in that: The image texture feature extraction module comprises: A texture statistical feature extraction unit, used to generate the texture statistical feature embedding representation vector based on the texture feature parameters of the denoised wood-plastic board surface grayscale image; The image texture distribution feature extraction unit is used to extract texture distribution features from the denoised surface grayscale image of the wood-plastic board using a deep learning model to obtain a surface texture feature map of the wood-plastic board.

4. The surface flatness detection device of the wood-plastic board according to claim 3 is characterized in that: The texture statistical feature extraction unit comprises: A texture feature parameter extraction subunit is used to extract a set of texture feature parameters from the denoised surface grayscale image of the wood-plastic board using a grayscale co-occurrence matrix, wherein the set of texture feature parameters includes angular second-order moment, contrast, correlation, variance, inverse difference moment, mean sum, variance sum, sum entropy, entropy, variance of difference, and difference entropy; The embedding coding subunit is used to process the set of texture feature parameters using the texture statistical feature embedding matrix to obtain the texture statistical feature embedding representation vector.

5. The device for detecting the surface flatness of a wood-plastic board according to claim 4, characterized in that: The image texture distribution feature extraction unit is used to: The denoised surface grayscale image of the wood-plastic board is input into a texture feature extractor based on a hole convolutional neural network model to obtain a surface texture feature map of the wood-plastic board.

6. The surface flatness detection device of the wood-plastic board according to claim 5, characterized in that: The cross-domain joint encoding module is used to: The texture statistical feature embedding representation vector and the surface texture feature map of the wood-plastic board are input into a cross-domain joint encoder based on an auxiliary modality to obtain a texture multi-scale fusion representation feature map assisted by the texture statistical feature.

7. The device for detecting the surface flatness of a wood-plastic board according to claim 6, characterized in that: The cross-domain joint encoding module is used to: Performing linear transformation and feature shape reshaping on the texture statistical feature embedding representation vector to obtain a texture statistical feature auxiliary modal modulation matrix; Performing a first linear transformation and a second linear transformation on the surface texture feature map of the wood-plastic board to obtain a first linear transformation surface texture feature map of the wood-plastic board and a second linear transformation surface texture feature map of the wood-plastic board; Performing feature shape reshaping on the first linear transformation wood-plastic board surface texture feature map and the second linear transformation wood-plastic board surface texture feature map to obtain a first linear transformation wood-plastic board surface texture feature matrix and a second linear transformation wood-plastic board surface texture feature matrix; Taking the first linear transformation wood-plastic board surface texture feature matrix and the second linear transformation wood-plastic board surface texture feature matrix as a key matrix and a value matrix, and taking the texture statistical feature auxiliary modal modulation matrix as a query matrix, inputting the query matrix, the key matrix and the value matrix into a Transformer structure to obtain an auxiliary modal attention query result matrix; Reshape and perform point convolution encoding on the auxiliary modality attention query result matrix to obtain an auxiliary modality attention query result feature map; The auxiliary modal attention query result feature map and the surface texture feature map of the wood-plastic board are multiplied by the position points to obtain the texture multi-scale fusion representation feature map assisted by the texture statistical feature.

8. The device for detecting the surface flatness of a wood-plastic board according to claim 7, characterized in that: The flatness detection module comprises: A defect detection unit, used for inputting the texture multi-scale fusion representation feature map assisted by the texture statistical feature into a classifier-based defect detector to obtain a defect detection result, wherein the defect detection result is used to indicate whether there is a defect; The flatness detection result generating unit is used for determining that the flatness detection result is that the flatness does not meet the requirements in response to the defect detection result being that the defect exists.

9. A method for detecting the surface flatness of a wood-plastic board, characterized in that: include: Acquire the surface image of the wood-plastic board collected by the camera; De-noising the surface image of the wood-plastic board to obtain a de-noised surface grayscale image of the wood-plastic board; Respectively extracting the texture statistical features and image texture distribution features of the denoised surface grayscale image of the wood-plastic board to obtain a texture statistical feature embedding representation vector and a wood-plastic board surface texture feature map; Performing cross-domain feature-assisted joint encoding on the texture statistical feature embedding representation vector and the surface texture feature map of the wood-plastic board to obtain a texture multi-scale fusion representation feature map assisted by texture statistical features; Based on the texture multi-scale fusion representation feature map assisted by the texture statistical features, it is determined whether the flatness of the wood-plastic board meets the requirements.

Images