A method, device, electronic device and storage medium for detecting the performance of a water-based coating

Abstract

An embodiment of the present invention provides a method for detecting the performance of waterborne coatings, which obtains the visual detection information and non-visual detection information of the waterborne coating to be tested under various performance detection methods. The visual detection information includes a visual matrix; the non-visual detection information is encoded to obtain a non-visual matrix with the same size resolution as the visual matrix; the non-visual matrix and the visual matrix are subjected to channel fusion to obtain a fusion matrix, in which each channel corresponds to one of the visual matrix or the non-visual matrix; the fusion matrix is input into a trained detection model for processing to obtain the performance detection result of the waterborne coating to be tested. There is no need for professional personnel to analyze the detection data, thus reducing the difficulty of detecting the performance of waterborne coatings.

Description

Technical Field

[0001] The present invention relates to the technical field of performance detection of waterborne coatings, and particularly relates to a method, device, electronic device and storage medium for detecting the performance of waterborne coatings. Background Art

[0002] Waterborne coatings use waterborne polymers as film-forming substances, reducing the emissions of VOCs, and having the advantages of safety, no fire hazards, being applicable for construction in humid environments, and easy cleaning. In recent years, due to the continuous improvement of environmental awareness, waterborne coatings have developed rapidly. The evaluation of the inherent performance of waterborne paints is an important indicator determining the quality of waterborne coatings. However, since there are many inherent performances to be detected for waterborne coatings and the detection data obtained by different detection methods are different, it requires very professional personnel to determine the performance of waterborne coatings based on the detection data. Therefore, there is a problem of high difficulty in detecting the performance of existing waterborne coatings. Summary of the Invention

[0003] Embodiments of the present invention provide a method for detecting the performance of waterborne coatings, aiming to solve the problem of high difficulty in detecting the performance of existing waterborne coatings. By unifying the dimensions of the detection information under each performance detection method, the non-visual detection information is unified into a matrix with the same size resolution as the visual detection information, and then the detection information under different dimensions is unified into detection information with matrix dimensions. After fusing the non-visual matrix and the visual matrix, the fused matrix is processed by a trained detection model to obtain the performance detection result of the waterborne coating to be detected, without the need for professional personnel to analyze the detection data, thereby reducing the difficulty of detecting the performance of waterborne coatings.

[0004] In a first aspect, embodiments of the present invention provide a method for detecting the performance of waterborne coatings, the method comprising:

[0005] Obtaining visual detection information and non-visual detection information of the waterborne coating to be detected under each performance detection method, wherein the visual detection information includes a visual matrix;

[0006] Performing encoding processing on the non-visual detection information to obtain a non-visual matrix having the same size resolution as the visual matrix;

[0007] Performing channel fusion on the non-visual matrix and the visual matrix to obtain a fused matrix, wherein each channel in the fused matrix corresponds to one of the visual matrix or the non-visual matrix;

[0008] Inputting the fused matrix into a trained detection model for processing to obtain the performance detection result of the waterborne coating to be detected.

[0009] Optionally, the encoding process of the non-visual detection information to obtain a non-visual matrix with the same size resolution as the visual matrix includes:

[0010] Determine the size resolution of the visual matrix;

[0011] Judge whether the type of the non-visual detection information is a numerical type;

[0012] If the type of the non-visual detection information is a numerical type, encode the non-visual detection information into a non-visual matrix with the size resolution through a first encoding method;

[0013] If the type of the non-visual detection result is a non-numerical type, encode the non-visual detection information into a non-visual matrix with the size resolution through a second encoding method.

[0014] Optionally, the step of if the type of the non-visual detection information is a numerical type, encoding the non-visual detection information into a non-visual matrix with the size resolution through a first encoding method includes:

[0015] If the type of the non-visual detection information is a numerical type, determine the length of the numerical sequence of the non-visual detection information;

[0016] Match a corresponding numerical encoder according to the length of the numerical sequence, and the output size resolution of the numerical encoder is the same as the size resolution;

[0017] Encode the non-visual detection information through the numerical encoder to obtain the non-visual matrix corresponding to the non-visual detection information.

[0018] Optionally, the step of if the type of the non-visual detection result is a non-numerical type, encoding the non-visual detection information into a non-visual matrix with the size resolution through a second encoding method includes:

[0019] If the type of the non-visual detection information is a non-numerical type, judge whether the type of the non-visual detection information is a text type;

[0020] If the type of the non-visual detection information is a text type, perform semantic recognition processing on the non-visual detection information to obtain the text semantic vector of the non-visual detection information;

[0021] Encode the text semantic vector through a preset vector encoder to obtain the non-visual matrix corresponding to the non-visual detection information, and the output size resolution of the vector encoder is the same as the size resolution.

[0022] Optionally, after determining whether the type of the non-visual detection information is a non-numerical type and then determining whether the type of the non-visual detection information is a text type, the method further includes:

[0023] If the type of the non-visual detection information is a non-text type, determine whether the type of the non-visual detection information is a mixed type, where the mixed type includes at least two types among a visual type, a numerical type, and a text type;

[0024] If the type of the non-visual detection information is a mixed type, obtain the weights of each type in the non-visual detection information;

[0025] According to the weights of each type in the non-visual detection information, match a corresponding hybrid encoder, where the hybrid encoder includes a positional attention mechanism, and the output size resolution of the hybrid encoder is the same as the size resolution;

[0026] Perform positional encoding on the non-visual detection information through the positional attention mechanism in the hybrid encoder to obtain a non-visual matrix corresponding to the non-visual detection information.

[0027] Optionally, the step of performing channel fusion on the non-visual matrix and the visual matrix to obtain a fusion matrix includes:

[0028] According to the detection time, the detection sample quantity, and the detection duration of the corresponding performance detection method for the non-visual matrix and the visual matrix, determine a fusion sequence corresponding to the non-visual matrix and the visual matrix;

[0029] Perform channel splicing on the non-visual matrix and the visual matrix in the fusion sequence to obtain a fusion matrix.

[0030] Optionally, before inputting the fusion matrix into a trained detection model for processing to obtain a performance detection result of the waterborne coating to be tested, the method further includes:

[0031] Obtain a training data set and a detection model to be trained, where the training data set includes sample fusion matrices, each sample fusion matrix corresponds to a set of performance labels, the obtaining method of the sample fusion matrix is the same as the obtaining method of the fusion matrix, and the detection model to be trained is a deep neural network model;

[0032] Input the training data set into the detection model to be trained for supervised training, and after training is completed, obtain a trained detection model.

[0033] In a second aspect, an embodiment of the present invention further provides a waterborne coating performance detection device, where the waterborne coating performance detection device includes:

[0034] A first acquisition module, configured to acquire visual detection information and non-visual detection information of the waterborne coating to be tested under various performance detection methods, where the visual detection information includes a visual matrix;

[0035] A first processing module, configured to perform encoding processing on the non-visual detection information to obtain a non-visual matrix having the same size resolution as the visual matrix;

[0036] A fusion module, configured to perform channel fusion on the non-visual matrix and the visual matrix to obtain a fusion matrix, where in the fusion matrix, each channel corresponds to one of the visual matrix or the non-visual matrix;

[0037] A second processing module, configured to input the fusion matrix into a trained detection model for processing to obtain a performance detection result of the waterborne coating to be tested.

[0038] In a third aspect, an embodiment of the present invention provides an electronic device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, where when the processor executes the computer program, the steps in the waterborne coating performance detection method provided by the embodiment of the present invention are implemented.

[0039] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps in the waterborne coating performance detection method provided by the embodiment of the invention are implemented.

[0040] In the embodiment of the present invention, visual detection information and non-visual detection information of the waterborne coating to be tested under various performance detection methods are acquired, where the visual detection information includes a visual matrix; the non-visual detection information is encoded to obtain a non-visual matrix having the same size resolution as the visual matrix; the non-visual matrix and the visual matrix are channel-fused to obtain a fusion matrix, where each channel in the fusion matrix corresponds to one of the visual matrix or the non-visual matrix; the fusion matrix is input into a trained detection model for processing to obtain a performance detection result of the waterborne coating to be tested. By unifying the dimensions of the detection information under various performance detection methods, the non-visual detection information is unified into a matrix with the same size resolution as the visual detection information, and then the detection information in different dimensions is unified into detection information in matrix dimensions. After fusing the non-visual matrix and the visual matrix, the trained detection model processes the fusion matrix to obtain a performance detection result of the waterborne coating to be tested, without the need for professional personnel to analyze the detection data, thereby reducing the difficulty of waterborne coating performance detection. Description of the Drawings

[0041] To more clearly illustrate the technical solutions in the embodiments of the present invention or in the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0042] Figure 1 is a flowchart of a method for detecting the performance of an aqueous coating provided by an embodiment of the present invention;

[0043] Figure 2 is a schematic structural diagram of a device for detecting the performance of an aqueous coating provided by an embodiment of the present invention;

[0044] Figure 3 is a schematic structural diagram of an electronic device provided by an embodiment of the present invention. Detailed implementation manners

[0045] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0046] As Figure 1 shown, Figure 1 is a method flowchart of a method for detecting the performance of an aqueous coating provided by an embodiment of the present invention. The method for detecting the performance of the aqueous coating includes the steps of:

[0047] 101. Obtain the visual detection information and non-visual detection information of the aqueous coating to be tested under various performance detection methods.

[0048] In the embodiments of the present invention, the above performance detection methods may be relevant performance detection methods for existing aqueous coatings. The performance of the above aqueous coating may include solid content, viscosity, hardness, gloss, adhesion, anti-blocking property, fineness, impact resistance, heat resistance, cold resistance, yellowing resistance, temperature change resistance, etc. Each performance may correspond to one or more performance detection methods. The above visual detection information may be data of an image type generated during the detection process, and the above non-visual detection information may be data of a non-image type generated during the detection process. The detection data generated by different performance detection methods may be data of an image type or data of a non-image type.

[0049] Different performances correspond to different performance detection methods. For example, the detection method for solid content can be to take a fixed amount of paint sample, bake it at a predetermined temperature, and obtain the remaining amount, which can be used as the solid content. For example, take 10 g of the waterborne paint sample to be tested, bake it at T = 160 °C for 2 hours, and obtain 3 - 5 g. These 3 - 5 g are the solid content of the waterborne paint to be tested. The above solid content is non - visual detection information.

[0050] The detection method for viscosity can be to take a sample 20 minutes after the waterborne paint is prepared, adjust the sample temperature to 15 °C in winter and 25 °C in summer, and test it with an Iwata #2 cup. Test it again the next day using the same method; through viscosity measurement, the polymerization degree of the waterborne paint to be tested and the use of additives can be observed. Viscosity varies greatly with temperature, so there are significant differences in viscosity in winter, summer, and rainy seasons. The above viscosity is non - visual detection information.

[0051] The detection method for hardness can be carried out according to the provisions of GB / T6739, using a Zhonghua brand 101 drawing pencil for testing; the hardness is divided into 14 grades such as 6H, 5H, 4H, 3H, 2H, H, F, HB, 2B, 3B, 4B, 5B, 6B, etc. The hardness of ordinary paints generally ranges between B - 3H. The hardness of single - component waterborne paint products is about HB, and that of two - component products is about 1H. The above hardness grades are non - visual detection information.

[0052] The detection method for glossiness can be to scrape a film of the waterborne paint to be tested with a certain thickness on a black glass plate, then bake it, and use a light meter to measure the baked black glass plate. For example, scrape a film with a thickness of 250 microns of the waterborne paint to be tested on the black glass plate, bake it at 45 °C for 2 hours, calibrate the light meter first, and then use the light meter to measure the baked black glass plate; Gloss is an index to evaluate the visual impression of the coating surface and is the result of the physical and optical interaction between light and the surface of the object. When light irradiates the surface of the substrate, light reflection will occur. However, if the surface is not smooth, the reflected light will not be in one direction but will be diffused in all directions, resulting in low image quality and giving a feeling of blurriness rather than brightness. Therefore, the gloss of the coating film is low. An optical photograph of the baked black glass plate can be taken to obtain an image of the black glass plate as visual detection information.

[0053] The adhesion can be measured by the comprehensive measurement method or the peel test method. In the comprehensive measurement method, a cutting tool is used to cut the coating layer to the substrate to form scratches of different shapes (grid or cross), and then an adhesive tape is adhered and torn off to check the peeling of the coating layer, and then the adhesion of the coating film is rated. The comprehensive measurement method can be the cross-cut method, the cross-hatch method and the conical mandrel method. The peel test method means that the measured adhesion refers to the force required to cause adhesion failure between the coating layers or between the coating layer and the substrate when a tensile force is applied to the coating layer of the test panel at a specified speed. It can be the peel-back method and the pull-off method; Adhesion is an important indicator of the adhesion degree between waterborne coatings and substrates, and it is also an indicator to characterize whether the waterborne coatings will fall off during long-term use of the product. If the comprehensive measurement method is adopted, the peeling image of the coating layer can be obtained by shooting in the comprehensive measurement method, and this peeling image is visual inspection information. If the peel test method is adopted, the tensile force applied to the coating layer of the test panel is non-visual inspection information.

[0054] The method for detecting anti-blocking property can be to stack multiple plates coated with the waterborne coating to be tested. The plates are placed from top to bottom, ensuring that two plates are in face-to-face contact or back-to-back contact. Apply a certain pressure on the plates, remove the pressure after drying, and separate the plates according to a preset rule. Specifically, six samples can be stacked (the plates are 150mm x 70mm), placed from top to bottom in the order of: one face up, two faces down, one face up, two faces down. Ensure that two plates are in face-to-face contact and two plates are in back-to-back contact. Apply a weight with a diameter of about 70mm and a mass of 500g on the samples. Dry at room temperature (25°C - 30°C) for 20 minutes on the surface, then heat up (to 70°C) and dry for 4 hours; finally, cool at room temperature (25°C - 30°C) for 60 minutes, place in an oven at 50°C for 4 hours, remove the pressure and separate the samples, and check the ease of separation and the degree of damage to the coating surface (visible traces). Generally speaking, the judgment of the anti-blocking property result can be expressed in two forms: "adhesion grade" and "surface damage grade", and each form is divided into different grades according to different degrees. The adhesion grade is divided into A, B, C, D, E. Among them, when the sample separation is free-fall separation, the adhesion grade is A; when the sample separation is slight tapping separation, the adhesion grade is B; when the sample separation is separation after applying a slight pulling force, the adhesion grade is C; when the sample separation is separation after applying a medium pulling force, the adhesion grade is C; when the sample separation is separation after applying a medium pulling force, the adhesion grade is D; when the sample separation is separation after applying a great pulling force, the adhesion grade is E. The surface damage grade is divided into 0, 1, 2, 3, 4. Among them, when there is no damage to the sample surface, the surface damage grade is 0; when the damage to the sample surface is less than 1%, the surface damage grade is 1; when the damage to the sample surface is between 1% - 5%, the surface damage grade is 2; when the damage to the sample surface is between 6% - 20%, the surface damage grade is 3; when the damage to the sample surface is between 21% - 50%, the surface damage grade is 4. Preferably, in the embodiment of the present invention, grading can be not carried out. Specifically, the force for separating the samples can be quantified. For example, the force during free-fall separation is the gravity G, the force during slight tapping separation is the gravity G + instantaneous force F1, and the force during separation after applying a pulling force is the gravity G + pulling force F2. The degree of damage to the above sample surface can also be represented by the surface image of the sample, and there is no need to grade the waterborne coating to be tested according to the degree of damage to the sample surface for the "surface damage grade". The force during separation is non-visual detection information, and the image obtained by photographing the surface of the separated sample is visual detection information.

[0055] The detection method of fineness can be to scrape a certain amount of the waterborne coating sample to be measured with a doctor blade fineness gauge under the condition of meeting the dispersion time to obtain the fineness data. For example, when the dispersion time is sufficient, take an appropriate amount of the waterborne coating sample to be measured and scrape it with a doctor blade fineness gauge, and read the fineness data within ten seconds. Fineness is a standard for checking the size of pigment particles or the degree of dispersion uniformity in waterborne coatings, expressed in micrometers (μm). The smaller the fineness, the smoother the coating film, and vice versa. The above detection method of fineness can obtain non-visual detection information

[0056] The detection method of impact resistance is to detect whether the sample board shows whitening or delamination under impact conditions. In the embodiments of the present invention, the image of the sample board captured under impact conditions is visual detection information.

[0057] The detection method of heat resistance can be to heat the coating sample board and observe the surface changes of the coating sample board. Specifically, the coating sample board can be heated in a blast thermostatic supply box or a high-temperature furnace. After reaching the predetermined temperature and time, then detect its physical and mechanical properties and observe the changes in the surface state of the coating sample board (such as delamination, wrinkling, bubbling, cracking, and discoloration, etc.). The changes in the surface state of the coating sample board can be an image captured. The image of the surface state of the coating sample board is visual detection information.

[0058] The detection method of cold resistance can be to store the coating sample board under low-temperature conditions for a certain period of time and then take it out to observe the changes in the coating sample board. Further, in the embodiments of the present invention, the taken-out coating sample board can be photographed to obtain an image of the coating sample board under low-temperature conditions. The image of the coating sample board under low-temperature conditions is visual detection information.

[0059] The detection method of yellowing resistance can be to irradiate the coating sample board with light and compare the color difference before and after irradiation. Further, in the embodiments of the present invention, the coating sample board before light irradiation can be photographed to obtain a pre-irradiation image, the coating sample board after light irradiation can be photographed to obtain a post-irradiation image, and the post-irradiation image is subtracted from the pre-irradiation image to obtain a color difference map. The color difference map is visual detection information. Yellowing resistance is a characterization index for the color retention performance of coatings during use. The smaller the color difference value, the better the color retention property. A coating with good performance should have a small color difference change under the test conditions. Adding a suitable anti-yellowing agent to the coating formulation can often improve its yellowing resistance.

[0060] The detection method of temperature change resistance can be to test the coating sample board under hot and cold cycling conditions and observe the changes in the coating sample board. Further, in the embodiments of the present invention, the taken-out coating sample board can be photographed to obtain an image of the coating sample board under hot and cold cycling conditions. The image of the coating sample board under hot and cold cycling conditions is visual detection information.

[0061] The method for detecting the properties of the waterborne coating according to the embodiment of the present invention is applied to a server. After obtaining relevant property data through the above property detection method, the property data can be directly uploaded to the server so that the server can obtain the visual detection information and non-visual detection information of the waterborne coating to be tested under each property detection method.

[0062] The above visual detection information may be an image. Since the images obtained by different property detection methods may have different size resolutions, the images obtained by different property detection methods can be linearly transformed to obtain a visual matrix with the same size resolution. Specifically, a fully convolutional network (FCN) can be used to perform full convolution processing on images with different size resolutions to output a visual matrix with the same size resolution.

[0063] If the images obtained by different property detection methods have the same size resolution, the corresponding images can be directly regarded as a visual matrix, and there is no need to linearly transform the images obtained by different property detection methods.

[0064] It should be noted that the above properties of the waterborne coating and the method for detecting the properties of the waterborne coating are only examples for understanding. With the development and standard improvement of the waterborne coating, new properties and new property detection methods may emerge, and the embodiments of the present invention can also be applied to new properties and new property detection methods.

[0065] 102. Encode the non-visual detection information to obtain a non-visual matrix with the same size resolution as the visual matrix.

[0066] In the embodiment of the present invention, when the server receives the visual detection information and non-visual detection information of the waterborne coating to be tested under each property detection method, the non-visual detection information can be encoded through an encoding algorithm preset in the server to obtain a non-visual matrix corresponding to the non-visual detection information. The size resolution of the above visual matrix can be m×n, indicating that the visual matrix has m rows and n columns.

[0067] Specifically, the non-video detection information can be vector-encoded to obtain an encoded vector of the non-video detection information, so that the non-video detection information can be expanded in the spatial dimension, and then the encoded vector is linearly transformed to transform the encoded vector into a matrix form. For example, if the non-video detection information is the hardness grade H, H can be vector-encoded to obtain the encoded vector [h n =(h1, h2,..., h n-1 , h n ), and the encoded vector [h nThe meaning expressed is still that the hardness grade of the water-based coating to be measured is H, and then a linear transformation is performed on the coding vector (h1, h2, …, h n-1 , h n ) to obtain a non-visual matrix [h m,n as follows:

[0068] h 1,1 , h 1,2 , …, h 1,n-1 , h 1,n

[0069] h 2,1 , h 2,2 , …, h 2,n-1 , h 2,n

[0070] …

[0071] H m-1,1 , h m-1,2 , …, h m-1,n-1 , h m-1,n

[0072] h m,1 , h m,2 , …, h m,n-1 , h m,n

[0073] Through the above coding process, a non-visual information can be expressed in the form of a matrix, and the meaning expressed by the non-visual matrix [h m,n is still that the hardness grade of the water-based coating to be measured is H. During the linear transformation process, a (m×1) operator matrix can be multiplied with the coding vector [h n through matrix multiplication to obtain the non-visual matrix [h m,n . In matrix multiplication, the coding vector [h n can be regarded as a (1×n) matrix, and the operator matrix (m×1) is multiplied with the coding vector (1×n) to obtain a (m×1)×(1×n) = (m×n) matrix as the non-visual matrix [h m,n .

[0074] 103. Perform channel fusion on the non-visual matrix and the visual matrix to obtain a fusion matrix.

[0075] In an embodiment of the present invention, after the server encodes non-visual detection information to obtain a non-visual matrix with the same resolution as the visual matrix, the non-visual matrix and the visual matrix can be channel-fused to obtain a fused matrix. A non-visual matrix can be regarded as a channel, and a visual matrix can also be regarded as a channel. Through channel fusion, the non-visual matrix and the visual matrix can be fused to obtain a fused matrix. The above-mentioned channel fusion can be to splice the non-visual matrix and the visual matrix as channels to obtain d×m×n, where d is the total number of channels. In the fused matrix d×m×n, each channel corresponds to a visual matrix or a non-visual matrix.

[0076] 104. Input the fused matrix into the trained detection model for processing to obtain the performance detection result of the waterborne coating to be tested.

[0077] In an embodiment of the present invention, a trained detection model is deployed in the server. After the server performs channel fusion on the visual matrix and the non-visual matrix, the server inputs the fused matrix into the trained detection model for processing to obtain the performance detection result of the waterborne coating to be tested, and sends the performance detection result to the client for the user to view.

[0078] The above-mentioned detection model can be a detection model based on a deep convolutional neural network, such as a detection model based on the R-CNN network, the Faster R-CNN network, the YOLO-V series network, or the centernet network. The standard format of the input data of the above-mentioned detection model can be d×m×n, and the output structure of the above-mentioned detection model can be d + 1, where d is the detection results of d performances, and the additional 1 is the comprehensive performance detection result.

[0079] Specifically, the above-mentioned detection model includes an input layer, a convolutional layer, a downsampling layer, a linear regression layer, and an output layer. The convolutional layer is used to extract the spatial features of each channel in the above-mentioned fused matrix, and the downsampling layer is used to downsample the spatial features extracted by the previous convolutional layer to reduce the dimension of the spatial features. The linear regression layer is used to perform linear regression on the finally extracted features to obtain a result vector, and the output layer is used to classify and output the result vector.

[0080] The output of the above-mentioned linear regression is (d + 1)×k, indicating that the output of the linear regression has d + 1 rows and k columns, and each row represents a feature vector corresponding to a performance.

[0081] The output layer includes d + 1 output neurons. Among them, d output neurons are used to classify d performances and output corresponding detection results, and 1 output neuron is used to classify the comprehensive performance and output corresponding detection results. Specifically, there is a trained classification operator in the output neuron. By multiplying the classification operator with the corresponding feature vector, the corresponding classification feature value is obtained. Each classification feature value corresponds to a performance detection classification. For example, the feature vector is the feature vector 1×k corresponding to the "surface damage level", and the classification operator in the corresponding output neuron is 1×k. Calculate the inner product of the feature vector 1×k and the classification operator 1×k to obtain the corresponding classification feature value y. If the surface damage level corresponding to y is 3, then the surface damage level of 3 is output. Another example is that the feature vector is the feature vector 1×k corresponding to the comprehensive performance, and the classification operator in the corresponding output neuron is 1×k. Calculate the inner product of the feature vector 1×k and the classification operator 1×k to obtain the corresponding classification feature value y. If the comprehensive performance corresponding to y is 0.95 (qualified if greater than 0.8), then the comprehensive performance is output as qualified.

[0082] In the embodiment of the present invention, the visual detection information and non-visual detection information of the water-based paint to be tested under each performance detection method are obtained. The visual detection information includes a visual matrix; the non-visual detection information is encoded to obtain a non-visual matrix with the same size resolution as the visual matrix; the non-visual matrix and the visual matrix are fused in channels to obtain a fusion matrix. In the fusion matrix, each channel corresponds to one of the visual matrix or the non-visual matrix; the fusion matrix is input into a trained detection model for processing to obtain the performance detection result of the water-based paint to be tested. By unifying the dimensions of the detection information under each performance detection method, the non-visual detection information is unified into a matrix with the same size resolution as the visual detection information, and then the detection information in different dimensions is unified into detection information in matrix dimensions. After fusing the non-visual matrix and the visual matrix, the trained detection model processes the fusion matrix to obtain the performance detection result of the water-based paint to be tested, without the need for professional personnel to analyze the detection data, thus reducing the difficulty of the performance detection of water-based paints.

[0083] Optionally, in the step of encoding the non-visual detection information to obtain a non-visual matrix with the same size resolution as the visual matrix, the size resolution of the visual matrix can be determined; it is judged whether the type of the non-visual detection information is a numerical type; if the type of the non-visual detection information is a numerical type, the non-visual detection information is encoded into a non-visual matrix with size resolution through a first encoding method; if the type of the non-visual detection result is a non-numerical type, the non-visual detection information is encoded into a non-visual matrix with size resolution through a second encoding method.

[0084] In an embodiment of the present invention, the non-visual detection information of the numerical type may be the detection information obtained by the detection method of solid content, the detection method of viscosity, the peeling test method of adhesion, and the detection method of fineness. The above non-visual detection information of the non-numerical type may be the detection information obtained by the detection method of hardness and the detection method of anti-blocking property.

[0085] After the server obtains the visual detection information and non-visual detection information of the waterborne coating to be tested under various performance detection methods, it can first determine the size resolution of the visual matrix. The size resolution of the above visual matrix can be determined according to the input size of the detection model. For example, if the input size of the detection model is m×n, then the size resolution of the above visual matrix is m×n, where m represents the number of rows, n represents the number of columns, and m×n represents the resolution, specifically indicating that there are m×n matrix units in the visual matrix. In the visual matrix, the value of the matrix unit can be a color value, and in the non-visual matrix, the value of the matrix unit can be the corresponding encoded value.

[0086] In a possible embodiment, there are images with different size resolutions in the above visual detection information, and the images with different size resolutions can be linearly transformed to obtain a visual matrix with a size resolution of m×n. Of course, the images with different size resolutions can also be scaled to obtain an image with a size resolution of m×n as the visual matrix.

[0087] After determining the size resolution of the visual matrix, the server can determine whether the type of the non-visual information is a numerical type. If it is determined that there is a non-visual information type that is a numerical type, the server will call the first encoding method to encode the non-visual information of the numerical type to obtain a non-visual matrix corresponding to the non-visual information of the numerical type.

[0088] Specifically, the above first encoding method may be to first determine the number of numerical values i in the non-visual information, and then match the encoding operator corresponding to the number of numerical values to encode the non-visual information. For the encoding operator corresponding to the number of numerical values i, its structure is i×m×n. The above non-visual information can be understood as a 1×i matrix. Multiplying the 1×i matrix by i×m×n can obtain a 1×m×n matrix, that is, an m×n matrix, and this matrix is the non-visual matrix corresponding to the non-visual information.

[0089] If the type of the non-visual information is a non-numerical type, the server will call the second encoding method to encode the non-visual information of the non-numerical type. The above non-numerical type may include text type and mixed type. Specifically, when the type of the non-visual information is a text type, the text type of the non-visual information can be encoded by the text encoding method, and when the type of the non-visual information is a mixed type, the mixed type of the non-visual information can be encoded by the mixed encoding method.

[0090] In an embodiment of the present invention, non-visual information is encoded into a matrix form through a corresponding encoding method to obtain a corresponding non-visual matrix. The non-visual matrix has the same size resolution as the visual matrix. Thus, by means of channel fusion of the non-visual matrix and the visual matrix, the visual information and the non-visual information are fused in the same dimension to obtain a fusion matrix that combines the visual information and the non-visual information. Only by using the fusion matrix to detect the performance of the waterborne coating can the detection speed of the waterborne coating performance detection be improved. At the same time, without the need for relevant personnel to view and analyze the detection data of each performance, the waterborne coating performance detection result can be obtained.

[0091] Optionally, in the step of encoding the non-visual detection information into a non-visual matrix with a size resolution through the first encoding method when the type of the non-visual detection information is a numerical type, if the type of the non-visual detection information is a numerical type, the numerical sequence length of the non-visual detection information can be determined; a corresponding numerical encoder is matched according to the numerical sequence length, and the output size resolution of the numerical encoder is the same as the size resolution; the non-visual detection information is encoded through the numerical encoder to obtain the non-visual matrix corresponding to the non-visual detection information.

[0092] In an embodiment of the present invention, the type of the above non-visual detection information is a numerical type, the above non-visual detection information includes a plurality of numerical values, the plurality of numerical values are arranged in the corresponding order according to the corresponding performance detection method, and the plurality of numerical values in the above non-visual detection information are a numerical sequence. The above first encoding method may be to first determine the numerical sequence length in the non-visual information, and then match a numerical encoder corresponding to the numerical sequence length to perform encoding processing on the non-visual information. In the server, numerical encoders corresponding to different numerical sequence lengths are stored, and after determining the numerical sequence length in the non-visual information, a numerical encoder corresponding to the numerical sequence length can be matched.

[0093] The above numerical encoder includes a corresponding encoding table. The encoding table includes the encoding vector [h n corresponding to the detection method, the encoding vector corresponding to the numerical sequence length, and the encoding vectors corresponding to each numerical value. The detection method, numerical sequence length, and numerical sequence of the non-visual detection information can be encoded through the encoding table to correspondingly obtain the encoding vector of the method number, the encoding vector of the sequence length, and the encoding vector of the numerical value. The encoding vector of the method number, the encoding vector of the sequence length, and the encoding vector of the numerical value are spliced to obtain the splicing vector corresponding to the non-visual detection information. The splicing vector can be i×n, where i represents the number of encoding vectors and n represents the length of the encoding vector. The splicing vector is encoded through the encoding operator in the numerical encoder.

[0094] For the encoding operator in the numerical encoder, its structure is i×m. By performing matrix multiplication on the transposed matrix m×i of the encoding operator and the above-mentioned concatenated vector i×n, a matrix of m×n can be obtained as the non-visual matrix corresponding to the non-visual information.

[0095] Optionally, in the step of encoding non-visual detection information into a non-visual matrix with dimensional resolution by the second encoding method when the type of the non-visual detection result is a non-numerical type, if the type of the non-visual detection information is a non-numerical type, it can be determined whether the type of the non-visual detection information is a text type; if the type of the non-visual detection information is a text type, semantic recognition processing can be performed on the non-visual detection information to obtain the text semantic vector of the non-visual detection information; the text semantic vector is encoded by a preset vector encoder to obtain the non-visual matrix corresponding to the non-visual detection information, and the output dimensional resolution of the vector encoder is the same as the dimensional resolution.

[0096] In the embodiments of the present invention, the above non-numerical type of non-visual detection information can be detection information obtained by a hardness detection method, an anti-adhesion detection method, etc. The above text type of non-visual detection information can be detection information obtained by a hardness detection method. The above non-text type of non-visual detection information can be detection information obtained by an anti-adhesion detection method.

[0097] After the server determines that the non-visual detection information is of a non-numerical type, it can further determine whether the type of the non-visual detection information is a text type. If the type of the non-visual detection information is a text type, semantic recognition processing can be performed on the non-visual detection information by a natural language processing model deployed in the server to obtain the text semantic vector of the non-visual detection information.

[0098] Specifically, the above natural language processing model can be the semantic extraction part in language models such as Neural Net Language Model (NNLM) model, Recurrent Neural Net Language Model (RNN) model, etc.

[0099] Furthermore, semantic recognition processing can also be performed on the text type of non-visual detection information by word2vec to obtain the text semantic vector of the non-visual detection information. By word2vec, each word in the non-visual detection information can be mapped to an L-dimensional space to obtain the word vector corresponding to each word, and the word vectors are concatenated in text order to obtain the text semantic vector R×L, where R represents the number of word vectors and L represents the length of the word vector.

[0100] After obtaining the text semantic vector R×L of the non-visual detection information, the server can transform the text semantic vector R×L into an m×n matrix by means of linear transformation, and use the m×n matrix as the non-visual matrix of the non-visual detection information.

[0101] The above vector encoder includes a first encoding operator L×n and a second encoding operator m×R. The text semantic vector R×L can be subjected to matrix multiplication processing with the first encoding operator L×n to obtain an intermediate matrix R×n. The second encoding operator m×R is subjected to matrix multiplication processing with the intermediate matrix R×n to obtain a final matrix m×n, and the final matrix m×n is used as the non-visual matrix of the non-visual detection information.

[0102] Optionally, after the step of determining whether the type of the non-visual detection information is a non-text type if the type of the non-visual detection information is a non-numerical type, if the type of the non-visual detection information is a non-text type, it can be determined whether the type of the non-visual detection information is a mixed type. The mixed type includes at least two types of visual type, numerical type, and text type; if the type of the non-visual detection information is a mixed type, the weights of each type in the non-visual detection information can be obtained; according to the weights of each type in the non-visual detection information, a corresponding hybrid encoder is matched. The hybrid encoder includes a position attention mechanism, and the output size resolution of the hybrid encoder is the same as the size resolution; the non-visual detection information is positionally encoded through the position attention mechanism in the hybrid encoder to obtain the non-visual matrix corresponding to the non-visual detection information.

[0103] In the embodiment of the present invention, when the server determines that the non-visual detection information is a non-text type, it can further determine whether the type of the non-visual detection information is a mixed type. For example, in the anti-adhesion detection method, the force for separating the sample is a numerical type, and the image obtained by photographing the surface of the separated sample is a visual type.

[0104] In a possible embodiment, if it is determined that the type of the non-visual detection information is a non-mixed type, the accurate and compliant non-visual detection information can be submitted by the detector corresponding to the non-visual detection information.

[0105] For the non-visual detection information of the mixed type, the weights of each type in the non-visual detection information can be obtained. According to the weights of each type in the non-visual detection information, a corresponding hybrid encoder is matched in the server. The above hybrid encoder is dedicated to encoding the non-visual detection information of the mixed type, so that the detection information of each type in the non-visual detection information is fused into an m×n matrix.

[0106] Specifically, the hybrid encoder includes a position attention mechanism, which can assign position attention parameters to different types of detection information. Through the position attention parameters, the positions of various types of detection information can be characterized in an m×n matrix. The above-mentioned hybrid encoder includes different types of encoding operators, such as encoding operators for numerical types, text types, and encoding operators for image linear transformation. Each type of encoding operator corresponds to a different weight. During the encoding process, for the matrix multiplication of each type of encoding operator, the corresponding weight of that type also needs to be multiplied.

[0107] The above weights can be determined according to the importance of each type of information in different performance detection methods. This importance can be determined by relevant experts or by dimensionality reduction processing using the principal component analysis method.

[0108] Optionally, in the step of performing channel fusion on the non-visual matrix and the visual matrix to obtain a fusion matrix, the fusion sequence corresponding to the non-visual matrix and the visual matrix can be determined according to the detection time, the detection sample size, and the detection duration of the corresponding performance detection method for the non-visual matrix and the visual matrix; the non-visual matrix and the visual matrix in the fusion sequence are concatenated in channels to obtain a fusion matrix.

[0109] In the embodiments of the present invention, the above detection time is the time from when the waterborne coating to be tested is produced to when the corresponding performance detection starts. The above detection sample size is the amount of the waterborne coating to be tested used in the corresponding performance detection method. The above detection duration is the length of time that the corresponding performance detection method lasts.

[0110] Specifically, the priority order of each detection information can be determined through a priority value, and the fusion sequence corresponding to the non-visual matrix and the visual matrix is determined according to the priority order. More specifically, the shorter the detection time, the larger the priority value; the longer the detection time, the smaller the priority value; the more the detection sample size, the larger the priority value; the smaller the detection sample size, the smaller the priority value; the longer the detection duration, the larger the priority value; the longer the detection duration, the smaller the priority value. The above priority value can be expressed by the following formula:

[0111] S = lnM - T1 + cos(T1) - T2 + cos(T2)

[0112] Wherein, the above S is the priority value, the above M is the sample size (by mass), the above T1 is the detection time, and the above T2 is the detection duration. Each detection information corresponds to a priority value. The non-visual detection information and the visual detection information are sorted from largest to smallest according to the priority value, and the corresponding non-visual matrix and visual matrix are concatenated in channels according to the above sorting to obtain a fusion matrix d×m×n.

[0113] Optionally, before the step of inputting the fusion matrix into the trained detection model for processing to obtain the performance detection result of the waterborne coating to be tested, a training data set and a detection model to be trained can also be obtained. The training data set includes sample fusion matrices, and each sample fusion matrix corresponds to a set of performance labels. The method for obtaining the sample fusion matrix is the same as the method for obtaining the fusion matrix. The detection model to be trained is a deep neural network model. The training data set is input into the detection model to be trained for supervised training, and after training is completed, a trained detection model is obtained.

[0114] In the embodiment of the present invention, the above-mentioned detection model to be trained is a detection model based on a deep convolutional neural network. For example, it can be a detection model based on the R-CNN network, Faster R-CNN network, YOLO-V series network, or centernet network. The standard format of the input data of the above-mentioned detection model to be trained can be d×m×n, and the output structure of the above-mentioned detection model to be trained can be d + 1, where d is the detection results of d performances, and the additional 1 is the comprehensive performance detection result.

[0115] Specifically, the above-mentioned detection model includes an input layer, a convolutional layer, a downsampling layer, a linear regression layer, and an output layer. The convolutional layer is used to extract the spatial features of each channel in the above-mentioned fusion matrix, and the downsampling layer is used to downsample the spatial features extracted by the previous convolutional layer to reduce the dimension of the spatial features. The linear regression layer is used to perform linear regression on the finally extracted features to obtain a result vector, and the output layer is used to classify and output the result vector.

[0116] The above-mentioned training data set includes sample fusion matrices, and each sample fusion matrix corresponds to a set of performance labels. A set of performance labels includes the performance labels corresponding to each performance detection method and a comprehensive performance label. The above-mentioned performance labels and comprehensive performance labels can be obtained by expert annotation. The method for obtaining the above-mentioned sample fusion matrix is the same as the method for obtaining the fusion matrix in the above-mentioned embodiment, and will not be elaborated here.

[0117] The training data set is input into the detection model to be trained for supervised training, and after the training is completed, the trained detection model is obtained. During the training process, cross-entropy loss is used to calculate the loss function. A sample fusion matrix is input into the detection model to be trained, and the sample fusion matrix is processed by the detection model to be trained to obtain the result of the sample fusion matrix. The result of the sample fusion matrix is calculated with the performance label of the sample fusion matrix to obtain the loss function between the result of the sample fusion matrix and the performance label of the sample fusion matrix. The smaller the loss function, the closer the result of the sample fusion matrix is to the performance label of the sample fusion matrix. Taking the minimization of the loss function as the optimization goal, the parameters of the detection model to be trained are adjusted through the backpropagation algorithm, and the above process is iterated. When the loss function of the detection model to be trained converges at the minimum or reaches the preset number of iterations, the training ends, and the trained detection model is obtained and deployed to the server.

[0118] It should be noted that the waterborne coating performance detection method provided by the embodiments of the present invention can be applied to devices such as smartphones, computers, and servers that can perform waterborne coating performance detection.

[0119] As Figure 2 shown, the embodiments of the present invention provide a waterborne coating performance detection device, and the waterborne coating performance detection device includes:

[0120] A first acquisition module 201, configured to acquire visual detection information and non-visual detection information of the waterborne coating to be detected under each performance detection method, where the visual detection information includes a visual matrix;

[0121] A first processing module 202, configured to perform encoding processing on the non-visual detection information to obtain a non-visual matrix having the same size resolution as the visual matrix;

[0122] A fusion module 203, configured to perform channel fusion on the non-visual matrix and the visual matrix to obtain a fusion matrix, where in the fusion matrix, each channel corresponds to one of the visual matrix or the non-visual matrix;

[0123] A second processing module 204, configured to input the fusion matrix into the trained detection model for processing to obtain the performance detection result of the waterborne coating to be detected.

[0124] Optionally, the first processing module 202 includes:

[0125] A first determination sub-module, configured to determine the size resolution of the visual matrix;

[0126] A judgment sub-module, configured to judge whether the type of the non-visual detection information is a numerical type;

[0127] The first processing sub-module is used to encode the non-visual detection information into a non-visual matrix with the size resolution through a first encoding method if the type of the non-visual detection information is a numerical type;

[0128] The second processing sub-module is used to encode the non-visual detection information into a non-visual matrix with the size resolution through a second encoding method if the type of the non-visual detection result is a non-numerical type.

[0129] Optionally, the first processing sub-module includes:

[0130] The first processing unit is used to determine the length of the numerical sequence of the non-visual detection information if the type of the non-visual detection information is a numerical type;

[0131] The second processing unit is used to match a corresponding numerical encoder according to the length of the numerical sequence, and the output size resolution of the numerical encoder is the same as the size resolution;

[0132] The first encoding unit is used to encode the non-visual detection information through the numerical encoder to obtain the non-visual matrix corresponding to the non-visual detection information.

[0133] Optionally, the second processing sub-module includes:

[0134] The third processing unit is used to judge whether the type of the non-visual detection information is a text type if the type of the non-visual detection information is a non-numerical type;

[0135] The fourth processing unit is used to perform semantic recognition processing on the non-visual detection information to obtain the text semantic vector of the non-visual detection information if the type of the non-visual detection information is a text type;

[0136] The second encoding unit is used to encode the text semantic vector through a preset vector encoder to obtain the non-visual matrix corresponding to the non-visual detection information, and the output size resolution of the vector encoder is the same as the size resolution.

[0137] Optionally, the device further includes:

[0138] The judgment module is used to judge whether the type of the non-visual detection information is a mixed type if the type of the non-visual detection information is a non-text type, and the mixed type includes at least two types of visual type, numerical type and text type;

[0139] The third processing module is used to obtain the weights of each type in the non-visual detection information if the type of the non-visual detection information is a mixed type;

[0140] A matching module, configured to match a corresponding hybrid encoder according to the weights of various types in the non-visual detection information, where the hybrid encoder includes a position attention mechanism, and the output size resolution of the hybrid encoder is the same as the size resolution;

[0141] An encoding module, configured to perform position encoding on the non-visual detection information through the position attention mechanism in the hybrid encoder to obtain a non-visual matrix corresponding to the non-visual detection information.

[0142] Optionally, the fusion module 203 includes:

[0143] A determination sub-module, configured to determine a fusion sequence corresponding to the non-visual matrix and the visual matrix according to the detection time, the detection sample quantity, and the detection duration of the corresponding performance detection method of the non-visual matrix and the visual matrix;

[0144] A fusion sub-module, configured to perform channel splicing on the non-visual matrix and the visual matrix in the fusion sequence to obtain a fusion matrix.

[0145] Optionally, the apparatus further includes:

[0146] A second acquisition module, configured to acquire a training data set and a detection model to be trained, where the training data set includes sample fusion matrices, each sample fusion matrix corresponds to a set of performance labels, the acquisition method of the sample fusion matrices is the same as the acquisition method of the fusion matrix, and the detection model to be trained is a deep neural network model;

[0147] A training module, configured to input the training data set into the detection model to be trained for supervised training, and obtain a trained detection model after the training is completed.

[0148] It should be noted that the waterborne coating performance detection device provided in the embodiments of the present invention can be applied to devices such as smartphones, computers, and servers that can perform waterborne coating performance detection.

[0149] The waterborne coating performance detection device provided in the embodiments of the present invention can implement each process implemented by the waterborne coating performance detection method in the above method embodiments, and can achieve the same beneficial effects. To avoid repetition, it will not be elaborated here.

[0150] See Figure 3 , Figure 3 is a schematic structural diagram of an electronic device provided in an embodiment of the present invention. As Figure 3 shown, it includes: a memory 302, a processor 301, and a computer program of the waterborne coating performance detection method stored on the memory 302 and executable on the processor 301, where:

[0151] The processor 301 is used to call the computer program stored in the memory 302 and execute the following steps:

[0152] Obtain the visual detection information and non-visual detection information of the water-based coating to be tested under various performance detection methods, where the visual detection information includes a visual matrix;

[0153] Encode the non-visual detection information to obtain a non-visual matrix with the same size resolution as the visual matrix;

[0154] Perform channel fusion on the non-visual matrix and the visual matrix to obtain a fusion matrix. In the fusion matrix, each channel corresponds to one of the visual matrix or the non-visual matrix;

[0155] Input the fusion matrix into the trained detection model for processing to obtain the performance detection result of the water-based coating to be tested.

[0156] Optionally, the encoding process of the non-visual detection information by the processor 301 to obtain a non-visual matrix with the same size resolution as the visual matrix includes:

[0157] Determine the size resolution of the visual matrix;

[0158] Judge whether the type of the non-visual detection information is a numerical type;

[0159] If the type of the non-visual detection information is a numerical type, encode the non-visual detection information into a non-visual matrix with the size resolution through a first encoding method;

[0160] If the type of the non-visual detection result is a non-numerical type, encode the non-visual detection information into a non-visual matrix with the size resolution through a second encoding method.

[0161] Optionally, the step that if the type of the non-visual detection information is a numerical type, the processor 301 encodes the non-visual detection information into a non-visual matrix with the size resolution through a first encoding method includes:

[0162] If the type of the non-visual detection information is a numerical type, determine the numerical sequence length of the non-visual detection information;

[0163] Match a corresponding numerical encoder according to the numerical sequence length, and the output size resolution of the numerical encoder is the same as the size resolution;

[0164] Encode the non-visual detection information through the numerical encoder to obtain the non-visual matrix corresponding to the non-visual detection information.

[0165] Optionally, when the type of the non-visual detection result is a non-numerical type, encoding the non-visual detection information into a non-visual matrix with the size resolution through a second encoding method by the processor 301 includes:

[0166] When the type of the non-visual detection information is a non-numerical type, determine whether the type of the non-visual detection information is a text type;

[0167] When the type of the non-visual detection information is a text type, perform semantic recognition processing on the non-visual detection information to obtain a text semantic vector of the non-visual detection information;

[0168] Encode the text semantic vector through a preset vector encoder to obtain a non-visual matrix corresponding to the non-visual detection information, and an output size resolution of the vector encoder is the same as the size resolution.

[0169] Optionally, after determining whether the type of the non-visual detection information is a text type when the type of the non-visual detection information is a non-numerical type, the method further includes performed by the processor 301:

[0170] When the type of the non-visual detection information is a non-text type, determine whether the type of the non-visual detection information is a mixed type, where the mixed type includes at least two types of a visual type, a numerical type, and a text type;

[0171] When the type of the non-visual detection information is a mixed type, obtain weights of each type in the non-visual detection information;

[0172] Match a corresponding hybrid encoder according to the weights of each type in the non-visual detection information, where the hybrid encoder includes a position attention mechanism, and an output size resolution of the hybrid encoder is the same as the size resolution;

[0173] Perform position encoding on the non-visual detection information through the position attention mechanism in the hybrid encoder to obtain a non-visual matrix corresponding to the non-visual detection information.

[0174] Optionally, the method of fusing the non-visual matrix and the visual matrix into a fusion matrix performed by the processor 301 includes:

[0175] Determine a fusion sequence corresponding to the non-visual matrix and the visual matrix according to detection time, detection sample quantity, and detection duration of a corresponding performance detection method of the non-visual matrix and the visual matrix;

[0176] Channel splice the non-visual matrix and the visual matrix in the fusion sequence to obtain a fusion matrix.

[0177] Optionally, before inputting the fusion matrix into the trained detection model for processing to obtain the performance detection result of the waterborne coating to be measured, the method executed by the processor 301 further includes:

[0178] Obtain a training data set and a detection model to be trained. The training data set includes sample fusion matrices, and each sample fusion matrix corresponds to a set of performance labels. The acquisition method of the sample fusion matrix is the same as that of the fusion matrix. The detection model to be trained is a deep neural network model;

[0179] Input the training data set into the detection model to be trained for supervised training, and after training is completed, obtain a trained detection model.

[0180] It should be noted that the electronic device provided in the embodiments of the present invention can be applied to devices such as smartphones, computers, and servers that can perform the waterborne coating performance detection method.

[0181] The electronic device provided in the embodiments of the present invention can implement each process implemented by the waterborne coating performance detection method in the above method embodiments, and can achieve the same beneficial effects. To avoid repetition, it will not be elaborated here.

[0182] The embodiments of the present invention further provide a computer-readable storage medium. A computer program is stored on the computer-readable storage medium. When the computer program is executed by a processor, it implements each process of the waterborne coating performance detection method or the application-side waterborne coating performance detection method provided in the embodiments of the present invention, and can achieve the same technical effects. To avoid repetition, it will not be elaborated here.

[0183] Those of ordinary skill in the art can understand that all or part of the processes of implementing the above method embodiments can be completed by instructing relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the above method embodiments. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM), etc.

[0184] The above-disclosed are only the preferred embodiments of the present invention. Of course, the scope of the rights of the present invention cannot be limited by this. Therefore, equivalent changes made according to the claims of the present invention still fall within the scope covered by the present invention.

Claims

1. A method for detecting the performance of an aqueous coating, characterized in that, The properties of the waterborne coating include solid content, viscosity, hardness, glossiness, adhesion, anti-blocking property, fineness, impact resistance, heat resistance, cold resistance, yellowing resistance, and temperature change resistance. The method includes the following steps: Obtain the visual detection information and non-visual detection information of the waterborne coating to be tested under various property detection methods. The visual detection information includes a visual matrix; Determine the size resolution of the visual matrix; judge whether the type of the non-visual detection information is a numerical type; if the type of the non-visual detection information is a numerical type, encode the non-visual detection information into a non-visual matrix with the size resolution through a first encoding method; if the type of the non-visual detection result is a non-numerical type, encode the non-visual detection information into a non-visual matrix with the size resolution through a second encoding method; Perform channel fusion on the non-visual matrix and the visual matrix to obtain a fusion matrix. In the fusion matrix, each channel corresponds to one of the visual matrix or the non-visual matrix; the step of performing channel fusion on the non-visual matrix and the visual matrix to obtain a fusion matrix includes: determining a fusion sequence corresponding to the non-visual matrix and the visual matrix according to the detection time, detection sample quantity, and detection duration of the corresponding property detection method of the non-visual matrix and the visual matrix; performing channel splicing on the non-visual matrix and the visual matrix in the fusion sequence to obtain a fusion matrix; Obtain a training data set and a detection model to be trained. The training data set includes sample fusion matrices, and each sample fusion matrix corresponds to a set of property labels. The method for obtaining the sample fusion matrix is the same as the method for obtaining the fusion matrix. The detection model to be trained is a deep neural network model; input the training data set into the detection model to be trained for supervised training, and after training is completed, obtain a trained detection model; input the fusion matrix into the trained detection model for processing to obtain the property detection result of the waterborne coating to be tested.

2. The waterborne coating performance detection method according to claim 1, characterized in that The step of, if the type of the non-visual detection information is a numerical type, encoding the non-visual detection information into a non-visual matrix with the size resolution through a first encoding method, includes: If the type of the non-visual detection information is a numerical type, determine the numerical sequence length of the non-visual detection information; Match a corresponding numerical encoder according to the numerical sequence length. The output size resolution of the numerical encoder is the same as the size resolution; Encode the non-visual detection information through the numerical encoder to obtain the non-visual matrix corresponding to the non-visual detection information.

3. The waterborne coating performance detection method according to claim 2, wherein, The step of, if the type of the non-visual detection result is a non-numerical type, encoding the non-visual detection information into a non-visual matrix with the size resolution through a second encoding method, includes: If the type of the non-visual detection information is a non-numerical type, judge whether the type of the non-visual detection information is a text type; If the type of the non-visual detection information is a text type, semantic recognition processing is performed on the non-visual detection information to obtain a text semantic vector of the non-visual detection information; The text semantic vector is encoded through a preset vector encoder to obtain a non-visual matrix corresponding to the non-visual detection information, and the output size resolution of the vector encoder is the same as the size resolution.

4. The waterborne coating performance detection method according to claim 3, characterized in that, After determining that if the type of the non-visual detection information is a non-numerical type, it is determined whether the type of the non-visual detection information is a text type, the method further includes: If the type of the non-visual detection information is a non-text type, it is determined whether the type of the non-visual detection information is a mixed type, and the mixed type includes at least two types among a visual type, a numerical type, and a text type; If the type of the non-visual detection information is a mixed type, weights of each type in the non-visual detection information are obtained; According to the weights of each type in the non-visual detection information, a corresponding hybrid encoder is matched, and the hybrid encoder includes a position attention mechanism, and the output size resolution of the hybrid encoder is the same as the size resolution; The non-visual detection information is position-encoded through the position attention mechanism in the hybrid encoder to obtain a non-visual matrix corresponding to the non-visual detection information.

5. An aqueous coating performance detection device, characterized in that, An aqueous coating performance detection device is used to execute the aqueous coating performance detection method according to any one of claims 1 to 4, and the aqueous coating performance includes solid content, viscosity, hardness, glossiness, adhesion, anti-blocking property, fineness, impact resistance, heat resistance, cold resistance, yellowing resistance, and temperature change resistance. The aqueous coating performance detection device includes: A first acquisition module, configured to acquire visual detection information and non-visual detection information of a to-be-detected aqueous coating under each performance detection method, where the visual detection information includes a visual matrix; A first processing module, configured to determine the size resolution of the visual matrix; determine whether the type of the non-visual detection information is a numerical type; if the type of the non-visual detection information is a numerical type, encode the non-visual detection information into a non-visual matrix with the size resolution through a first encoding method; if the type of the non-visual detection result is a non-numerical type, encode the non-visual detection information into a non-visual matrix with the size resolution through a second encoding method; A fusion module, configured to perform channel fusion on the non-visual matrix and the visual matrix to obtain a fusion matrix, and in the fusion matrix, each channel corresponds to one of the visual matrix or the non-visual matrix; performing channel fusion on the non-visual matrix and the visual matrix to obtain a fusion matrix includes: determining a fusion sequence corresponding to the non-visual matrix and the visual matrix according to the detection time, detection sample quantity, and detection duration of the corresponding performance detection method of the non-visual matrix and the visual matrix; performing channel splicing on the non-visual matrix and the visual matrix in the fusion sequence to obtain a fusion matrix; A second acquisition module, configured to acquire a training data set and a detection model to be trained, wherein the training data set includes sample fusion matrices, each sample fusion matrix corresponds to a set of performance labels, the acquisition method of the sample fusion matrix is the same as that of the fusion matrix, and the detection model to be trained is a deep neural network model; A training module, configured to input the training data set into the detection model to be trained for supervised training, and obtain a trained detection model after the training is completed; A second processing module, configured to input the fusion matrix into the trained detection model for processing, and obtain a performance detection result of the water-based paint to be measured.

6. An electronic device, characterized in that, Comprising: A memory, a processor, and a computer program stored on the memory and executable on the processor, wherein when the processor executes the computer program, the steps in the water-based paint performance detection method according to any one of claims 1 to 4 are implemented.

7. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium, and when the computer program is executed by the processor, the steps in the water-based paint performance detection method according to any one of claims 1 to 4 are implemented.

Images