Coating film quality detection system for intelligent mirror board glass

Abstract

The invention discloses a coating quality detection system for intelligent mirror board glass, and relates to the technical field of machine vision, and the coating quality detection system comprises the following steps: obtaining a coating image of target mirror board glass; substituting the coating image into an image recognition model to obtain a coating result; judging a film coating result; if the film coating result is unqualified, emitting a beam of polarized light through spectrum equipment, and irradiating the target mirror board glass at a preset angle to obtain a spectrogram; substituting the spectrogram into a multi-layer film optical model to obtain a thickness value of each layer of film in the coating film; the coating film is composed of multiple layers of films; and calculating the thickness error of each layer of film according to the thickness value of each layer of film and the preset thickness value of each layer of film, comparing the thickness error of each layer of film with the preset error thickness value of each layer of film to obtain a coating thickness state, and sending the coating thickness state to the terminal. According to the invention, the problem that the thicknesses of all layers of films are not uniform is solved, and the detection efficiency of the coating quality is improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of machine vision, and particularly relates to a film coating quality detection system for intelligent mirror panel glass. BACKGROUND

[0002] Machine vision technology is responsible for collecting and analyzing the visual information of the object to be detected in the field of industrial detection, so as to ensure that the information can accurately identify the target features. Machine vision technology converts the physical object into a digital image that can be analyzed through a series of vision devices such as industrial cameras, lenses, light sources, image processors, etc., and combines algorithm models to extract and determine image features, forming an intelligent detection process to evaluate product quality, and providing precision for quality control and efficiency improvement of industrial production.

[0003] The prior art (publication number: CN117191747B) discloses a detection system for the quality of the coating film of float glass, which comprises a workbench and a photosensitive instrument arranged on the workbench for detecting the light transmittance of the coating film. A carrying assembly for carrying the glass is slidably connected to the workbench along the length direction of the workbench. A first moving mechanism is arranged in the workbench for driving the carrying assembly to move. A fixing frame is slidably connected to the workbench along the length direction. A cleaning block for wiping the glass is slidably connected to the lower part of the fixing frame along the length direction of the fixing frame. A second light source is installed on the cleaning block. A second moving mechanism is arranged in the workbench for driving the fixing frame to move. The surface of the glass is cleaned by the cleaning block to remove dust and dirt on the surface of the glass. Then the glass is irradiated by the second light source. The scratches and wear on the coating film of the glass are displayed by the photosensitive instrument to detect whether the quality of the glass coating film meets the standard.

[0004] The above-mentioned patent mainly detects the scratches and wear on the glass coating film to evaluate the quality of the coating film. However, the existing intelligent glass usually adopts multiple layers of film, and stress occurs between the thicknesses of the layers, which leads to uneven thickness of the layers and reduces the detection efficiency of the coating film quality. SUMMARY

[0005] The purpose of the present application is to solve the problem of uneven thickness of the layers of the existing intelligent glass, which leads to uneven thickness of the layers and reduces the detection efficiency of the coating film quality. A film coating quality detection system for intelligent mirror panel glass is proposed.

[0006] In the first aspect of the present application, a film coating quality detection system for intelligent mirror panel glass is proposed, which comprises: An image recognition module: a film coating image is substituted into an image recognition model to obtain a film coating result; The first coating judgment module judges the coating result, if the coating result is unqualified, the target mirror panel glass is marked as abnormal and sent to the terminal, if the coating result is qualified, a polarized light is emitted through a spectrum device, and a spectrum diagram is obtained by irradiating the target mirror panel glass at a preset angle; The multi-layer film optical module substitutes the spectrum diagram into a multi-layer film optical model to obtain thickness values of each layer film in the coating; the coating is composed of multiple layers of films; The second coating judgment module calculates the thickness error of each layer film by comparing the thickness values of each layer film with the preset thickness values of each layer film, compares the thickness error of each layer film with the preset thickness error value of each layer film to obtain the coating thickness state, and sends the coating thickness state to the terminal.

[0007] Optionally, the image recognition module is improved based on the original YOLOv8 model, and the specific improvements include: In the backbone structure, the fourth layer, the sixth layer and the eighth layer of the original YOLOv8 model are replaced by a multi-scale enhancement module; The multi-scale enhancement module is composed of a plurality of multi-scale enhancement units connected in sequence; The network structure of the multi-scale enhancement unit is as follows: The input feature of the multi-scale enhancement unit is determined as the input feature, the first deep convolution feature is obtained by substituting the input feature into the DConv layer, the first full connection feature is obtained by substituting the first deep convolution feature into the full connection layer, the attention feature is obtained by substituting the full connection feature into the cascaded group attention module, and the second deep convolution feature is obtained by substituting the attention feature into the DConv layer; the second full connection feature is obtained by substituting the obtained second deep convolution feature into the full connection layer; The input feature and the second full connection feature are fused to obtain the output feature.

[0008] Optionally, the original YOLOv8 model is improved, and the specific improvements include: The ninth layer in the backbone structure of the original YOLOv8 model is connected with the channel attention module, and the channel attention module is connected with the eleventh layer and the twenty-first layer of the neck structure, respectively; The specific calculation steps of the channel attention module include: Wherein, E is the output feature of the previous layer of the channel attention module as the input feature map, Y1, Y2, Y3, Y4 are the feature maps generated in the calculation process; Y5 represents the output feature map of the channel attention module; f represents the operator symbol, the superscript represents the convolution kernel size, the subscript conv represents the standard convolution, Globolpooling represents the global pooling, Sigmoid represents the first activation function, ReLU represents the second activation function, FC represents the full connection, represents element-wise addition.

[0009] Optionally, the improvement based on the original YOLOv8 model further includes: In the neck network, the C2f module in the original YOLOv8 model is replaced by a spatial feature extraction module; The working principle of the spatial feature extraction module includes: The Bottlenk module in the C2f module is replaced by a spatial feature extraction module; The specific working principle of the spatial feature extraction module is: The feature input into the spatial feature extraction module is determined as an input feature tensor; The input feature tensor is convolved with a convolution kernel of 1x1 to obtain a convolution feature tensor; The convolution feature tensor is processed by multiple branches to obtain a first branch feature tensor, a second branch feature tensor and a third branch feature tensor; The first branch feature tensor is convolved with a convolution kernel of 5x5 to obtain a first convolution feature tensor, the second convolution feature tensor is convolved with a convolution kernel of 9x9 to obtain a second convolution feature tensor, and the third branch feature tensor is convolved with a convolution kernel of 13x13 to obtain a third convolution feature tensor; The first convolution feature tensor, the second convolution feature tensor and the third convolution feature tensor are subjected to a feature fusion operation to obtain a fusion feature tensor; The fusion feature tensor and the input feature tensor are added element by element to obtain a first feature tensor; After the first feature tensor is normalized, it is convolved with a convolution kernel of 1x1 to obtain an output feature tensor.

[0010] Optionally, the first film plating judgment module further includes a principle process of the spectral diagram, which includes: According to the preset angle and the preset wavelength, the target mirror panel glass is irradiated to obtain a plurality of reflected lights; according to the optical path difference of the plurality of reflected lights, interference is generated to obtain light intensity information, and then the light intensity signal is converted into an electrical signal; According to the electrical signal, the optical parameters are analyzed; According to the light parameters and the preset wavelength, a spectral diagram is generated.

[0011] Optionally, the multi-layer film optical module comprises a dynamic optimization module. The dynamic optimization module is configured to perform dynamic optimization on the spectrum to obtain thickness values of the layers of film. The dynamic optimization comprises the following steps: initializing the predicted thickness values of the layers of film in the spectrum as a population; the population is composed of multiple chromosomes; calculating the predicted thickness values of the layers of film according to a fitness formula to obtain a highest fitness value; taking the chromosome corresponding to the highest fitness value as an optimization target, and iteratively processing the optimization target through quantum evolution; the quantum evolution operation comprises dynamic quantum crossover and dynamic quantum mutation; if the iterative algorithm reaches a maximum number of iterations or meets a convergence error, outputting the optimal chromosome and the thickness values of the layers of film corresponding to the optimal chromosome.

[0012] Optionally, the step of calculating the predicted thickness values of the layers of film according to a fitness formula to obtain a highest fitness value comprises the following steps: calculating the fitness values of the chromosomes in the population; sorting the fitness values through a sorting algorithm to obtain a fitness sequence; extracting the highest value in the fitness sequence as the highest fitness value; and taking the chromosome corresponding to the optimal fitness value as an optimization target for iteration; the fitness formula: wherein, SY represents the fitness value, n represents the total number of wavelength points, i = 1, 2, …, n, the wavelength is divided into n equal parts, represents the actual amplitude ratio of the i-th wavelength point, represents the theoretical amplitude ratio of the i-th wavelength point, represents the actual phase difference of the i-th wavelength point, represents the theoretical phase difference of the i-th wavelength point.

[0013] Optionally, the second film coating judgment module is further configured to calculate the thickness error of each layer of film by calculating the thickness value of each layer of film and the preset thickness value of each layer of film. The thickness error formula comprises: wherein, represents the thickness error of the j-th layer of film, represents the thickness value of the j-th layer of film, represents the preset thickness value of each layer of film.

[0014] Optionally, the optimization target is iterated by quantum evolution, comprising: The specific process of the quantum is: by obtaining the fitness value of any two chromosomes, if the fitness value exceeds the fitness threshold, dynamic quantum crossover is performed to generate a new chromosome.

[0015] Optionally, the second coating judging module is further configured to compare the thickness error of each layer of film with a preset error thickness value of each layer of film to obtain a coating thickness state, comprising: If the thickness error of each layer of film is less than the preset error thickness value of each layer of film, the coating thickness state is normal, the target mirror panel glass is marked normally and sent to the terminal.

[0016] If the thickness error of each layer of film is greater than or equal to the preset error thickness value of each layer of film, the coating thickness state is abnormal, the target mirror panel glass is marked abnormally and sent to the terminal.

[0017] The beneficial effects of the present application are: The present application provides a coating quality detection system for intelligent mirror panel glass, which can enhance the accuracy of defect recognition and improve the recognition speed by using an image recognition model to detect intelligent mirror panel glass. Then, the multi-layer film structure of the coating is processed to generate a corresponding spectrum, and the specific thickness value of each layer of film is calculated by combining the multi-layer film optical model, thereby avoiding uneven thickness of each layer of film, and finally improving the coating quality detection efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0018] The present application will be further described below with reference to the accompanying drawings.

[0019] Figure 1 A flowchart of a coating quality detection system for intelligent mirror panel glass provided by the present application embodiment is provided. Figure 2 A network structure diagram of a YOLOv8 model provided by the present application embodiment is provided. Figure 3 A network structure diagram of an image recognition model provided by the present application embodiment is provided. Figure 4 A network structure diagram of a multi-scale enhancement unit provided by the present application embodiment is provided. DETAILED DESCRIPTION

[0020] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments of the present application. The term "and / or" in this document is only used to describe the association relationship of associated objects, and can represent three relationships, for example, A and B can represent three cases of A alone, A and B together, and B alone. In addition, the description of "first", "second" and the like in the present application is only for the purpose of description, and cannot be understood as indicating or implying the relative importance of the indicated technical features or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first", "second" can explicitly or implicitly include at least one of the features. In addition, the technical solutions of each embodiment can be combined with each other, but it must be based on the fact that a person skilled in the art can realize it. When the combination of technical solutions appears contradictory or unachievable, it should be considered that the combination of technical solutions does not exist, and is not within the protection scope of the present application.

[0021] Based on the embodiments in the present application, all other embodiments obtained by a person skilled in the art without creative labor are within the scope of protection of the present application.

[0022] The embodiments of the present application provide a coating quality detection system for intelligent mirror panel glass. Referring to Figure 1 , Figure 1 The flow chart of the coating quality detection system for intelligent mirror panel glass provided by the embodiments of the present application. The system includes the following modules: Image acquisition module: acquiring the coating image of the target mirror panel glass through the camera device; Image recognition module: obtaining the coating result by substituting the coating image into the image recognition model; First coating judgment module: judging the coating result, if the coating result is unqualified, marking the target mirror panel glass as abnormal and sending it to the terminal; if the coating result is qualified, emitting a polarized light through the spectrum device, and irradiating the target mirror panel glass at a preset angle to obtain a spectrum graph; Multi-layer film optical module: substituting the spectrum graph into the multi-layer film optical model to obtain the thickness value of each layer of film in the coating; the coating is composed of multiple layers of film; Second coating judgment module: calculating the thickness error of each layer of film by comparing the thickness value of each layer of film with the preset thickness value of each layer of film, comparing the thickness error of each layer of film with the preset thickness error value of each layer of film to obtain the coating thickness state, and sending the coating thickness state to the terminal.

[0023] The film quality detection system of the intelligent mirror panel glass provided by the embodiment of the application can detect the intelligent mirror panel glass through an image recognition model, improve the defect recognition capability and recognition efficiency, process the multilayer film of the coated film to obtain a spectrum, obtain the thickness value of each layer of film according to a multilayer film optical model, and further solve the uneven thickness between each layer of film, thereby improving the detection efficiency of the coated film quality.

[0024] Specifically, the camera equipment (industrial area array camera, etc.) has unqualified coating results, indicating that cracks, scratches, etc. appear on the coating surface; the coating result is qualified, indicating that no texture abnormalities, scratches, etc. appear on the coating surface, and vice versa; the preset angle and the preset error thickness value of each layer of film are obtained according to the working experience of the staff; and the spectrum equipment (spectrum ellipsometer).

[0025] In an implementation manner, Figure 2 , Figure 2 The network structure diagram of the original YOLOv8 model provided by the embodiment of the application is as follows, Figure 3 , Figure 3 The network structure diagram of the image recognition model provided by the embodiment of the application is as follows, and the image recognition model is improved based on the original YOLOv8 model. The specific improvements include: In the backbone structure, the fourth layer, the sixth layer and the eighth layer of the original YOLOv8 model are replaced by a multi-scale enhancement module; The multi-scale enhancement module is obtained by sequentially connecting a plurality of multi-scale enhancement units; Figure 4 , Figure 4 The network structure diagram of the multi-scale enhancement unit provided by the embodiment of the application is as follows, and the specific network structure of the multi-scale enhancement unit is as follows: The feature input into the multi-scale enhancement unit is determined as an input feature, the input feature is substituted into a DConv layer to obtain a first deep convolution feature; the first deep convolution feature is substituted into a full connection layer to obtain a first full connection feature, the full connection feature is substituted into a cascaded group attention module to obtain an attention feature, and the attention feature is substituted into a DConv layer to obtain a second deep convolution feature; and the second deep convolution feature is substituted into a full connection layer to obtain a second full connection feature; The input feature and the second full connection feature are fused to obtain an output feature.

[0026] In one implementation, a multi-scale enhancement module is introduced in the key layer of the backbone structure. Through the sequential linkage of multiple units, different sizes of coating defect features can be fully captured, and the problem of missing small and irregular defects in the original model is solved. The design of the double DConv layer in the multi-scale enhancement unit strengthens the deep feature extraction capability and can accurately mine the deep feature information of subtle defects such as coating texture abnormalities, scratches, and spots. The cascaded group attention module can focus on the core defect area, suppress background interference, and improve feature recognition. The fusion of input features and second fully connected features further enriches the feature expression dimension, ensuring that the model can still accurately identify defects in complex coating scenarios.

[0027] In one implementation, the original YOLOv8 model is improved. The specific improvements include: The ninth layer in the backbone structure of the original YOLOv8 model is connected to the channel attention module, and the channel attention module is connected to the eleventh layer and the twenty-first layer of the neck structure, respectively. The specific calculation steps of the channel attention module include: Wherein, E is the output feature of the previous layer of the channel attention module as the input feature map, Y1, Y2, Y3, and Y4 are feature maps generated during the calculation; Y5 represents the output feature map of the channel attention module; f represents the operator symbol, the superscript represents the convolution kernel size, the subscript conv represents the standard convolution, Globolpooling represents the global pooling, Sigmoid represents the first activation function, ReLU represents the second activation function, and FC represents the full connection. represents element-wise addition.

[0028] In one implementation, the channel weight is generated by full connection and activation function, and the multi-stage features are fused by element-wise addition, which can highlight the key features of coating defects (such as scratches and spots), reduce background interference, and effectively improve the accuracy of defect recognition.

[0029] In one implementation, the original YOLOv8 model is improved. The specific improvements include: In the neck network, the C2f module in the original YOLOv8 model is replaced by a spatial feature extraction module. The working principle of the spatial feature extraction module includes: The Bottlenk module in the C2f module is replaced by a spatial feature extraction module. The specific working principle of the spatial feature extraction module is: The feature input into the spatial feature extraction module is determined as an input feature tensor; The input feature tensor is convolved with a convolution kernel of 1x1 to obtain a convolution feature tensor; The convolution feature tensor is processed by multiple branches to obtain a first branch feature tensor, a second branch feature tensor and a third branch feature tensor; The first branch feature tensor is convolved by a convolution kernel 5x5 to obtain a first convolution feature tensor, the second convolution feature tensor is convolved by a convolution kernel 9x9 to obtain a second convolution feature tensor, and the third branch feature tensor is convolved by a convolution kernel 13x13 to obtain a third convolution feature tensor; The first convolution feature tensor, the second convolution feature tensor and the third convolution feature tensor are subjected to a feature fusion operation to obtain a fused feature tensor; The fused feature tensor and the input feature tensor are added element by element to obtain a first feature tensor; After the first feature tensor is subjected to normalization processing, the output feature tensor is obtained by convolution of a convolution kernel 1x1.

[0030] In an implementation, the Bottlenk modules in the C2f module are all replaced by spatial feature extraction modules, which effectively improves the calculation efficiency.

[0031] In an implementation, the principle process of the spectral diagram includes: According to the preset angle and the preset wavelength, a plurality of reflected lights are irradiated on the target mirror panel glass; the optical path difference of the plurality of reflected lights is used to generate interference to obtain light intensity information, and the light intensity signal is converted into an electrical signal; According to the electrical signal, optical parameters are obtained by analysis; According to the light parameters and the preset wavelength, a spectral diagram is generated.

[0032] In an implementation, a polarized light with a preset wavelength (300nm-900nm) and a 2nm step size is scanned at an incident angle of 65°, and a plurality of reflected lights are obtained after the polarized light is reflected and interfered by the interface of a multilayer film, and are received by a detector and converted into an electrical signal. The electrical signal is calculated to obtain corresponding optical parameters (the optical parameters are specifically amplitude ratio, phase difference, etc.), the preset wavelength is taken as the abscissa, the light parameters are taken as the ordinate, a scatter diagram is formed, and the scatter diagram is determined as a spectral diagram.

[0033] In an implementation, the multilayer film optical module includes a dynamic optimization module: The dynamic optimization module is used to dynamically optimize the spectral diagram to obtain thickness values of each layer of film; The specific process of dynamic optimization includes: The predicted thickness values of each layer of film in the spectral diagram are initialized as a population; the population is composed of multiple chromosomes; The predicted thickness values of each layer of film are calculated according to the fitness formula to obtain the highest fitness value; The chromosome corresponding to the highest fitness value is taken as the optimization target, and the optimization target is iterated through quantum evolution; the quantum evolution operation includes dynamic quantum crossover and dynamic quantum mutation. If the iterative algorithm reaches the maximum number of iterations or meets the convergence error, the optimal chromosome and the corresponding thickness values of each layer film are output.

[0034] In an implementation manner, the predicted thickness values of each layer film are obtained through an inverse problem solving method of the spectrum diagram, and the specific process of the inverse problem solving method is that the spectrum diagram (containing the optical parameters of the layer film) of the to-be-detected plated film is collected, the measured spectrum is compared and fitted with the theoretical spectrum, and finally the predicted thickness values of each layer film are adjusted.

[0035] In an implementation manner, the highest fitness value is obtained by calculating the predicted thickness values of each layer film according to the fitness formula, and the highest fitness value includes: Each fitness value is obtained by calculating each chromosome in the population. The fitness values are sorted by a sorting algorithm to obtain a fitness sequence. The highest value of the fitness sequence is extracted as the highest fitness value. And the chromosome corresponding to the optimal fitness value is taken as the optimization target for iteration. The fitness formula is: Wherein, SY represents the fitness value, n represents the total number of wavelength points, i=1, 2, …, n, the wavelength is divided into n equal parts, represents the actual amplitude ratio of the i-th wavelength point, represents the theoretical amplitude ratio of the i-th wavelength point, represents the actual phase difference of the i-th wavelength point, represents the theoretical phase difference of the i-th wavelength point.

[0036] In an implementation manner, each fitness value is obtained by calculating each chromosome in the population. The fitness values are sorted by a sorting algorithm to obtain a fitness sequence; the sorting algorithm can be selection sorting, bubble sorting, etc. The highest value of the fitness sequence is extracted as the highest fitness value. And the chromosome corresponding to the optimal fitness value is taken as the optimization target for iteration. The chromosome with a high fitness value is obtained through the fitness formula. The smaller the difference between the actual amplitude ratio and the theoretical amplitude ratio, the larger the corresponding fitness value. The smaller the difference between the actual phase difference and the theoretical phase difference, the larger the corresponding fitness value. It is indicated that the thickness values of each mode layer of the plated film are more accurately detected.

[0037] In an implementation manner, the second plated film judgment module calculates the thickness error of each layer film by calculating the thickness values of each layer film and the preset thickness values of each layer film. The thickness error formula comprises: wherein, represents the thickness error of the jth layer film, represents the thickness value of the jth layer film, represents the preset thickness value of each layer film of the jth layer film.

[0038] In an implementation, the optimization target is iterated by quantum evolution, comprising: The specific process of quantum is: by obtaining the fitness value of any two chromosomes, if the fitness value exceeds the fitness threshold, dynamic quantum crossover is performed to generate new chromosomes.

[0039] In an implementation, the quantum crossover probability is calculated according to the dynamic quantum crossover strategy of the chromosome fitness value; If the quantum crossover probability exceeds the upper limit of the preset crossover threshold, it means that the crossover probability is too large, the probability of iteration stagnation is greater, and the chromosome with high fitness value will be destroyed; if the quantum crossover probability exceeds the lower limit of the preset crossover threshold, it means that the crossover probability is too small, and the speed of the iteration process is reduced. Therefore, by dynamically adjusting the quantum crossover probability, the efficiency of the iteration process is improved, and it is also helpful to promote the information mixing in the population and generate new high-quality chromosomes.

[0040] In an implementation, the second coating film judgment module compares the thickness error of each layer film with the preset thickness error value of each layer film to obtain the coating film thickness state, comprising: If the thickness error of each layer film is less than the preset thickness error value of each layer film, the coating film thickness state is normal, the target mirror panel glass is marked normally and sent to the terminal; If the thickness error of each layer film is greater than or equal to the preset thickness error value of each layer film, the coating film thickness state is abnormal, the target mirror panel glass is marked as abnormal and sent to the terminal.

[0041] In an implementation, the coating film thickness state comprises normal and abnormal; when the thickness error of each layer film is less than the preset thickness error value of each layer film, it means that the coating film thickness state is normal, and the target mirror panel glass is marked as normal and sent to the related terminal; when the thickness error of each layer film is greater than or equal to the preset thickness error value of each layer film, it means that the coating film thickness state is abnormal, and the thickness of the target mirror panel glass is not uniform, and the target mirror panel glass is marked as abnormal and sent to the related terminal.

[0042] The above has described one embodiment of the present application in detail, but the content is only the preferred embodiment of the present application, and cannot be considered as limiting the scope of the present application. Any equivalent changes and improvements made according to the scope of the present application should still belong to the scope of the claims of the present application.

Claims

1. A coating quality inspection system for intelligent mirror panel glass, characterized in that, The system includes: Image acquisition module: Acquires images of the coating on the target mirror panel glass; Image recognition module: Obtains the coating result by substituting the coating image into the image recognition model; First coating judgment module: judges the coating result. If the coating result is unqualified, the target mirror glass is marked as abnormal and sent to the terminal; if the coating result is qualified, a segment of polarized light is emitted through the spectral device and the target mirror glass is irradiated at a preset angle to obtain a spectral image. Multilayer film optical module: Substitute the spectrum into the multilayer film optical model to obtain the thickness value of each layer in the coating; the coating is composed of multiple layers; The second coating judgment module calculates the thickness error of each layer of film by comparing the thickness value of each layer with the preset thickness value of each layer of film, compares the thickness error of each layer of film with the preset thickness error value of each layer of film to obtain the coating thickness status, and sends the coating thickness status to the terminal.

2. The coating quality inspection system for intelligent mirror panel glass according to claim 1, characterized in that, The image recognition module is also used to improve the image recognition model based on the original YOLOv8 model. Specific improvements include: In the backbone structure, the fourth, sixth and eighth layers of the original YOLOv8 model are replaced with multi-scale enhancement modules; A multi-scale enhancement module is formed by sequentially connecting multiple multi-scale enhancement units. The specific network structure of the multi-scale enhancement unit is as follows: The features input to the multi-scale enhancement unit are determined as input features. The input features are substituted into the DConv layer to obtain the first deep convolutional features. The first deep convolutional features are substituted into the fully connected layer to obtain the first fully connected features. The fully connected features are substituted into the cascaded group attention module to obtain attention features. The attention features are substituted into the DConv layer to obtain the second deep convolutional features. The obtained second deep convolutional features are substituted into the fully connected layer to obtain the second fully connected features. The input features are fused with the second fully connected features to obtain the output features.

3. The coating quality inspection system for intelligent mirror panel glass according to claim 2, characterized in that, The improvements based on the original YOLOv8 model include: The ninth layer of the backbone structure of the original YOLOv8 model is connected to the channel attention module, and the channel attention module is connected to the eleventh and twenty-first layers of the neck structure, respectively. The specific calculation steps of the channel attention module include: Where E is the input feature map, which is the output feature of the layer above the channel attention module; Y1, Y2, Y3, and Y4 are the feature maps generated during the calculation process; Y5 represents the output feature map of the channel attention module; f represents the operator, the superscript indicates the kernel size, the subscript conv indicates standard convolution, Globolpooling indicates global pooling, Sigmoid indicates the first activation function, ReLU indicates the second activation function, and FC indicates fully connected. This indicates element-wise addition.

4. The coating quality inspection system for intelligent mirror panel glass according to claim 2, characterized in that, The improvements based on the original YOLOv8 model include: In the neck network, the C2f module in the original YOLOv8 model is replaced with a spatial feature extraction module; The working principle of the spatial feature extraction module includes: Replace all Bottlenk modules in the C2f module with spatial feature extraction modules; The specific working principle of the spatial feature extraction module is as follows: The features input to the spatial feature extraction module are defined as the input feature tensor; The input feature tensor is convolved with a 1×1 kernel to obtain the convolutional feature tensor; The convolutional feature tensor is processed into a multi-branch form to obtain a first branch feature tensor, a second branch feature tensor, and a third branch feature tensor. The first branch feature tensor is convolved with a 5×5 kernel to obtain the first convolutional feature tensor, the second convolutional feature tensor is convolved with a 9×9 kernel to obtain the second convolutional feature tensor, and the third branch feature tensor is convolved with a 13×13 kernel to obtain the third convolutional feature tensor. A feature fusion operation is performed on the first convolutional feature tensor, the second convolutional feature tensor, and the third convolutional feature tensor to obtain a fused feature tensor; The first feature tensor is obtained by adding the fused feature tensor and the input feature tensor element by element. After normalizing the first feature tensor, a 1×1 convolution kernel is applied to obtain the output feature tensor.

5. The coating quality inspection system for intelligent mirror panel glass according to claim 1, characterized in that, The first coating determination module also includes the following principle process for the spectral graph: Multiple reflected lights are obtained by irradiating the target mirror glass at a preset angle and wavelength; the light intensity information is obtained by generating interference based on the optical path difference of the multiple reflected lights; and the light intensity signal is then converted into an electrical signal. Optical parameters are obtained by analyzing the electrical signal. A spectrum is generated based on the light parameters and the preset wavelength.

6. The coating quality inspection system for intelligent mirror panel glass according to claim 1, characterized in that, The multilayer optical module includes: a dynamic optimization module. The dynamic optimization module is used to dynamically optimize the spectrum to obtain the thickness values ​​of each film layer; The specific process of dynamic optimization includes: The predicted thickness values ​​of each membrane layer in the spectrum are used as the initial population; the population consists of multiple chromosomes. The highest fitness value is obtained by calculating the predicted thickness of each membrane layer according to the fitness formula. The chromosome corresponding to the highest fitness value is used as the optimization target, and the optimization target is iterated through quantum evolution; the quantum evolution operation includes: dynamic quantum crossover and dynamic quantum mutation. If the iterative algorithm reaches the maximum number of iterations or meets the convergence error, it outputs the optimal chromosome and the corresponding thickness values ​​of each membrane.

7. The coating quality inspection system for intelligent mirror panel glass according to claim 6, characterized in that, The highest fitness value obtained by calculating the predicted thickness of each membrane layer according to the fitness formula includes: The fitness values ​​of each chromosome in the population are obtained by calculating the fitness values ​​of each chromosome. The fitness values ​​are sorted using a sorting algorithm to obtain a fitness sequence. Extract the highest value from the fitness sequence as the highest fitness value; The chromosome corresponding to the optimal fitness value is then used as the optimization target for iterative optimization. Fitness formula: Where SY represents the fitness value, n represents the total number of wavelength points, i = 1, 2, ..., n, and the wavelengths are divided into n equal parts. This represents the actual amplitude ratio at the i-th wavelength point. This represents the theoretical amplitude ratio at the i-th wavelength point. This represents the actual phase difference at the i-th wavelength point. This represents the theoretical phase difference at the i-th wavelength point.

8. The coating quality inspection system for intelligent mirror panel glass according to claim 1, characterized in that, The second coating determination module is further used to calculate the thickness error of each film layer based on the thickness value of each film layer and the preset thickness value of each film layer. The thickness error of each film layer includes: The thickness error formula includes: ,in, This represents the thickness error of the j-th film. This represents the thickness value of the j-th film layer. This represents the preset thickness values ​​of each layer of the j-th layer.

9. The coating quality inspection system for intelligent mirror panel glass according to claim 6, characterized in that, The optimization objective is iterated through quantum evolution, including: The specific process of quantum crossover is as follows: by obtaining the fitness values ​​of any two chromosomes, if the fitness values ​​exceed the fitness threshold, dynamic quantum crossover is performed to generate new chromosomes.

10. The coating quality inspection system for intelligent mirror panel glass according to claim 1, characterized in that, The second coating determination module is further used to compare the thickness error of each layer of film with the preset thickness error value of each layer of film to obtain the coating thickness status, including: If the thickness error of each film layer is less than the preset thickness error value of each film layer, the coating thickness status is normal, the target mirror panel glass is marked normally and sent to the terminal. If the thickness error of each film layer is greater than or equal to the preset thickness error value of each film layer, the coating thickness status is abnormal, the target mirror glass is marked as abnormal and sent to the terminal.

Images