Rare earth reflective heat insulation metal plate and roll coating quality detection method thereof

Abstract

The application relates to the technical field of heat-insulating metal plates, and particularly discloses a rare earth reflective heat-insulating metal plate and a roll coating quality detection method thereof. The rare earth reflective heat-insulating metal plate is a metal plate with high reflective capacity for sunlight and infrared rays and heat-insulating function. The surface coating of the metal plate is composed of a multilayer overlapping structure of a rare earth composite reflective material coating and a rare earth composite hollow microsphere heat-insulating material coating. In the preparation process of the rare earth reflective heat-insulating metal plate, machine vision is used to construct a roll coating quality detection scheme. The scheme can automatically identify defects on the surface of the coating, such as bubbles, cracks and unevenness, and realize mass roll coating quality control, so as to guarantee the heat-insulating performance and durability of the final product.

Description

Technical Field

[0001] This application relates to the field of heat-insulating metal plate technology, and more specifically, to a rare earth reflective heat-insulating metal plate and a method for testing the quality of its roller coating. Background Technology

[0002] With the development of technology and the increasing awareness of environmental protection, the requirements for the energy-saving performance of materials in the construction and industrial sectors are becoming increasingly stringent. Especially in building envelopes, heat loss in components such as walls and roofs is a significant factor affecting building energy consumption. To effectively reduce heat exchange between the interior and exterior environments of buildings and improve building energy efficiency, the development of new thermal insulation materials has become a research hotspot.

[0003] Traditional thermal insulation materials primarily rely on their physical properties (such as low thermal conductivity) to achieve insulation effects, but these properties suffer from drawbacks such as heavy weight and inconvenient installation. In recent years, with the development of nanotechnology and new materials science, technologies that utilize the optical properties of specific substances or materials to reflect solar radiation and achieve efficient thermal insulation have gained increasing attention. Among these, rare earth elements, due to their unique spectral characteristics and chemical stability, show great potential for application in reflective thermal insulation coatings.

[0004] Rare earth reflective heat insulation coating is a functional coating with rare earth compounds as its main components. It can effectively reflect the near-infrared part of sunlight while maintaining good visible light transmittance, thereby significantly reducing the surface temperature of objects without affecting lighting and achieving the purpose of energy saving and consumption reduction.

[0005] However, in the actual application process, the quality of the coating directly affects the thermal insulation performance of the final product. To ensure coating quality, precise testing is usually required. Currently, the main method for testing coating quality relies on manual sampling inspection. This method is not only inefficient but also easily influenced by the subjective judgment of the inspectors, making it difficult to guarantee the objectivity and accuracy of the test results.

[0006] Therefore, an optimized rare-earth reflective heat-insulating metal plate and its preparation method are desired. Summary of the Invention

[0007] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide a novel rare-earth reflective heat-insulating metal plate and its roll coating quality inspection method. The rare-earth reflective heat-insulating metal plate is a metal plate with high reflectivity to sunlight and infrared rays and heat insulation function. Its surface coating consists of a multi-layered overlapping structure of a rare-earth composite reflective material coating and a rare-earth composite hollow microsphere heat-insulating material coating. During the preparation process of the rare-earth reflective heat-insulating metal plate, machine vision is used to construct a roll coating quality inspection scheme. This scheme can automatically identify defects on the coating surface, such as bubbles, cracks, and unevenness, achieving large-scale roll coating quality control, thereby ensuring the heat insulation performance and durability of the final product.

[0008] Accordingly, according to one aspect of this application, a method for preparing a rare-earth reflective heat-insulating metal plate is provided, comprising: providing a rare-earth reflective heat-insulating coating and a metal plate; performing surface treatment on the metal plate to obtain a surface-treated metal plate; applying the rare-earth reflective heat-insulating coating to the surface-treated metal plate using a roll coating process to obtain a rare-earth reflective heat-insulating metal plate; and performing roll coating quality inspection on the rare-earth reflective heat-insulating metal plate to obtain a roll coating quality inspection result; wherein, performing roll coating quality inspection on the rare-earth reflective heat-insulating metal plate to obtain a roll coating quality inspection result includes: acquiring an image of the coating surface state captured by an industrial camera; and performing roll coating quality inspection on the rare-earth reflective heat-insulating metal plate to obtain a roll coating quality inspection result. The surface condition image of the coating is subjected to texture feature extraction based on grid granularity to obtain a set of texture feature vectors of grid granularity of the coating surface; the set of texture feature vectors of grid granularity of the coating surface is subjected to feature saliency modulation dynamic aggregation to obtain a global saliency aggregation representation vector of the texture grid granularity of the coating surface; based on the feature difference between each texture feature vector of the coating surface in the set of texture feature vectors of grid granularity of the coating surface and the global saliency aggregation representation vector of the texture grid granularity of the coating surface, the roll coating quality inspection result is generated, wherein the roll coating quality inspection result is used to indicate whether the coating surface is dense.

[0009] In the above-mentioned method for preparing rare earth reflective heat-insulating metal plates, the process of extracting texture features based on grid granularity from the surface state image of the coating to obtain a set of texture feature vectors of the coating surface grid granularity includes: dividing the surface state image of the coating into grids to obtain a set of surface state grid granularity image blocks; and inputting each surface state grid granularity image block in the set of surface state grid granularity image blocks into a texture feature extractor based on a multi-layer dilated convolution structure to obtain the set of texture feature vectors of the coating surface grid granularity.

[0010] In the above-mentioned method for preparing rare-earth reflective heat-insulating metal plates, the set of feature vectors of the coating surface mesh grain texture is dynamically aggregated with feature saliency modulation to obtain a globally saliency aggregated representation vector of the coating surface texture mesh grain. This includes: determining an initial center vector for clustering the coating surface mesh grain texture features based on the feature distribution field of the set of coating surface mesh grain texture feature vectors; and performing saliency modulation dynamic aggregation on the set of coating surface mesh grain texture feature vectors based on the spatial span of each coating surface mesh grain texture feature vector in the set relative to the initial center vector of the coating surface mesh grain texture feature cluster to obtain a globally saliency aggregated representation vector of the coating surface texture mesh grain.

[0011] In the above-mentioned method for preparing rare-earth reflective heat-insulating metal plates, the initial center vector for clustering the coating surface mesh grain texture feature is determined based on the feature distribution field of the set of coating surface mesh grain texture feature vectors. This includes: calculating the static energy factor of each coating surface mesh grain texture feature vector in the set of coating surface mesh grain texture feature vectors to obtain a set of static energy factors for coating surface mesh grain texture features; and selecting the coating surface mesh grain texture feature vector corresponding to the maximum value in the set of static energy factors for coating surface mesh grain texture features as the initial center vector for clustering the coating surface mesh grain texture features.

[0012] In the above-mentioned method for preparing rare earth reflective heat-insulating metal plates, calculating the static energy factor of each coating surface mesh grain texture feature vector in the set of coating surface mesh grain texture feature vectors to obtain the set of coating surface mesh grain texture feature static energy factors includes: calculating the kurtosis of the coating surface mesh grain texture feature vectors, and inputting the kurtosis into the sigmoid activation function to obtain the coating surface mesh grain texture feature static energy factor.

[0013] In the above-mentioned method for preparing rare-earth reflective heat-insulating metal plates, based on the spatial span of each coating surface mesh grain texture feature vector in the set of coating surface mesh grain texture feature vectors relative to the initial center vector of the coating surface mesh grain texture feature clustering, the set of coating surface mesh grain texture feature vectors is subjected to saliency modulation dynamic aggregation to obtain the globally saliency aggregated representation vector of the coating surface texture mesh grain. This includes: the spatial span between each coating surface mesh grain texture feature vector in the set of coating surface mesh grain texture feature vectors and the initial center vector of the coating surface mesh grain texture feature clustering, and The set of static energy factors of the coating surface mesh grain texture features is used to calculate the dynamic aggregated energy factor of each coating surface mesh grain texture feature vector to obtain the set of dynamic aggregated energy factors of the coating surface mesh grain texture features; the set of dynamic aggregated energy factors of the coating surface mesh grain texture features is input into a gated mask unit to obtain the set of dynamic aggregated weight factors of the coating surface mesh grain texture features; based on the set of dynamic aggregated weight factors of the coating surface mesh grain texture features, the weighted sum of the set of coating surface mesh grain texture feature vectors is calculated to obtain the global salient aggregated representation vector of the coating surface texture mesh grain.

[0014] In the above-mentioned method for preparing rare earth reflective heat-insulating metal plates, the dynamic aggregation energy factor of the surface mesh grain texture feature vectors of each coating surface is calculated to obtain a set of dynamic aggregation energy factors of the surface mesh grain texture features. This includes: using the square of the number of feature vectors separated by the surface mesh grain texture feature vectors and the initial center vector of the surface mesh grain texture feature cluster as a spatial span coefficient; and calculating the weighted ratio between the product of the static energy factor of the surface mesh grain texture feature vectors and the static energy factor of the initial center vector of the surface mesh grain texture feature cluster cluster and the spatial span coefficient to obtain the dynamic aggregation energy factor of the surface mesh grain texture features.

[0015] In the above-mentioned method for preparing rare earth reflective heat-insulating metal plates, the roll coating quality inspection result is generated based on the feature differences between each coating surface mesh grain texture feature vector in the set of coating surface mesh grain texture feature vectors and the global significant aggregated representation vector of coating surface texture mesh grain. This includes: calculating the coating surface state semantic difference coefficient between each coating surface mesh grain texture feature vector in the set of coating surface mesh grain texture feature vectors and the global significant aggregated representation vector of coating surface texture mesh grain to obtain a global distribution representation vector of coating surface state composed of multiple coating surface state semantic difference coefficients; and inputting the global distribution representation vector of coating surface state into a classifier-based roll coating quality inspection module to obtain the roll coating quality inspection result.

[0016] In the above-mentioned method for preparing rare earth reflective heat-insulating metal plates, calculating the coating surface state semantic difference coefficient between each coating surface mesh grain texture feature vector in the set of coating surface mesh grain texture feature vectors and the coating surface texture mesh grain global significant aggregated representation vector to obtain a coating surface state global distribution characterization vector composed of multiple coating surface state semantic difference coefficients includes: calculating the Mahalanobis distance between each coating surface mesh grain texture feature vector in the set of coating surface mesh grain texture feature vectors and the coating surface texture mesh grain global significant aggregated representation vector as the coating surface state semantic difference coefficient to obtain the coating surface state global distribution characterization vector.

[0017] According to another aspect of this application, a rare earth reflective heat-insulating metal plate is provided, characterized in that the rare earth reflective heat-insulating metal plate is prepared by the method described above for preparing rare earth reflective heat-insulating metal plates.

[0018] Compared with existing technologies, the rare earth reflective heat-insulating metal plate and its roller coating quality inspection method described in this application are characterized by a metal plate with high reflectivity to sunlight and infrared rays and heat insulation function. The surface coating consists of a multi-layered overlapping structure of rare earth composite reflective material coating and rare earth composite hollow microsphere heat-insulating material coating. During the preparation of the rare earth reflective heat-insulating metal plate, machine vision is used to construct a roller coating quality inspection scheme. This scheme can automatically identify defects on the coating surface, such as bubbles, cracks, and unevenness, achieving large-scale roller coating quality control and thus ensuring the heat insulation performance and durability of the final product. Attached Figure Description

[0019] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0020] Figure 1 This is a flowchart illustrating the process of performing roll coating quality testing on the rare earth reflective heat-insulating metal plate according to the preparation method of the rare earth reflective heat-insulating metal plate in the embodiments of this application, in order to obtain the roll coating quality testing results.

[0021] Figure 2 This is a flowchart of step S2 in the method for preparing a rare earth reflective heat-insulating metal plate according to an embodiment of this application.

[0022] Figure 3This is a flowchart of step S3 in the method for preparing a rare earth reflective heat-insulating metal plate according to an embodiment of this application.

[0023] Figure 4 This is a flowchart of step S4 in the method for preparing a rare earth reflective heat-insulating metal plate according to an embodiment of this application. Detailed Implementation

[0024] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0025] This application proposes a method for preparing a rare earth reflective heat-insulating metal plate, the method comprising: providing a rare earth reflective heat-insulating coating and a metal plate; performing surface treatment on the metal plate to obtain a surface-treated metal plate; applying the rare earth reflective heat-insulating coating to the surface-treated metal plate by a roller coating process to obtain a rare earth reflective heat-insulating metal plate; and performing roller coating quality testing on the rare earth reflective heat-insulating metal plate to obtain roller coating quality testing results.

[0026] In the technical solution of this application, the metal plate can be made of aluminum plate, stainless steel plate or other metal materials. The rare earth reflective heat insulation coating is prepared by the following method: Step 1: Add solvent, dispersant and defoamer to the dispersion tank and disperse at low speed until uniform; Step 2: Continue to add rare earth pigments and fillers and precipitated barium sulfate to the dispersion tank and disperse at high speed until uniform to obtain a first mixture; Step 3: Grind the first mixture to obtain a first mixture powder; Step 4: Mix the first mixture powder, pure acrylic emulsion, wetting agent and film-forming aid and stir evenly to obtain a second mixture; Step 5: Add thickener to adjust the viscosity of the second mixture to obtain a third mixture; Step 6: Add bactericide to the third mixture and stir until uniform to obtain the rare earth reflective heat insulation coating.

[0027] Meanwhile, in the viscosity adjustment process of rare earth reflective heat-insulating coatings, this application further introduces intelligent industrial control technology to adjust the amount of thickener applied. First, a target state image of the mixture is generated based on the target viscosity value input by the user. Then, image processing technology based on deep learning is used to extract and enhance the image features of the target state image and the actual surface state image of the mixture to extract the surface state features of the materials. Based on the feature differences between the actual surface state of the mixture and the target state, the amount of thickener applied is intelligently recommended. In this way, the amount of thickener applied can be optimized and controlled, improving the production efficiency and quality stability of the coating.

[0028] To ensure good adhesion between the coating and the metal plate, the metal plate surface needs to be pre-treated, including but not limited to degreasing, pickling, and chemical conversion treatment (such as phosphating or chromating). This removes oil, oxides, and other impurities from the metal surface and forms a protective film, improving the durability and adhesion of subsequent coatings. Next, a high-efficiency roller coating process is used to evenly coat the pre-treated metal plate with the rare earth reflective heat-insulating coating. After coating, the coated metal plate is placed in a dedicated oven for drying and curing, allowing the solvent in the coating to evaporate rapidly and inducing physical or chemical changes in the coating to form a robust and durable coating structure.

[0029] Traditional roll coating quality inspection methods rely primarily on manual sampling, which is inefficient and susceptible to subjective judgment by inspectors, making it difficult to guarantee the objectivity and accuracy of the results. Therefore, this application further introduces computer vision technology for roll coating quality inspection. Specifically, an industrial camera is first used to acquire images of the coating surface condition. Then, deep learning-based computer vision technology is used to extract texture features from the coating surface images at the network granularity level. Using the globally salient texture features of the coating surface images as a reference, the semantic differences between the texture features of each network region and the globally salient texture features of the image are used to capture the global texture distribution characteristics of the coating surface, thereby achieving coating surface density detection. This allows for the automatic identification of defects on the coating surface, such as bubbles, cracks, and unevenness, enabling large-scale roll coating quality control and ensuring the thermal insulation performance and durability of the final product.

[0030] Figure 1 This is a flowchart illustrating the roll coating quality inspection of the rare earth reflective heat-insulating metal plate according to an embodiment of this application, to obtain the roll coating quality inspection results. Figure 1 As shown, the roll coating quality inspection of the rare earth reflective heat-insulating metal plate includes: S1, acquiring a coating surface state image captured by an industrial camera; S2, extracting texture features based on grid granularity from the coating surface state image to obtain a set of coating surface grid granularity texture feature vectors; S3, performing feature saliency modulation dynamic aggregation on the set of coating surface grid granularity texture feature vectors to obtain a global saliency aggregation representation vector of coating surface texture grid granularity; S4, generating the roll coating quality inspection result based on the feature differences between each coating surface grid granularity texture feature vector in the set of coating surface grid granularity texture feature vectors and the global saliency aggregation representation vector of coating surface texture grid granularity, wherein the roll coating quality inspection result is used to indicate whether the coating surface is dense.

[0031] In the above-described method for preparing rare-earth reflective heat-insulating metal plates, step S1 involves acquiring a surface condition image of the coating using an industrial camera. It should be understood that using an industrial camera to acquire surface condition images of the coating provides high-resolution visual information about the coating surface, enabling automated visual quality inspection, thereby improving inspection efficiency and shortening the production cycle. Furthermore, using an industrial camera to acquire surface condition images of the coating is a non-contact inspection method, avoiding potential damage to the coating or material from physical contact, while also reducing interference with the production environment.

[0032] Furthermore, in this application's technical solution, a high-resolution industrial camera is used. A high-resolution industrial camera can capture minute details on the coating surface, such as bubbles, cracks, and unevenness, which is crucial for subsequent defect detection. Secondly, the stability and reliability of the industrial-grade camera are also important considerations. Industrial-grade cameras can operate stably for extended periods in various industrial environments, unaffected by environmental factors such as temperature and humidity, ensuring the consistency and reliability of image acquisition. Simultaneously, a uniform light source is set up, which reduces shadows and reflections, ensuring image clarity and consistency. Multi-angle light source settings can illuminate the coating surface from different directions, capturing surface features from more perspectives. For example, using a ring light source can provide uniform illumination, while using a directional light source can highlight surface texture and defects. In addition, the color temperature and intensity of the light source also need to be adjusted according to the specific coating color and surface characteristics to obtain the best image effect. Fixing the industrial camera in a suitable position, ensuring that the distance and angle between the camera and the coating surface are consistent, can reduce image distortion and loss of detail. Optionally, in other examples of this application, multiple cameras can be set up to photograph the same coating surface from different angles to obtain more comprehensive information. For example, installing multiple cameras at different locations on the production line can enable comprehensive monitoring of the coated surface.

[0033] In this embodiment, step S1 further includes image acquisition. First, during image acquisition, it is essential to ensure that the metal plate has been coated and dried. After coating, the coating surface needs to be cleaned to remove dust and impurities to avoid interfering with image acquisition. Next, the coated metal plate is placed on a fixed platform, ensuring its position is stable and flat. This step is crucial because any movement or tilting will cause image distortion. Before starting the industrial camera, camera parameters, such as exposure time and gain, need to be adjusted to obtain optimal image quality. Excessive exposure time will result in overexposure, while insufficient exposure time will result in underexposure. Excessive gain will introduce noise, while insufficient gain will result in unclear image details. Therefore, the optimal parameter settings need to be found through experimentation based on the actual coating surface characteristics and lighting conditions. The camera shutter is then triggered to acquire an image of the coating surface. In practice, multiple images can be acquired to ensure data reliability and integrity. For example, multiple images can be acquired under different lighting conditions, or the same coating surface can be photographed from different positions and angles to obtain more information. The acquired images need to be promptly saved to a designated storage device for subsequent processing and analysis.

[0034] In an optional embodiment of this application, step S1 further includes image preprocessing, which can improve image quality and detection accuracy. First, geometric correction is performed to eliminate image distortion caused by camera lens distortion and platform unevenness. Geometric correction can be achieved using a calibration plate and calibration algorithms. The calibration plate typically has markers at known locations, through which the camera's intrinsic and extrinsic parameters can be calculated, thereby correcting the geometric distortion of the image. Second, illumination correction is performed to eliminate brightness inconsistencies caused by uneven light sources. Illumination correction can be achieved through image histogram equalization or grayscale transformation. These methods can adjust the brightness and contrast of the image, making the details of the coating surface more apparent. Furthermore, multi-scale illumination correction methods can be used to further improve image quality by performing illumination correction at different scales. In an optional embodiment of this application, the image preprocessing also includes image enhancement. By adjusting the image contrast, the details of the coating surface can be made more apparent. Contrast enhancement can be achieved through linear stretching or nonlinear transformation, for example, using histogram equalization or gamma correction. Additionally, filtering algorithms can be used to remove noise from the image, improving image quality. Common filtering algorithms include median filtering, Gaussian filtering, and bilateral filtering. These algorithms can effectively remove random noise from images while preserving the details of the coating surface. Furthermore, the image preprocessing also includes image quality assessment. First, the image resolution is assessed to ensure it is high enough to clearly display the details of the coating surface. Insufficient resolution will prevent the capture of minute defects, affecting subsequent detection accuracy. Second, the image sharpness is assessed to ensure there is no blurring or defocusing. Insufficient sharpness will lead to loss of detail, affecting the accuracy of the detection results. Finally, the image illumination uniformity is assessed to ensure there is no significant unevenness in brightness. Uneven illumination will reduce image contrast, affecting defect identification. If the image quality is unsatisfactory, troubleshooting is required. First, check if the camera, light source, and platform are functioning properly. For example, check if the camera lens is dusty or dirty, check if the brightness and color temperature of the light source meet requirements, and check if the platform is flat and stable. Second, adjust camera parameters, such as exposure time and gain, according to the actual situation to optimize image quality. For example, if the image is too dark, the exposure time can be increased appropriately; if the image is overexposed, the exposure time can be decreased appropriately. Through these adjustments, the image quality can be gradually improved, ensuring the smooth progress of subsequent processing and analysis.

[0035] In an optional embodiment of this application, step S1 further includes data transmission and storage to ensure the security and integrity of image data. First, the acquired images are transmitted to a computer or server in real time for subsequent processing and analysis. Data transmission can be achieved via wired or wireless means. Wired transmission offers higher stability and reliability, while wireless transmission is more flexible and convenient. Regardless of the method used, the real-time nature and security of data transmission must be ensured to prevent data loss or corruption. Data backup is another crucial step. Regularly backing up image data ensures its security and integrity. Backups can be implemented using external storage devices or cloud storage services. External storage devices offer higher storage capacity and access speeds, while cloud storage services are more flexible and convenient. Regardless of the method used, the integrity and recoverability of the backup data must be ensured to prevent data loss due to equipment failure or human error. File naming and database storage are also important aspects of data management. Naming image files according to certain rules facilitates subsequent searching and management. For example, date, time, and location information can be used as part of the filename to quickly locate a specific image file. Storing image data in a database enables centralized management and querying of image data. The database can record information such as the acquisition time, location, and parameter settings of each image, which facilitates subsequent quality inspection and data analysis.

[0036] In the above-mentioned method for preparing rare-earth reflective heat-insulating metal plates, step S2 involves extracting texture features based on grid granularity from the surface state image of the coating to obtain a set of grid granularity texture feature vectors for the coating surface. Wherein, Figure 2 This is a flowchart of step S2 in the preparation method of the rare-earth reflective heat-insulating metal plate according to an embodiment of this application. Figure 2 As shown, step S2 includes: S21, performing mesh segmentation on the coating surface state image to obtain a set of coating surface state mesh granularity image blocks; S22, inputting each coating surface state mesh granularity image block in the set of coating surface state mesh granularity image blocks into a texture feature extractor based on a multi-layer dilated convolution structure to obtain a set of coating surface mesh granularity texture feature vectors.

[0037] Specifically, in step S21, the coating surface state image is divided into grids to obtain a set of coating surface state grid-granular image blocks. That is, to fully capture the detailed information of the coating surface, this application further divides the coating surface state image into grids to refine the granularity of feature extraction, facilitating detailed analysis of coating features within each grid region, thereby more accurately evaluating the uniformity and density of the coating. Furthermore, compared to directly processing the entire image, dividing the image into multiple small blocks reduces the amount of data processed per operation, thereby reducing computational complexity and time required, and improving the real-time performance of detection. In addition, grid division helps capture local details of the coating surface, such as tiny bubbles, cracks, and non-uniformities, which may be overlooked in large-size images. By analyzing the texture features within each grid region, a more comprehensive understanding of the overall state of the coating surface can be achieved.

[0038] In an optional embodiment of this application, the grid segmentation method is uniform segmentation, which divides the image evenly into several small blocks of the same size. A 1024x1024 pixel image can be segmented into 16x16 small blocks of 64x64 pixels each. The advantage of uniform segmentation is its simplicity and ease of implementation; each small block has a consistent size and shape, facilitating subsequent processing and analysis. The specific operation steps are as follows: First, determine the grid size. Select an appropriate grid size based on the characteristics of the coating surface and the detection requirements. For example, a 64x64 pixel grid size can be selected. Then, calculate the number of grids. Calculate the number of grids to be segmented based on the total image size and the grid size. For example, for a 1024x1024 pixel image, segmenting it into a 64x64 pixel grid yields 16x16 grids. Finally, segment the image. Use image processing software or an image processing library in a programming language to segment the image, uniformly dividing the large image into multiple small blocks.

[0039] In an optional embodiment of this application, the mesh segmentation method is adaptive segmentation. Adaptive segmentation dynamically divides the image based on local features, allowing for smaller meshes in areas with complex textures and larger meshes in areas with simple textures. The advantage of adaptive segmentation is its ability to more flexibly handle regions of varying complexity, improving the targeting and accuracy of feature extraction. The specific operation steps are as follows: First, feature extraction is performed. Preliminary feature extraction is conducted on the image; for example, edge detection, texture analysis, and other methods can be used to extract local features. Then, feature analysis is performed. Based on the extracted features, the complexity of each region is analyzed. For example, the edge density or texture complexity of each region can be calculated. Finally, the mesh is dynamically divided. Based on the results of the feature analysis, the mesh is dynamically divided. For regions with high complexity, smaller meshes are used; for regions with low complexity, larger meshes are used.

[0040] In an optional embodiment of this application, the mesh segmentation method is overlapping segmentation, which means that adjacent small blocks have a certain overlap area during segmentation. For example, a 1024x1024 pixel image can be segmented into 16x16 64x64 pixel blocks, but each block overlaps by 16 pixels. The advantage of overlapping segmentation is that it avoids feature truncation at the boundaries, ensuring that each feature can be fully analyzed in multiple blocks. The specific operation steps are as follows: First, determine the mesh size and overlap ratio. Based on the characteristics of the coating surface and the detection requirements, select an appropriate mesh size and overlap ratio. For example, a 64x64 pixel mesh size and an overlap ratio of 25% can be selected. Then, calculate the number of meshes. Based on the total image size, mesh size, and overlap ratio, calculate the number of meshes to be segmented. For example, for a 1024x1024 pixel image, segmenting it into a 64x64 pixel mesh with an overlap ratio of 25% yields 21x21 meshes. Finally, segment the image. Image segmentation can be performed using image processing software or image processing libraries in programming languages ​​to divide a large image into multiple smaller blocks with overlapping areas.

[0041] Specifically, in step S22, each coating surface state grid granularity image block in the set of coating surface state grid granularity image blocks is input into a texture feature extractor based on a multi-layer dilated convolution structure to obtain a set of coating surface grid granularity texture feature vectors. It should be understood that, considering the multi-scale characteristics of coating surface state features, this application further employs a texture feature extractor based on a multi-layer dilated convolution structure to extract features from each coating surface state grid granularity image block in the set of coating surface state grid granularity image blocks. Those skilled in the art should know that dilated convolution, by inserting holes in the convolution kernel, can expand the receptive field without reducing spatial resolution, helping to retain more detailed information. Furthermore, the multi-layer dilated convolution structure, through layer-by-layer in-depth learning and abstraction of texture features, forms a hierarchical feature representation, which not only captures local detailed information in the coating surface state grid granularity image blocks but also effectively integrates contextual information over a wider range, contributing to a more comprehensive understanding of the texture state of the coating surface.

[0042] Specifically, texture feature extraction is the process of extracting information describing surface texture from an image. Texture features can reflect the local structure and patterns of an image, which is particularly important for defect detection on coated surfaces. Common texture feature extraction methods include Gray-Level Co-occurrence Matrix (GLCM), Local Binary Pattern (LBP), and deep learning-based methods. In this application, a texture feature extractor based on a multi-layer dilated convolutional structure is used. This method combines the powerful representational capabilities of Convolutional Neural Networks (CNNs) with the expanded receptive field characteristics of dilated convolutions, enabling it to more effectively capture multi-scale texture features of coated surfaces. The core of texture feature extraction lies in extracting feature vectors that describe the surface texture from the image. These feature vectors can be used for subsequent quality detection and classification tasks. For example, the Gray-Level Co-occurrence Matrix (GLCM) extracts statistical features describing the texture by calculating the gray-level relationships between pixel pairs in the image; the Local Binary Pattern (LBP) extracts local pattern features describing the texture by comparing each pixel with its neighboring pixels. However, these traditional methods have limitations when dealing with complex textures and multi-scale features, while deep learning-based methods, especially multi-layer dilated convolutional structures, can overcome these limitations.

[0043] Multi-layer dilated convolutional structures are a special type of convolutional neural network architecture that expands the receptive field without reducing spatial resolution by inserting holes (i.e., skipping certain pixels) into the convolutional kernel. This characteristic allows multi-layer dilated convolutional structures to capture a wider range of contextual information while preserving more detail. Specifically, multi-layer dilated convolutional structures have the following characteristics: dilated convolution expands the receptive field of each kernel by inserting holes, thus capturing a wider range of contextual information. This is particularly advantageous for capturing multi-scale texture features of coated surfaces. For example, a 3x3 convolutional kernel with a dilation rate of 1 has a receptive field of 3x3; with a dilation rate of 2, the receptive field is 7x7. Through multi-layer dilated convolution, the receptive field can be progressively expanded to capture texture information at different scales. Unlike ordinary convolution operations, dilated convolution does not reduce the spatial resolution of the image, thus preserving more detail. This is crucial for detecting minute defects (such as bubbles, cracks, etc.) on coated surfaces. Ordinary convolution reduces the spatial resolution of the image after each convolution operation, while dilated convolution maintains the spatial resolution by inserting holes, thus preserving detailed information. Multi-layered dilated convolutional structures learn and abstract texture features layer by layer, forming hierarchical feature representations. This hierarchical feature representation not only captures local details but also effectively integrates contextual information over a wider range, contributing to a more comprehensive understanding of the texture state of the coated surface. For example, shallow convolutional layers can capture local details, while deep convolutional layers can capture broader contextual information, forming multi-layered feature representations.

[0044] In an optional embodiment of this application, the multi-layer dilated convolutional structure consists of multiple convolutional layers, each using a different size dilated convolutional kernel. In this application, a 3x3 convolutional kernel can be used with dilation rates of 1, 2, 4, etc., progressively expanding the receptive field. Each convolutional layer is followed by an activation function (such as ReLU) and a pooling layer (such as max pooling) to enhance the non-linear representation of features and reduce the size of the feature map. In the multi-layer dilated convolutional structure, the convolutional operation of each layer extracts texture features at different scales. The first convolutional layer uses a 3x3 convolutional kernel with a dilation rate of 1, which can capture local detail features; the second convolutional layer uses a 3x3 convolutional kernel with a dilation rate of 2, which can capture a slightly larger range of contextual information; the third convolutional layer uses a 3x3 convolutional kernel with a dilation rate of 4, which can capture an even larger range of contextual information. Through multi-layer convolutional operations, high-level texture features are gradually abstracted. In each convolutional layer, the dilated convolutional operation captures texture features at different scales. For example, a smaller void ratio (e.g., 1) can capture local details, while a larger void ratio (e.g., 4) can capture a wider range of contextual information. Through multiple convolutional operations, high-level texture features are gradually abstracted. These features can more comprehensively describe the texture state of the coated surface.

[0045] In a multi-layered dilated convolutional architecture, each convolutional operation generates a new feature map. These feature maps contain rich texture information, reflecting different scale features of the coated surface. For example, the feature map generated by the first convolutional layer may reflect local details, the feature map generated by the second convolutional layer may reflect contextual information over a slightly larger area, and the feature map generated by the third convolutional layer may reflect even more contextual information. Through multiple convolutional operations, high-level texture features are progressively abstracted, forming a multi-layered feature representation. After the last convolutional operation, the resulting feature map is flattened into a one-dimensional vector. Each image patch corresponds to a feature vector, and these feature vectors contain rich texture information.

[0046] In the above-mentioned method for preparing rare-earth reflective heat-insulating metal plates, step S3 involves dynamically aggregating the set of feature vectors of the coating surface mesh grain size texture to obtain a globally significant aggregated representation vector of the coating surface texture mesh grain size. It should be understood that, since the quality assessment of the coating surface needs to consider its overall state, in order to reveal the global texture distribution characteristics of the coating surface, this application further performs a global aggregation analysis on the set of feature vectors of the coating surface mesh grain size texture to mine the globally significant texture features of the coating surface. This serves as a reference benchmark to assess the texture feature difference distribution of each local mesh region, thereby effectively revealing the uniformity and density of the coating. Accordingly, this application proposes a dynamic aggregation method based on feature distribution field to process the set of feature vectors of the coating surface mesh grain size texture. This method dynamically adjusts the feature aggregation weights based on the correlation interaction patterns between the various coating surface mesh grain size texture feature vectors, thereby effectively mining the globally significant texture feature patterns of the coating surface.

[0047] Figure 3 This is a flowchart of step S3 in the preparation method of the rare-earth reflective heat-insulating metal plate according to an embodiment of this application. Figure 3 As shown, step S3 includes: S31, determining the initial center vector for clustering the coating surface mesh grain texture features based on the feature distribution field of the set of coating surface mesh grain texture feature vectors; S32, performing saliency modulation dynamic aggregation on the set of coating surface mesh grain texture feature vectors based on the spatial span of each coating surface mesh grain texture feature vector in the set of coating surface mesh grain texture feature vectors relative to the initial center vector for clustering the coating surface mesh grain texture features to obtain the global saliency aggregated representation vector of the coating surface texture mesh grain.

[0048] Specifically, step S31 includes: calculating the static energy factor of each coating surface mesh grain texture feature vector in the set of coating surface mesh grain texture feature vectors to obtain a set of coating surface mesh grain texture feature static energy factors; selecting the coating surface mesh grain texture feature vector corresponding to the maximum value in the set of coating surface mesh grain texture feature static energy factors as the initial center vector for the clustering of coating surface mesh grain texture features. The calculation of the static energy factor of each coating surface mesh grain texture feature vector in the set of coating surface mesh grain texture feature vectors to obtain the set of coating surface mesh grain texture feature static energy factors includes: calculating the kurtosis of the coating surface mesh grain texture feature vectors, and inputting the kurtosis into a sigmoid activation function to obtain the coating surface mesh grain texture feature static energy factors.

[0049] That is, firstly, the static energy factor is calculated based on the kurtosis of the texture feature vector of each coating surface mesh to evaluate its stability and importance in the feature space. Then, the texture feature vector of the coating surface mesh corresponding to the largest static energy factor is selected as the initial center vector for the clustering of texture features of the coating surface mesh, so as to determine the cluster core of the global texture feature distribution of the coating surface, thereby providing a stable reference point for subsequent feature aggregation.

[0050] The above step S31 can be expressed by the formula as follows:

[0051]

[0052]

[0053]

[0054]

[0055] in, This represents the set of feature vectors representing the mesh grain texture of the coating surface. , , , and These represent the first, second, and third features in the set of mesh grain texture feature vectors on the coating surface, respectively. The, the The and the first Each coating surface mesh grain texture feature vector The value is the number of the mesh grain texture feature vectors on the coating surface. Indicates the first In the feature vector of the mesh grain texture of the coating surface, the first... Feature values ​​at each position, and Respectively represent the first The eigenmean and squared eigenvariance of the feature vector of each coating surface mesh grain texture. This indicates the calculation of the expected value of the set. This represents the sigmoid activation function. Indicates the first The static energy factor of the feature vector of the mesh grain texture of the coating surface. This indicates taking the index corresponding to the maximum value. The index representing the largest static energy factor in the set of static energy factors representing the mesh grain texture features of the coating surface. This represents the coating surface mesh grain texture feature vector corresponding to the index of the largest static energy factor in the set of static energy factors of the coating surface mesh grain texture features. This represents the initial center vector for clustering the mesh-grained texture features of the coating surface.

[0056] Specifically, step S32 includes: calculating the dynamic aggregation energy factor of each coating surface mesh grain texture feature vector based on the spatial span between each coating surface mesh grain texture feature vector in the set of coating surface mesh grain texture feature vectors and the initial center vector of the coating surface mesh grain texture feature clustering, and the set of static energy factors of the coating surface mesh grain texture features, to obtain a set of dynamic aggregation energy factors of the coating surface mesh grain texture features; inputting the set of dynamic aggregation energy factors of the coating surface mesh grain texture features into a gating mask unit to obtain a set of dynamic aggregation weight factors of the coating surface mesh grain texture features; and calculating the weighted sum of the set of coating surface mesh grain texture feature vectors based on the set of dynamic aggregation weight factors of the coating surface mesh grain texture features to obtain the global salient aggregation representation vector of the coating surface texture mesh grain.

[0057] In a specific example of this application, calculating the dynamic aggregation energy factor of each coating surface mesh grain texture feature vector to obtain a set of dynamic aggregation energy factors of coating surface mesh grain texture features includes: using the square of the number of feature vectors separated by the coating surface mesh grain texture feature vector and the initial center vector of the coating surface mesh grain texture feature cluster as a spatial span coefficient, and calculating the weighted ratio between the product of the static energy factor of the coating surface mesh grain texture feature vector and the static energy factor of the initial center vector of the coating surface mesh grain texture feature cluster cluster and the spatial span coefficient to obtain the dynamic aggregation energy factor of the coating surface mesh grain texture features.

[0058] In a specific example of this application, inputting the set of dynamically aggregated energy factors of the coating surface mesh grain texture features into a gating mask unit includes: inputting the set of dynamically aggregated energy factors of the coating surface mesh grain texture features into a sigmoid function for normalization processing to obtain a normalized set of dynamically aggregated energy factors of the coating surface mesh grain texture features; and deactivating the normalized set of dynamically aggregated energy factors of the coating surface mesh grain texture features based on a preset mask threshold to obtain a set of dynamically aggregated weight factors of the coating surface mesh grain texture features.

[0059] In other words, considering that adjacent regions in an image often have more similar texture characteristics, this application further derives the dynamic aggregation energy factor of each coating surface mesh granularity texture feature vector based on the spatial span of each coating surface mesh granularity texture feature vector relative to the initial center vector of the coating surface mesh granularity texture feature cluster, and combines their respective static energy factors. That is, by considering the feature space proximity of each coating surface mesh granularity texture feature vector relative to the core of the coating surface global texture feature distribution cluster and the importance differences of its intrinsic attributes, its contribution to the global texture feature pattern of the coating surface is more accurately measured. Then, based on a gating mask mechanism, the obtained dynamic aggregation energy factor is nonlinearly transformed and filtered to generate corresponding weights, which are used to perform weighted aggregation of the set of coating surface mesh granularity texture feature vectors, thereby enhancing the expression of important features and suppressing the influence of irrelevant or noisy features, to obtain a representative global salient aggregation representation vector of coating surface texture mesh granularity.

[0060] Step S32 above can be expressed by the formula as follows:

[0061]

[0062]

[0063]

[0064]

[0065]

[0066] in, Indicates the first The static energy factor of the feature vector of the mesh grain texture of the coating surface. The static energy factor represents the initial center vector of the clustering of the mesh-grained texture features on the coating surface. and For different weight parameters, Indicates the first The number of feature vectors separating each coating surface mesh grain texture feature vector from the initial center vector of the coating surface mesh grain texture feature clustering. Indicates the first Dynamic aggregation energy factor of the feature vector of mesh grain size of a coating surface Indicates the first A normalized coating surface mesh grain texture feature dynamic aggregation energy factor Indicates masking processing, To preset the mask threshold, The value is the number of the mesh grain texture feature vectors on the coating surface. Indicates the first Dynamic aggregation weighting factor for the mesh grain texture features of the coating surface This represents the globally significant aggregated representation vector of the texture mesh granularity of the coating surface.

[0067] In the above-mentioned method for preparing rare-earth reflective heat-insulating metal plates, step S4 generates the roll coating quality inspection result based on the feature differences between each coating surface mesh grain texture feature vector in the set of coating surface mesh grain texture feature vectors and the globally significant aggregated representation vector of coating surface texture mesh grain. The roll coating quality inspection result is used to indicate whether the coating surface is dense. Figure 4 This is a flowchart of step S4 in the preparation method of the rare-earth reflective heat-insulating metal plate according to an embodiment of this application. Figure 4 As shown, step S4 includes: S41, calculating the coating surface state semantic difference coefficient between each coating surface mesh grain texture feature vector in the set of coating surface mesh grain texture feature vectors and the coating surface texture mesh grain global saliency aggregated representation vector to obtain a coating surface state global distribution representation vector composed of multiple coating surface state semantic difference coefficients; S42, inputting the coating surface state global distribution representation vector into the classifier-based roll coating quality detection module to obtain the roll coating quality detection result.

[0068] Specifically, in step S41, the semantic difference coefficient of the coating surface state between each coating surface mesh grain texture feature vector in the set of coating surface mesh grain texture feature vectors and the global significant aggregated representation vector of the coating surface texture mesh grain is calculated to obtain a global distribution representation vector of the coating surface state composed of multiple coating surface state semantic difference coefficients. That is, by further calculating the semantic difference coefficient between each coating surface mesh grain texture feature vector and the global significant aggregated representation vector of the coating surface texture mesh grain, the difference between the texture features of each local region in the coating surface state image and the main texture features of the image as a whole is quantified. Based on the semantic difference coefficients of each coating surface state, the global texture distribution pattern of the coating surface is revealed, capturing the overall texture distribution uniformity information of the coating surface state image, thus providing an important basis for evaluating the density of the coating. In a specific example of this application, the Mahalanobis distance between each coating surface mesh grain texture feature vector in the set of coating surface mesh grain texture feature vectors and the global significant aggregated representation vector of the coating surface texture mesh grain is calculated as the semantic difference coefficient of the coating surface state to obtain the global distribution representation vector of the coating surface state.

[0069] Specifically, in step S42, the global distribution representation vector of the coating surface state is input into the roll coating quality detection module based on a classifier to obtain the roll coating quality detection result. In the technical solution of this application, the classifier, through training, can effectively learn the coating surface texture distribution uniformity information contained in the global distribution representation vector of the coating surface state, and distinguish between dense and non-dense states of the coating surface accordingly. This enables automatic classification and evaluation of roll coating quality, provides real-time feedback for the production process, guides production adjustments, and ensures the consistency and reliability of coating quality.

[0070] In the technical solution of this application, since the global salient aggregated representation vector of the coating surface texture mesh granularity is obtained by clustering analysis of the set of coating surface mesh granularity texture feature vectors, it is not the true semantic measurement center of the surface state. Therefore, when using the global salient aggregated representation vector of the coating surface texture mesh granularity to calculate the semantic difference coefficient of the coating surface state of each coating surface mesh granularity texture feature vector in the set of coating surface mesh granularity texture feature vectors, additional interference information will be introduced. The interference component of the global distribution representation vector of the coating surface state will ensure that the overall feature manifold of the global distribution representation vector of the coating surface state in the high-dimensional feature space has fine-grained feature structure holes. The existence of fine-grained feature structure holes will not only make the semantic coverage of the feature corresponding class probability label of the global distribution representation vector of the coating surface state insufficient, but also cause the outlier class regression inference mapping to be off-target, affecting the accuracy of the roll coating quality detection result obtained by the roll coating quality detection module based on the classifier of the global distribution representation vector of the coating surface state.

[0071] In the technical solution of this application, the global distribution representation vector of the coating surface state is input into the roll coating quality detection module based on the classifier to obtain the roll coating quality detection result, including: inputting the global distribution representation vector of the coating surface state into the pre-classifier based on the Softmax function to obtain the global fine-grained class probability label vector of the coating surface state;

[0072] Calculate the product between the global fine-grained class probability label vector of the coating surface state and its transpose vector to obtain the global fine-grained full label domain modulation matrix of the coating surface state;

[0073] Divide the eigenvalues ​​at each position in the global fine-grained full-label domain modulation matrix of the coating surface state by the square root of the scale value of the global distribution representation vector of the coating surface state to obtain the global fine-grained full-label domain scale modulation matrix of the coating surface state.

[0074] The global fine-grained full-label domain scale modulation matrix of the coating surface state is input into a probabilistic domain attention sparse module based on a multi-level gating function to obtain the global fine-grained full-label domain scale modulation matrix of the coating surface state.

[0075] Using the global distribution representation vector of the coating surface state as the query feature vector, the global fine-grained full-label domain scale modulation matrix of the coating surface state is multiplied with the global distribution representation vector of the coating surface state to obtain the optimized global distribution representation vector of the coating surface state; and

[0076] The optimized global distribution representation vector of the coating surface state is input into the classifier-based roll coating quality detection module to obtain the roll coating quality detection result.

[0077] The optimization process of the global distribution representation vector of the coating surface state is expressed by the following formula:

[0078]

[0079]

[0080]

[0081]

[0082]

[0083] in, The first element representing the global distribution characterization vector of the coating surface state Feature values ​​at each position, The scale value represents the global distribution representation vector of the coating surface state. The first element represents the global fine-grained class probability label vector of the coating surface state. Feature values ​​at each position, This represents the global fine-grained class probability label vector representing the surface state of the coating. Represents matrix multiplication. Represents the transpose of a vector. This represents the global fine-grained, full-label-domain modulation matrix representing the surface state of the coating. Indicates the predetermined threshold. The first element representing the global fine-grained full-label domain scale modulation matrix of the coating surface state. Feature values ​​at each position, Indicates the pre-defined hyperparameters. Represents the mask function, The first element representing the global fine-grained, full-label-domain sparse modulation matrix of the coating surface state. Feature values ​​at each position, This represents a global fine-grained, full-label-domain sparse modulation matrix representing the surface state of the coating. This represents the global distribution representation vector of the optimized coating surface state.

[0084] Therefore, by adding a class-probability domain-level optimizable perturbation to the original feature vector to strengthen the entanglement of the dependency relationship between different variables in the original feature vector based on label domain modulation, the source domain feature vector can be optimized by class-probability query-based scaling modulation. In this way, the significant intrinsic modal information in the source domain features can be preserved more effectively, and the interference components in the feature distribution can be effectively masked. This improves the adversarial robustness of the manifold representation of the feature vector, thereby improving the accuracy of the roll coating quality detection result obtained by the classifier-based roll coating quality detection module.

[0085] According to another aspect of this application, a rare-earth reflective heat-insulating metal plate is also provided, which is prepared by the method described above. The performance indicators of the rare-earth reflective heat-insulating coating are shown in Table 1.

[0086] Table 1 Performance Indicators of Rare Earth Reflective Thermal Insulation Metal Panels

[0087]

[0088] As shown in Table 1, the rare-earth reflective heat-insulating metal plate described in this application embodiment possesses excellent heat insulation and reflective properties, effectively reducing the surface temperature of the metal plate while maintaining good physical and chemical stability. Furthermore, the rare-earth reflective heat-insulating metal plate also exhibits good water resistance and aging resistance, ensuring the long-term stability and reliability of the coating in various environments. Its salt spray resistance also indicates that the metal plate can effectively resist corrosion and extend its service life when used in marine environments or areas with high salt spray levels. Simultaneously, the short drying time and moderate coating thickness meet the requirements for rapid drying and coating protection performance in industrial applications.

[0089] Specifically, the reflective heat-insulating coating has the following technical features:

[0090] First, it utilizes dual-reflection technology, achieving a reflectivity >92% and full-spectrum reflection across the 150-2000nm wavelength range. Ordinary traditional reflective coatings only offer single-reflection and lack dual-reflection capabilities. By employing rare-earth composite materials to reflect the solar spectrum, the molecular microstructure forms an overlapping, layered double-layer film system, altering the coating's single-reflection capability to achieve dual-reflection. When sunlight shines on the coating, the first layer reflects and removes over 85% of infrared, ultraviolet, and visible light, while the second layer reflects and removes light transmitted through the first layer, further increasing the reflectivity by 5-8%. This dual-reflection effectively prevents solar heat accumulation and inhibits heat buildup; the average reflectivity under dual-reflection is >90%.

[0091] Secondly, the hemispherical emissivity of the reflective heat-insulating coating is >85%, enhancing infrared thermal radiation, achieving heat dissipation and cooling, and its radiation band is: Excellent reflective sun-protective and cooling paints must possess superior spatial thermal radiation performance. Utilizing rare-earth composite coating ultrafine hollow microsphere technology and materials, a special microscopically textured, high-radiation surface is formed, increasing the surface area several times over. This allows for the radiant removal of solar heat that is not reflected off the surface using a wavelength of 2-15.5μm, further reducing the surface temperature and achieving effective heat dissipation and cooling control.

[0092] Furthermore, the reflective heat-insulating coating has a vacuum heat shield layer that blocks heat, prevents heat exchange and conduction, and employs a microsphere vacuum shield layer composite structure technology to reduce the coating's thermal conductivity and thus reduce heat conduction. The coating cures to form a tangentially bonded, three-dimensional spatial structure. Microparticles of different sizes fill the gaps between molecules, ultimately forming a composite heat shield layer that effectively prevents heat exchange and conduction between the internal and external heat energy, avoiding internal temperature rise. A 1mm thick rare-earth reflective heat-insulating coating layer reflects approximately 90%-95% of radiant heat, equivalent to a 10mm thick polystyrene foam with an R-value of 20.

[0093] Furthermore, the reflective heat-insulating coating utilizes both water-based and oil-based coating technologies. The oil-based coating technology offers superior characteristics such as high weather resistance, aging resistance, long lifespan, salt spray resistance, waterproofing, corrosion resistance, and strong adhesion.

[0094] Furthermore, the aforementioned reflective heat-insulating coating is suitable for spraying and roller coating, and is suitable for making pre-roll-coated metal plates with a film thickness controlled between 80um and 200um. It can be industrialized, has strong processability, and is easy to punch, cut, and fold, thus overcoming the weaknesses of on-site construction.

[0095] In particular, when the reflective heat-insulating coating is used to make a pre-roll-coated metal plate and to obtain the rare earth reflective heat-insulating metal plate, since the thickness of the reflective heat-insulating coating is only 80um-200um, the obtained rare earth reflective heat-insulating metal plate is easy to punch, cut, fold, and roll, thus facilitating carrying, packaging, and processing.

[0096] Finally, it should be noted that the embodiments described above are only some, not all, of the embodiments of the present invention. The detailed description of the embodiments of the present invention is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

Claims

1. A roll coating production process for rare earth reflective heat-insulating metal plates, characterized in that, include: We provide rare earth reflective heat-insulating coatings and metal panels; The metal plate is subjected to surface treatment to obtain a surface-treated metal plate; The rare earth reflective heat-insulating coating is applied to the surface-treated metal plate by a roller coating process to obtain a rare earth reflective heat-insulating metal plate. The rare earth reflective heat insulation metal plate is subjected to roll coating quality inspection to obtain roll coating quality inspection results, thereby realizing automatic classification and evaluation of roll coating quality, providing real-time feedback for the production process, guiding production adjustments, and ensuring the consistency and reliability of coating quality; The process of performing roll coating quality inspection on the rare earth reflective heat-insulating metal plate to obtain roll coating quality inspection results includes: acquiring a coating surface state image captured by an industrial camera; extracting texture features based on grid granularity from the coating surface state image to obtain a set of coating surface grid granularity texture feature vectors; performing feature saliency modulation dynamic aggregation on the set of coating surface grid granularity texture feature vectors to obtain a global saliency aggregation representation vector of coating surface texture grid granularity; and generating the roll coating quality inspection result based on the feature differences between each coating surface grid granularity texture feature vector in the set of coating surface grid granularity texture feature vectors and the global saliency aggregation representation vector of coating surface texture grid granularity, wherein the roll coating quality inspection result is used to indicate whether the coating surface is dense; The process of extracting texture features based on grid granularity from the coating surface state image to obtain a set of coating surface grid granularity texture feature vectors includes: dividing the coating surface state image into grids to obtain a set of coating surface state grid granularity image blocks; and inputting each coating surface state grid granularity image block in the set of coating surface state grid granularity image blocks into a texture feature extractor based on a multi-layer dilated convolution structure to obtain the set of coating surface grid granularity texture feature vectors. The roll coating quality inspection result is generated based on the feature differences between each coating surface mesh grain texture feature vector in the set of coating surface mesh grain texture feature vectors and the globally significant aggregated representation vector of coating surface texture mesh grain, including: The semantic difference coefficients of the coating surface state between each coating surface mesh grain texture feature vector in the set of coating surface mesh grain texture feature vectors and the global salient aggregated representation vector of the coating surface texture mesh grain are calculated to obtain a global distribution representation vector of the coating surface state composed of multiple coating surface state semantic difference coefficients. In this process, the semantic difference coefficients between each coating surface mesh grain texture feature vector and the global salient aggregated representation vector of the coating surface texture mesh grain are calculated to quantify the difference between the texture features of each local region in the coating surface state image and the main texture features of the image as a whole. Based on the semantic difference coefficients of each coating surface state, the global texture distribution pattern of the coating surface is revealed, and the overall texture distribution uniformity information of the coating surface state image is captured. The global distribution representation vector of the coating surface state is input into the classifier-based roll coating quality detection module to obtain the roll coating quality detection result.

2. The roll coating production process for rare earth reflective heat-insulating metal plates according to claim 1, characterized in that, The set of feature vectors for the textured mesh size of the coating surface is dynamically aggregated using feature saliency modulation to obtain a globally saliency aggregated representation vector for the textured mesh size of the coating surface, including: Based on the feature distribution field of the set of feature vectors of the mesh grain texture of the coating surface, the initial center vector for clustering the feature of the mesh grain texture of the coating surface is determined; Based on the spatial span of each coating surface mesh grain texture feature vector in the set of coating surface mesh grain texture feature vectors relative to the initial center vector of the coating surface mesh grain texture feature clustering, the set of coating surface mesh grain texture feature vectors is dynamically aggregated with saliency modulation to obtain the global saliency aggregated representation vector of the coating surface texture mesh grain.

3. The roll coating production process for rare earth reflective heat-insulating metal plates according to claim 2, characterized in that, Based on the feature distribution field of the set of feature vectors of the coating surface mesh grain texture, the initial center vector for clustering the feature of the coating surface mesh grain texture is determined, including: Calculate the static energy factor of each coating surface mesh grain texture feature vector in the set of coating surface mesh grain texture feature vectors to obtain the set of static energy factors of coating surface mesh grain texture features; The coating surface mesh grain texture feature vector corresponding to the maximum value in the set of static energy factors of the coating surface mesh grain texture features is selected as the initial center vector for clustering the coating surface mesh grain texture features.

4. The roll coating production process for rare earth reflective heat-insulating metal plates according to claim 3, characterized in that, Calculate the static energy factor of each coating surface mesh grain texture feature vector in the set of coating surface mesh grain texture feature vectors to obtain the set of static energy factors of coating surface mesh grain texture features, including: The kurtosis of the texture feature vector of the coating surface mesh is calculated, and the kurtosis is input into the sigmoid activation function to obtain the static energy factor of the texture feature of the coating surface mesh.

5. The roll coating production process for rare earth reflective heat-insulating metal plates according to claim 4, characterized in that, Based on the spatial span of each coating surface mesh grain texture feature vector in the set of coating surface mesh grain texture feature vectors relative to the initial center vector of the coating surface mesh grain texture feature clustering, the set of coating surface mesh grain texture feature vectors is dynamically aggregated with saliency modulation to obtain the globally saliency aggregated representation vector of the coating surface texture mesh grain, including: Based on the spatial span between each coating surface mesh grain texture feature vector in the set of coating surface mesh grain texture feature vectors and the initial center vector of the coating surface mesh grain texture feature cluster, and the set of static energy factors of the coating surface mesh grain texture features, the dynamic aggregation energy factor of each coating surface mesh grain texture feature vector is calculated to obtain the set of dynamic aggregation energy factors of the coating surface mesh grain texture features. The set of dynamic aggregation energy factors of the coating surface mesh grain texture features is input into the gated mask unit to obtain the set of dynamic aggregation weight factors of the coating surface mesh grain texture features; The set of dynamic aggregation weight factors for the texture features of the coating surface mesh granularity is used to calculate the weighted sum of the set of texture feature vectors of the coating surface mesh granularity to obtain the global salient aggregated representation vector of the texture mesh granularity of the coating surface.

6. The roll coating production process for rare earth reflective heat-insulating metal plates according to claim 5, characterized in that, Calculate the dynamic aggregation energy factor of the mesh grain texture feature vectors of each coating surface to obtain a set of dynamic aggregation energy factors of the coating surface mesh grain texture features, including: The spatial span coefficient is obtained by taking the square of the number of feature vectors that are separated from the feature vector of the coating surface mesh grain texture feature vector and the initial center vector of the clustering of the coating surface mesh grain texture feature vector as the spatial span coefficient, and calculating the weighted ratio between the product of the static energy factor of the feature vector of the coating surface mesh grain texture feature vector and the static energy factor of the initial center vector of the clustering of the coating surface mesh grain texture feature vector and the spatial span coefficient.

7. The roll coating production process for rare earth reflective heat-insulating metal plates according to claim 6, characterized in that, Calculate the coating surface state semantic difference coefficient between each coating surface mesh grain texture feature vector in the set of coating surface mesh grain texture feature vectors and the coating surface texture mesh grain global salient aggregated representation vector to obtain a coating surface state global distribution representation vector composed of multiple coating surface state semantic difference coefficients, including: The Mahalanobis distance between each coating surface mesh grain texture feature vector in the set of coating surface mesh grain texture feature vectors and the global salient aggregated representation vector of coating surface texture mesh grain is calculated as the semantic difference coefficient of the coating surface state to obtain the global distribution representation vector of the coating surface state.

Images