Automatic control system and method for coating machine

Abstract

The present application provides an automatic control system and method for a coating machine, which relates to the field of intelligent control. The system first obtains the slurry coating state image and the slurry characteristic text description of the target detection area, and then uses machine learning and natural language analysis technology to perform image encoding on the slurry coating state image, and semantic encoding on the slurry characteristic text description, and generates the detection surface density corresponding to the slurry in the target detection area based on the cross-modal significant interaction of the core clues of the two, so as to adjust the coating machine in combination with the target surface density corresponding to the slurry. In this way, by fusing multimodal data, it is possible to reduce the dependence on environmental factors of a single sensor or parameter estimation method, better understand and process complex coating processes, and make adaptive adjustments according to the characteristics of different slurries, with high flexibility.

Description

Technical Field

[0001] The present application relates to the field of intelligent control, and more particularly to an automatic control system and method for a coating machine in an embodiment of the present application. Background Art

[0002] In modern manufacturing, especially in industries involving slurry coating processes (such as coating material production, etc.), the requirements for coating quality are getting higher and higher. Traditionally, these processes often rely on manual experience for adjustment and control, which is not only inefficient, but also easily affected by human factors, resulting in unstable product quality.

[0003] In this regard, the existing patent CN116351659A proposes a coating machine control method and device, and a coating machine, which obtains the detection surface density corresponding to the slurry in the target detection area in real time during the coating process, and uses the coating control model trained by deep learning to input the detection surface density and the target surface density into this model to determine the best coating setting information, and then adjusts the motor position, pump speed, gap and other parameters through simulation, and continuously iterates until the simulation surface density meets the target requirements, and finally automatically adjusts the coating machine settings according to the actual production situation to ensure that the expected coating effect is achieved. This improves production efficiency, reduces uncertainty caused by human intervention, and improves the consistency of product quality.

[0004] In this patent, the detection surface density corresponding to the slurry in the target detection area is obtained by direct measurement by the sensor or estimation based on the coating die output parameters. However, the sensor is easily affected by environmental factors (such as temperature, humidity, dust, etc.), resulting in inaccurate measurement results. The estimation method based on the coating die output parameters relies on a series of assumptions (such as flow rate, density, etc.). If these parameters change during the actual production process, the estimation results may be inaccurate.

[0005] Therefore, an optimized automatic control solution for the coating machine is desired. Summary of the invention

[0006] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides an automatic control system and method for a coating machine, which first obtains the slurry coating state image and the slurry characteristic text description of the target detection area, and then uses machine learning and natural language analysis technology to perform image encoding on the slurry coating state image, and semantic encoding on the slurry characteristic text description, and generates the detection surface density corresponding to the slurry in the target detection area based on the cross-modal significant interaction representation of the core clues of the two, so as to adjust the coating machine in combination with the target surface density corresponding to the slurry. In this way, by fusing multimodal data, it is possible to reduce the dependence on environmental factors of a single sensor or parameter estimation method, better understand and process complex coating processes, and make adaptive adjustments according to the characteristics of different slurries, with high flexibility.

[0007] According to one aspect of the present application, an automatic control method for a coating machine is provided, comprising: when the coating machine performs a slurry coating operation, obtaining a slurry coating state image of a target detection area captured by a camera; extracting a characteristic text description of the slurry from a slurry feature database; based on the slurry coating state image and the characteristic text description of the slurry, obtaining a slurry coating state-slurry attribute cross-modal significant interaction feature, comprising: encoding the slurry coating state image and the characteristic text description of the slurry and cross-domain combining the slurry coating state-slurry attribute to obtain the slurry coating state-slurry attribute cross-modal significant interaction feature; based on the slurry coating state-slurry attribute cross-modal significant interaction feature, obtaining a detection surface density corresponding to the slurry in the target detection area; and adjusting the coating machine based on the detection surface density corresponding to the slurry in the target detection area and the target surface density.

[0008] According to another aspect of the present application, an automatic control system for a coating machine is provided, which includes: a slurry coating state image acquisition module, used to acquire a slurry coating state image of a target detection area captured by a camera when the coating machine performs a slurry coating operation; a slurry characteristic text description module, used to extract a characteristic text description of the slurry from a slurry characteristic database; a slurry coating state-slurry attribute cross-modal significant interaction module, used to obtain a slurry coating state-slurry attribute cross-modal significant interaction feature based on the slurry coating state image and the characteristic text description of the slurry; a slurry detection surface density acquisition module, used to obtain a detection surface density corresponding to the slurry in the target detection area based on the slurry coating state-slurry attribute cross-modal significant interaction feature; a coating machine adjustment module, used to adjust the coating machine based on the detection surface density corresponding to the slurry in the target detection area and the target surface density.

[0009] Compared with the prior art, the present application provides an automatic control system and method for a coating machine, which first obtains the slurry coating state image and slurry characteristic text description of the target detection area, and then uses machine learning and natural language analysis technology to perform image encoding on the slurry coating state image, semantic encoding on the slurry characteristic text description, and based on the core clues of the two, the cross-modal significant interaction representation generates the detection surface density corresponding to the slurry in the target detection area, so as to adjust the coating machine in combination with the target surface density corresponding to the slurry. In this way, by fusing multimodal data, it is possible to reduce the dependence on environmental factors of a single sensor or parameter estimation method, better understand and process complex coating processes, and make adaptive adjustments according to the characteristics of different slurries, with high flexibility. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0011] Figure 1 Flow chart of an automatic control method of a coating machine according to an embodiment of the present application.

[0012] Figure 2 Schematic diagram of data flow of an automatic control method for a coating machine according to an embodiment of the present application.

[0013] Figure 3 A flowchart for encoding the slurry coating state image and the characteristic text description of the slurry and cross-domain combination of the slurry coating state and slurry properties in the automatic control method of the coating machine according to an embodiment of the present application to obtain the cross-modal significant interaction features of the slurry coating state and slurry properties.

[0014] Figure 4 The present invention provides a flowchart for cross-domain joint encoding of the slurry coating state-slurry attribute core clues on the slurry coating state image encoding feature map and the slurry characteristic text semantic encoding vector in the automatic control method of the coating machine according to an embodiment of the present application to obtain a slurry coating state-slurry attribute cross-modal significant interaction encoding vector.

[0015] Figure 5 The present invention is a flowchart for calculating the slurry coating state-slurry attribute core clue weaving template matrix between the slurry coating state image coding feature map and the slurry property text semantic coding vector in the automatic control method of the coating machine according to the embodiment of the present application.

[0016] Figure 6 4 is a system block diagram of an automatic control system of a coating machine according to an embodiment of the present application. DETAILED DESCRIPTION

[0017] Various exemplary embodiments, features and aspects of the present application will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0018] The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

[0019] In addition, in order to better illustrate the present application, numerous specific details are given in the following specific embodiments. It should be understood by those skilled in the art that the present application can also be implemented without certain specific details. In some examples, methods, means, components and circuits well known to those skilled in the art are not described in detail in order to highlight the subject matter of the present application.

[0020] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0021] In modern manufacturing, the quality requirements for slurry coating processes (such as coating material production) are increasing, but the traditional control method that relies on manual experience is inefficient and easily affected by human factors, resulting in unstable product quality. Patent CN116351659A proposes a coating machine control method that combines the detection surface density with the target surface density through a deep learning model, simulates and optimizes parameters such as motor position, pump speed, and gap, and automatically adjusts the coating machine settings to ensure the coating effect and improve efficiency and consistency. However, there are limitations in obtaining the detection surface density: sensor measurements are easily interfered by environmental factors, and the estimation method based on the output parameters of the coating die relies on assumptions. Parameter changes in actual production may lead to inaccurate results.

[0022] In response to the above technical problems, the present application proposes an automatic control method for a coating machine. Figure 1 It is a flow chart of an automatic control method of a coating machine according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of the automatic control method of the coating machine according to the embodiment of the present application. Figure 1 and Figure 2As shown, the automatic control method of the coating machine according to the embodiment of the present application includes: S110, when the coating machine performs a slurry coating operation, obtaining a slurry coating state image of the target detection area captured by a camera; S120, extracting a characteristic text description of the slurry from a slurry feature database; S130, based on the slurry coating state image and the characteristic text description of the slurry, obtaining a slurry coating state-slurry attribute cross-modal significant interaction feature; S140, based on the slurry coating state-slurry attribute cross-modal significant interaction feature, obtaining a detection surface density corresponding to the slurry in the target detection area; S150, based on the detection surface density corresponding to the slurry in the target detection area and the target surface density, adjusting the coating machine.

[0023] Accordingly, the technical concept of the present application is to obtain the slurry coating state image of the target detection area collected by the camera, and extract the characteristic text description of the slurry from the slurry feature database, and use the technology based on machine learning and natural language analysis to image encode the slurry coating state image, and semantically encode the characteristic text description of the slurry, so as to automatically generate the detection surface density corresponding to the slurry in the target detection area according to the core clue cross-modal significant interaction between the slurry coating state image encoding feature and the slurry characteristic text semantic encoding feature, and adjust the coating machine in combination with the target surface density corresponding to the slurry in the target detection area. The present application collects the slurry coating state image of the target detection area through the camera, and combines the characteristic text description extracted from the slurry feature database, providing a richer data source. This multimodal data fusion can effectively reduce the dependence of a single sensor or parameter estimation method on environmental factors. In addition, it can better understand and handle complex coating processes, make adaptive adjustments according to the characteristics of different slurries, and have high flexibility.

[0024] In the above-mentioned automatic control method of the coating machine, the step S110, when the coating machine performs the slurry coating operation, obtains the slurry coating state image of the target detection area collected by the camera. It should be understood that in order to ensure that the camera can clearly capture the slurry coating state in the target detection area, the camera needs to be fixed in a stable position near the coating head. This position should allow the camera field of view to cover the entire coating width and avoid vibration or other interference factors generated during the coating process affecting the imaging quality. In addition, considering that the coating environment may have unfavorable conditions such as dust and moisture, the camera itself also needs to have a certain level of protection to resist the influence of these external environmental factors. At the same time, in order to reduce the impact of light changes on the imaging effect, appropriate lighting equipment is usually configured around the camera to ensure that consistent image quality can be obtained whether it is day or night, or under lighting conditions in different seasons. Next, considering that different coating processes require different resolutions and frame rates, for example, for high-speed coating production lines, higher frame rates help capture details in fast movement, while for low-speed production lines, the frame rate can be appropriately reduced to save storage space and processing time, and different resolutions and frame rates need to be set for different production lines. Furthermore, since some noise interference is inevitable in the actual production environment, such as randomly appearing bright spots or dark spots, these will cause trouble for subsequent analysis. Therefore, before the original image is sent to the next step of processing, it is usually necessary to perform a series of preprocessing operations, including but not limited to denoising, contrast enhancement, and correction of geometric distortion. Among them, denoising can be achieved by applying various filters, such as Gaussian filters for smoothing images, while median filters are good at removing isolated noise points. Contrast enhancement can help highlight the difference between the slurry coating and the background, making subsequent feature extraction easier. As for geometric distortion correction, it is mainly to compensate for the barrel or pincushion distortion that may occur when shooting with a wide-angle lens to ensure the accuracy of the measurement results.

[0025] In the above-mentioned automatic control method of the coating machine, the step S120 extracts the characteristic text description of the slurry from the slurry characteristic database. It should be understood that in the technical solution of the present application, the slurry characteristic database contains a wealth of slurry-related information, such as specific parameters such as physical properties, pH value, reducing agent type, or key indicators such as functional phase characteristics and chemical corrosion resistance. Next, in actual use, when it is necessary to extract the characteristic text description of a specific slurry from the database, the system will automatically match the relevant records according to the query conditions entered by the user. This link relies on the powerful database query function. By writing accurate search statements through SQL or other similar query languages, the relevant information of the target slurry can be quickly located. At the same time, in order to improve the query efficiency, indexing technology can also be used to accelerate the data retrieval speed, and the fuzzy matching algorithm can be combined to expand the search range to cover possible synonyms or synonymous expressions. However, relying solely on traditional database queries is not enough to fully meet the needs of current complex application scenarios. Therefore, advanced natural language processing technology is introduced on this basis to further enhance the intelligence level of the system. Specifically, this technology can help convert unstructured text content into structured data form, making it easier for computers to understand and operate. For example, through learning and training on a large amount of literature, the machine learning model can automatically identify and extract descriptive sentences related to a specific slurry, and then reorganize them into an easy-to-understand standard format according to the predefined template and output them to the user. More importantly, this method can not only process fixed entries statically stored in the database, but also dynamically adapt to new knowledge points, continuously enriching and improving the existing knowledge system.

[0026] In the above-mentioned automatic control method of the coating machine, the step S130, based on the slurry coating state image and the characteristic text description of the slurry, obtains the cross-modal significant interaction feature of the slurry coating state-slurry attribute, including: encoding the slurry coating state image and the characteristic text description of the slurry and cross-domain combination of the slurry coating state-slurry attribute to obtain the cross-modal significant interaction feature of the slurry coating state-slurry attribute. It should be understood that the slurry coating state image can reflect the texture, thickness distribution and possible defects of the coating surface, and the characteristic text description of the slurry contains key attributes such as slurry viscosity, fluidity, and solid content. By extracting the information of the two and combining them with the slurry coating state-slurry attribute cross-domain, the image feature vector and the text feature vector can be mapped to the same semantic space so that they can be compared and fused in the same dimension, thereby retaining the key information in the original data and revealing the deep association between the image and text modalities.

[0027] Figure 3The present invention is a flowchart of encoding the slurry coating state image and the characteristic text description of the slurry and cross-domain combination of the slurry coating state and slurry attribute to obtain the cross-modal significant interaction feature of the slurry coating state and slurry attribute in the automatic control method of the coating machine according to the embodiment of the present application. Figure 3 As shown, in an embodiment of the present application, the step S130, encoding the slurry coating state image and the characteristic text description of the slurry and cross-domain combining the slurry coating state and the slurry attribute to obtain the slurry coating state-slurry attribute cross-modal significant interaction feature, includes: S131, extracting the slurry coating state feature from the slurry coating state image to obtain a slurry coating state image encoding feature map; S132, semantically encoding the characteristic text description of the slurry to obtain a slurry characteristic text semantic encoding vector; S133, cross-domain combining the slurry coating state-slurry attribute core clues of the slurry coating state image encoding feature map and the slurry characteristic text semantic encoding vector to obtain a slurry coating state-slurry attribute cross-modal significant interaction encoding vector as the slurry coating state-slurry attribute cross-modal significant interaction feature.

[0028] In an embodiment of the present application, the step S131, extracting the slurry coating state features from the slurry coating state image to obtain the slurry coating state image coding feature map, includes: using a slurry coating state feature extractor based on DenseNet to perform feature extraction on the slurry coating state image to obtain the slurry coating state image coding feature map. It should be understood that, considering that the slurry coating state image contains a large amount of pixel information, the key information in the coating state image, such as the distribution uniformity of the slurry, thickness changes, whether there are defects, etc., is often not intuitively visible and hidden in complex pixel data. Based on this, in the present application, the slurry coating state features are extracted from the slurry coating state image to capture subtle texture and color changes, and the slurry coating state image coding feature map is obtained. In particular, in a specific example of the present application, the slurry coating state image can be feature extracted using a slurry coating state feature extractor based on DenseNet to obtain the slurry coating state image coding feature map. Among them, DenseNet is an advanced deep learning model. Compared with the traditional convolutional neural network (CNN), the biggest feature of DenseNet is that each layer is directly connected to all other layers. This densely connected design can not only promote the propagation of features throughout the network, but also effectively reduce the gradient vanishing problem and improve parameter utilization. Specifically, in DenseNet, the output of each layer is not only the input of the next layer, but also the additional input of all subsequent layers. This design enables each layer in the network to receive feature information from all previous layers, thereby achieving full flow and fusion of features. In the feature extraction of slurry coating state images, the coating state images usually contain a large amount of pixel information, including both global features (such as the overall uniformity of the coating) and local details (such as edge contours or tiny defects). Traditional convolutional neural networks may find it difficult to take into account both global and local features at the same time, while DenseNet, through its dense connection mechanism, can capture and retain rich feature information at different levels. For example, at the lower level, the network can extract the basic texture and color changes of the image; at the higher level, these low-level features can be further integrated to form higher-order semantic representations, such as the trend of coating thickness changes or the contrast difference in a specific area. In this way, different feature information in the image can fully flow and merge between layers, which helps the network capture richer and more comprehensive features. Features such as subtle textures, edge information, and contrast changes in different areas in the slurry coating can be better extracted and retained, thereby having a more comprehensive understanding of the slurry coating status.

[0029] In an embodiment of the present application, the step S132, semantically encoding the text description of the characteristics of the slurry to obtain the semantic encoding vector of the slurry characteristics text, includes: using a semantic encoder based on a bidirectional recurrent neural network model to semantically encode the text description of the characteristics of the slurry to obtain the semantic encoding vector of the slurry characteristics text. It should be understood that considering that the slurry characteristics text contains information about various characteristics of the slurry such as physics, chemistry, and rheology, these information are interrelated and complex, and these semantic information have an important influence and role in the determination of the surface density. Therefore, in the technical solution of the present application, the text description of the characteristics of the slurry is semantically encoded to effectively capture the deep semantic information contained in the text, and the semantic encoding vector of the slurry characteristics text is obtained. In particular, in a specific example of the present application, the text description of the characteristics of the slurry can be semantically encoded by using a semantic encoder based on a bidirectional recurrent neural network model to obtain the semantic encoding vector of the slurry characteristics text. It is worth mentioning that compared with the traditional unidirectional recurrent neural network, the bidirectional recurrent neural network (Bi-RNN) can simultaneously process forward and reverse sequence information, and fully capture the contextual dependencies in the text. Specifically, Bi-RNN consists of two independent RNNs: one is responsible for processing the input sequence from front to back, and the other is responsible for processing the input sequence from back to front. This design enables the model to not only consider the previous information at any time, but also combine the later information to make more accurate predictions or encoding. For example, when describing the slurry as "having good fluidity and being stable even at high temperatures", Bi-RNN can simultaneously use the information before and after "fluidity" to accurately understand the semantics of this feature in the entire text. In this way, the various characteristics of the slurry expressed in the text and their interrelationships can be accurately reflected, providing a reliable semantic basis for subsequent analysis and decision-making.

[0030] Further, in the step S133, the slurry coating state image coding feature map and the slurry property text semantic coding vector are cross-domain jointly coded with the slurry coating state-slurry property core clues to obtain the slurry coating state-slurry property cross-modal significant interaction coding vector. It should be understood that considering that the slurry coating state image coding feature map mainly reflects the intuitive visual performance of the slurry during the coating process, such as the shape, distribution, surface state and other information of the coating. The slurry property text semantic coding vector focuses on describing the intrinsic physical, chemical and rheological properties of the slurry, such as viscosity, composition, drying characteristics, etc. These two data modalities provide information about slurry coating from different angles and are highly complementary. At the same time, there must be a potential causal relationship and mutual influence between the intrinsic properties of the slurry and the actual coating state. For example, the high viscosity of the slurry may cause uneven coating. Based on this, the present application performs cross-domain joint encoding of the slurry coating state-slurry attribute core clues on the slurry coating state image encoding feature map and the slurry characteristic text semantic encoding vector to obtain the slurry coating state-slurry attribute cross-modal significant interaction encoding vector. In particular, this method regards both as key elements and builds an association template and interaction system. By using the attention mechanism, heterogeneous transformation and template constraint learning, the semantic associations and complementary attributes between multimodal features are fully integrated to produce a joint representation with highly consistent semantics and excellent interactivity, so as to more comprehensively capture information related to the slurry surface density.

[0031] Figure 4 The present invention is a flowchart of performing cross-domain joint coding of the slurry coating state-slurry attribute core clues on the slurry coating state image coding feature map and the slurry characteristic text semantic coding vector in the automatic control method of the coating machine according to the embodiment of the present application to obtain the slurry coating state-slurry attribute cross-modal significant interaction coding vector. Figure 4 As shown, in an embodiment of the present application, the step S133, performs cross-domain joint encoding of the slurry coating state image coding feature map and the slurry property text semantic coding vector with the slurry coating state-slurry property core clues to obtain the slurry coating state-slurry property cross-modal significant interaction coding vector, including: S1331, calculating the slurry coating state-slurry property core clue weaving template matrix between the slurry coating state image coding feature map and the slurry property text semantic coding vector; S1332, based on the slurry coating state-slurry property core clue weaving template matrix, performs cross-modal fine-grained interaction on the slurry coating state image coding feature map and the slurry property text semantic coding vector to obtain the slurry coating state-slurry property cross-modal significant interaction coding vector.

[0032] Figure 5The present invention is a flow chart of calculating the slurry coating state-slurry attribute core clue weaving template matrix between the slurry coating state image coding feature map and the slurry property text semantic coding vector in the automatic control method of the coating machine according to the embodiment of the present application. Figure 5 As shown, in an embodiment of the present application, the step S1331, calculates the slurry coating state-slurry attribute core clue weaving template matrix between the slurry coating state image coding feature map and the slurry characteristic text semantic coding vector, including: S1331-1, performing core clue point convolution extraction on the slurry characteristic text semantic coding vector to obtain the slurry characteristic text semantic core clue coding vector; S1331-2, extracting the slurry coating state image feature core clue coding vector from the slurry coating state image coding feature map; S1331-3, constructing the slurry coating state-slurry attribute core clue weaving template matrix between the slurry characteristic text semantic core clue coding vector and the slurry coating state image feature core clue coding vector.

[0033] Specifically, in step 1331-1, the core clue point convolution extraction is performed on the slurry characteristic text semantic coding vector to obtain the slurry characteristic text semantic core clue coding vector, which is expressed as the core clue point convolution formula: ;in, is the slurry characteristic text semantic encoding vector, is the point convolutional coding, is the bias vector, is the activation function, is the slurry characteristic text semantic core clue encoding vector. It should be understood that this process can be regarded as a refiner of the modal space, aiming to draw an abstract and highly condensed representation from the input features to form the slurry characteristic text semantic core clue encoding vector, so that it can focus more on the key information of the slurry characteristics and effectively reduce redundant expressions. Specifically, point convolution is an efficient feature extraction method that can strengthen semantically significant modal specific features while maintaining the integrity of high-dimensional information. Unlike traditional convolution operations, point convolution directly acts on the feature dimension of each sample without involving sliding window operations in space or time. This design makes point convolution particularly suitable for processing high-dimensional data such as slurry characteristic text semantic encoding vectors, because it can directly capture the global relationship between features without considering local structures. In this way, by performing point convolution operations on the slurry characteristic text semantic encoding vector, key features that have an important impact on the coating process can be more accurately identified and extracted, thereby maintaining the integrity of high-dimensional information and strengthening semantically significant modal specific features. At the same time, point convolution effectively avoids redundant expression of information and focuses on the extraction of important elements of the modality.

[0034] Specifically, in the step S1331-2, a core clue coding vector of the slurry coating state image feature is extracted from the slurry coating state image coding feature map, and is expressed as a core clue coding formula: ;in, is the slurry coating state image coding feature map, is the meshing operation, and They are the first, second, and third local feature maps in the set of slurry coating state image encoding. and The slurry coating state image encodes the local feature map, is a convolutional code with a convolution kernel of 3×3. yes function, It is the first in the set of local enhancement feature maps of the slurry coating state image encoding. The slurry coating state image encodes the local enhanced feature map, is a global pooling operation, It is the first in the set of local enhancement feature vectors of the slurry coating state image coding. The slurry coating state image encodes the local enhanced feature vector, and Take the maximum and minimum values ​​respectively. and is the modulation parameter, and They are The mean and variance of The first local significant coefficient in the set of the slurry coating state image encoding The slurry coating state image encodes the local significant coefficients, Expressed as a natural constant The exponential function value with base , is the number of vectors in the set of local enhancement feature vectors for slurry coating state image coding, is the core clue encoding vector of the slurry coating state image feature. It should be understood that since the slurry coating state image encoding feature map usually carries stronger spatial dimension information, the processing method of this stage is different from the slurry characteristic text semantic encoding vector, which uses the regional attention mechanism to ensure the co-retention of spatial information and semantic features. Specifically, by introducing the attention mechanism, the model can dynamically adjust the importance weights of different regions, so that the significant regions are given priority and the spatial perception ability of the model is enhanced. For example, when analyzing the slurry coating state image, some regions may show obvious coating thickness changes, surface unevenness or potential defects. These regions are crucial for evaluating the coating quality, so they need to be given higher weights. Among them, the regional attention mechanism generates an attention map by calculating the correlation between each region and the global context to guide the subsequent feature extraction process. This design not only improves the model's sensitivity to key areas, but also effectively avoids the interference of irrelevant areas on the results. In addition to the regional attention mechanism, the convolution operation can maintain the integrity of high-dimensional spatial information while strengthening the structured information expression ability of the modality through its unique local receptive field characteristics. The convolution operation extracts local features of different scales and levels by scanning the slurry coating state image encoding feature map layer by layer. For example, at the lower level, the convolution operation can capture basic features such as edges and textures in the image; while at the higher level, these low-level features can be integrated to form higher-order semantic representations, such as coating thickness distribution, surface uniformity, etc., to better adapt to the spatial complexity of the coating state image. In this way, the salient areas of the slurry coating state image encoding feature map are prioritized through the re-weighting of attention, and at the same time, the local receptive field based on the convolution operation can enrich the structured information expression ability of the modality.

[0035] In an embodiment of the present application, the step S1331-3, constructing the slurry coating state-slurry attribute core clue weaving template matrix between the slurry characteristic text semantic core clue coding vector and the slurry coating state image feature core clue coding vector, includes: S1331-31, using a feature mapping function to process the slurry characteristic text semantic core clue coding vector and the slurry coating state image feature core clue coding vector to obtain a slurry coating state-slurry attribute core association matrix; S1331-32, processing the slurry characteristic text semantic core clue coding vector and the slurry coating state image feature core clue coding vector to obtain a slurry coating state-slurry attribute core association matrix; The clue coding vector performs semantic alignment of the different modal mapping residuals to obtain the slurry coating state-slurry attribute semantic alignment vector and the slurry attribute-slurry coating state semantic alignment vector; S1331-33, calculate the slurry coating state-slurry attribute significant difference matrix between the slurry coating state-slurry attribute semantic alignment vector and the slurry attribute-slurry coating state semantic alignment vector; S1331-34, based on the slurry coating state-slurry attribute significant difference matrix, the slurry coating state-slurry attribute core association matrix is ​​fine-grained significantly compensated to obtain the slurry coating state-slurry attribute core clue weaving template matrix. Among them, the step S1331-3 is expressed by the formula: ;in, is the transpose operation, and is a feature mapping function, such as a linear mapping or a nonlinear kernel function, is the core correlation matrix of slurry coating state and slurry properties, It is subtracted by position point. and They are the slurry coating state-slurry attribute semantic alignment vector and the slurry attribute-slurry coating state semantic alignment vector. To calculate the square of the vector's two-norm, is the logarithmic function value with base 2, is matrix multiplication, is the slurry coating state-slurry property significant difference matrix, It is the slurry coating state - the slurry properties core clue weaving template matrix.

[0036] It should be understood that after the extraction of the two modal feature core clues, the slurry characteristic text semantic core clue encoding vector and the slurry coating state image feature core clue encoding vector, is completed, the core association matrix construction phase is entered. The template matrix is ​​the core prior representation of the cross-modal interaction engine, and its function can be compared to a graph structure that establishes an explicit association between the two modalities. The edge weights of this graph structure are driven by the intrinsic relationship between the core clues of the modalities, which can reflect the global semantic constraints between the modalities and can also capture the fine-grained relationship between local significant areas. In specific implementation, it can be calculated by dot product similarity measurement, attention mapping mechanism or other methods.

[0037] In particular, the slurry feature text semantic core clue encoding vector for heterogeneous modality feature core clue representation and the core clue encoding vector of the slurry coating state image feature , through its residual mechanism based on different modal mapping, a preliminary structured semantic alignment of common explicit associations is performed in the shared representation space to obtain the slurry coating state-slurry attribute semantic alignment vector and slurry attribute-slurry coating state semantic alignment vector Then, based on the global semantic spatialization representation of the graph structure under explicit modeling of mapping preference, the residual mechanism is further used to capture the "missing" information in the alignment process under different mapping modalities, and adaptive compensation is performed through explicit association to further capture the fine-grained significant difference information between modalities in the initial alignment process. In this way, the contradiction between the alignment complexity of the intrinsic driving mapping of cross-modal differences and the intuitiveness of the explicit association of the graph structure can be optimized, and the illusion of global semantic constraints between modalities under the condition of misalignment of fine-grained relationships between local salient regions can be reduced.

[0038] In an embodiment of the present application, the step S1332, based on the slurry coating state-slurry attribute core clue weaving template matrix, performs cross-modal fine-grained interaction on the slurry coating state image coding feature map and the slurry attribute text semantic coding vector to obtain the slurry coating state-slurry attribute cross-modal significant interaction coding vector, including: S1332-1, performing feature decoupling along the channel dimension on the slurry coating state image coding feature map to obtain a set of local feature matrices of the slurry coating state image; S1332-2, using the slurry attribute text semantic coding vector as a query A query vector is obtained, each local feature matrix of the slurry coating state image in the set of local feature matrices of the slurry coating state image is used as a key matrix and the slurry coating state-slurry attribute core clue weaving template matrix is ​​used as a priori information constraint matrix, and cross-modal heterogeneous transformation constraint encoding is performed on it to obtain a set of slurry coating state-slurry attribute cross-modal fine-grained interaction encoding vectors; S1332-3, calculating the positional mean vector of the set of the slurry coating state-slurry attribute cross-modal fine-grained interaction encoding vectors to obtain the slurry coating state-slurry attribute cross-modal significant interaction encoding vector.

[0039] Specifically, in the step S1332-1, the feature decoupling along the channel dimension is performed on the slurry coating state image coding feature map to obtain a set of local feature matrices of the slurry coating state image, which is expressed as follows using the feature decoupling formula: ;in, is the feature decoupling operation along the channel dimension, and is the first, second, and third local feature matrix of the slurry coating state image. and The local feature matrix of the slurry coating state image. It should be understood that the core idea of ​​feature decoupling is to decompose the complex high-dimensional feature map into multiple low-dimensional local feature matrices, so as to achieve refined processing of different channel information in the image encoding feature map. This local separation can not only effectively reduce the redundancy of the data, but also better capture the subtle differences in complex content between modalities. In this way, this feature-level local separation provides a unitized modeling input for the next stage of cross-modal query-key interaction, ensuring a more refined capture of complex content between modalities. Specifically, the feature decoupling operation along the channel dimension aims to reveal the semantic information hidden inside the slurry coating state image encoding feature map by independently processing different channels. Among them, each channel can be regarded as a specific feature representation, such as color, texture or edge information. By separating and reorganizing these channels into local feature matrices, the model can more clearly understand the role of each feature in the overall image.

[0040] Specifically, in step S1332-2, the slurry characteristic text semantic coding vector is used as the query vector, each slurry coating state image local feature matrix in the set of the slurry coating state image local feature matrix is ​​used as the key matrix, and the slurry coating state-slurry attribute core clue weaving template matrix is ​​used as the prior information constraint matrix, and cross-modal heterogeneous conversion constraint coding is performed on them to obtain a set of slurry coating state-slurry attribute cross-modal fine-grained interactive coding vectors, which is expressed as a cross-modal heterogeneous conversion constraint coding formula: ;in, yes The scale is the width of the matrix multiplied by the height of the matrix, is the normalization function, is the first in the set of cross-modal fine-grained interaction encoding vectors of slurry coating state and slurry attributes. A slurry coating state-slurry attribute cross-modal fine-grained interaction encoding vector. It should be understood that in this process, the slurry property text semantic encoding vector is used as a query vector (Query Vector), and the response is obtained from the local feature matrix set of the slurry coating state image through the attention mechanism. These response sets are also subject to the prior constraints of the core clue weaving template matrix between modalities. This constraint mechanism further improves the representation ability of the results by strengthening the standardization and accuracy in modal interaction. The main difference between heterogeneous transformation and native Transformer is the introduction of the core clue weaving template matrix between modalities, which makes the dynamic weight distribution or attention parameters between modalities subject to its prior constraints so that the feature interaction revolves around the template information woven by the core clues, thereby ensuring the high coupling and interaction directionality between modal representations at the algorithm level.

[0041] Specifically, in step S1332-3, the position mean vector of the set of the slurry coating state-slurry attribute cross-modal fine-grained interaction coding vectors is calculated to obtain the slurry coating state-slurry attribute cross-modal significant interaction coding vector, which is expressed as the position mean calculation formula: ;in, is the number of vectors in the set of the slurry coating state-slurry attribute cross-modal fine-grained interaction encoding vectors, It is the slurry coating state-slurry attribute cross-modal significant interaction coding vector. It should be understood that the set of slurry coating state-slurry attribute cross-modal fine-grained interaction coding vectors is generated as the modal interaction is completed, and each slurry coating state-slurry attribute cross-modal fine-grained interaction coding vector in the set reflects the combined representation of a certain area in the local feature matrix of the slurry coating state image and the slurry characteristic text semantic coding vector feature. Then, the position mean vector of the set of the slurry coating state-slurry attribute cross-modal fine-grained interaction coding vector is calculated to obtain the slurry coating state-slurry attribute cross-modal significant interaction coding vector. Here, in order to further compress the processing cost and retain the main semantic information, these slurry coating state-slurry attribute cross-modal fine-grained interaction coding vectors are condensed by position mean calculation to form the final slurry coating state-slurry attribute cross-modal significant interaction coding vector.

[0042] In the above-mentioned automatic control method of the coating machine, the step S140, based on the cross-modal significant interaction feature of the slurry coating state-slurry attribute, obtains the detection surface density corresponding to the slurry in the target detection area, including: inputting the cross-modal significant interaction coding vector of the slurry coating state-slurry attribute into the detection surface density estimator based on the decoder to obtain the detection surface density corresponding to the slurry in the target detection area. That is, the cross-domain significant interaction coding vector of the slurry coating state-slurry attribute slurry coating state-slurry attribute obtained by cross-domain combination of core clues using the slurry coating state image coding feature map and the slurry characteristic text semantic coding vector is decoded to automatically generate the detection surface density corresponding to the slurry in the target detection area. In particular, the decoder structure has unique advantages in processing the conversion task from abstract features to specific outputs. It can gradually parse and reconstruct the input coding vector in a layer-by-layer progressive manner according to the patterns and rules learned during the training process, and mine the key information related to the surface density. For example, during the training process, the decoder can learn the nonlinear relationship between the different dimensional features and the surface density in the encoding vector, so as to accurately map the encoding vector to the corresponding surface density value. Specifically, the output result of the decoder is a set of continuous or discrete values, which is used to represent the surface density distribution of the slurry coating in the target detection area. For example, in a typical embodiment, the decoder outputs a two-dimensional matrix, in which each element corresponds to a local position in the target detection area and contains the surface density value at that position, which is further used to evaluate the overall uniformity of the coating, the consistency of thickness, and the presence or absence of potential defects.

[0043] In the above-mentioned automatic control method of the coater, the step S150 adjusts the coater based on the detection surface density and the target surface density corresponding to the slurry in the target detection area. It should be understood that the difference between the detection surface density and the target surface density intuitively reflects the deviation of the current coater in the execution of the coating process. For example, changes in factors such as slurry flow, coating speed, and coating pressure will affect the final surface density. By adjusting the coater based on the detection surface density and the target surface density corresponding to the slurry in the target detection area, the various process parameters of the coater can be adjusted in a targeted manner, such as appropriately increasing or decreasing the slurry flow, adjusting the coating speed, changing the coating pressure, etc., so that the subsequent coating process is closer to the ideal state, optimizing the overall coating process, and improving production efficiency.

[0044] Specifically, based on the detection surface density and target surface density corresponding to the slurry in the target detection area, the coating machine is adjusted, including the following steps: First, using the coating control model trained by deep learning, based on historical coating data, the detection surface density and target surface density are input into this model to determine the best coating setting information, covering adjustment suggestions for multiple key parameters such as motor position, pump speed, and gap. Among them, this model adopts an advanced neural network architecture, combining convolutional neural networks (CNN) for extracting image features (such as coating thickness distribution) and recurrent neural networks (RNN) for capturing time series data (such as coating speed changes). In this way, the model can fully understand the complex laws in the coating process and provide reliable guidance for subsequent adjustments. The core of this process is to predict how to adjust equipment parameters (such as motor position, pump speed, gap, etc.) to achieve the ideal coating effect through the model.

[0045] Next, the parameters such as motor position, pump speed, and gap are adjusted through simulation, and the simulation is iterated continuously until the simulation surface density meets the target requirements. This is to verify whether the settings recommended by the model can effectively achieve the target surface density, and further optimize on this basis. Among them, simulation adjustment usually relies on virtual simulation technology to build a digital twin model of the coating process, so as to complete parameter testing without interfering with actual production. Specifically, this stage adopts a real-time feedback mechanism to ensure that each adjustment result can be quickly reflected in the system and speed up the optimization process. According to the parameter adjustment suggestions output by the initial model, the system will gradually change the position of the motor-driven coating head, adjust the speed of the slurry delivery pump, and adjust the gap between the coating head and the substrate. After each adjustment, the new simulation surface density will be recalculated and compared with the target value until the difference between the two meets the preset threshold. This iterative optimization method not only improves the adjustment accuracy, but also significantly reduces the trial and error cost, laying a solid foundation for subsequent actual production.

[0046] Finally, the coating machine settings are automatically adjusted according to the actual production situation to ensure the expected coating effect. This step is designed to enable the system to continuously optimize its performance in a dynamically changing environment and ensure the continuity and stability of production. To this end, the system collects various parameters in the coating process (such as slurry flow, coating speed, coating pressure, etc.) in real time through a sensor network and compares and analyzes them with the detection surface density. Once the deviation is found to be beyond the allowable range, the system will immediately trigger the automatic adjustment mechanism. For example, when the coating thickness fluctuates, the system will automatically adjust the coating speed or slurry flow to restore the stable state; if the abnormal situation is caused by equipment failure or raw material quality fluctuations, the system will also start the emergency plan to ensure that the production process is not significantly affected. In addition, the automatic adjustment function relies on an adaptive control system, which integrates a variety of intelligent algorithms such as fuzzy control and genetic algorithms to further enhance the flexibility and robustness of the system. With the support of these algorithms, the system can better cope with complex and changing production environments and maximize production efficiency while ensuring coating quality.

[0047] These three steps together constitute a complete closed-loop control system, which ensures the efficiency and accuracy of the coating process from the application of theoretical models to continuous optimization in actual operations. In the entire process, the predictive ability of the deep learning model, the optimization means of virtual simulation, and the real-time response of the adaptive control system cooperate with each other to form an organic whole. In this way, not only the production efficiency is improved, but also the uncertainty caused by human intervention is reduced, and the consistency of product quality is improved.

[0048] In summary, the automatic control method of the coating machine based on the embodiment of the present application is explained, which first obtains the slurry coating state image and the slurry characteristic text description of the target detection area, and then uses machine learning and natural language analysis technology to perform image encoding on the slurry coating state image, and semantic encoding on the slurry characteristic text description, and generates the detection surface density corresponding to the slurry in the target detection area based on the cross-modal significant interaction of the core clues of the two, so as to adjust the coating machine in combination with the target surface density corresponding to the slurry. In this way, by fusing multimodal data, it is possible to reduce the dependence on environmental factors of a single sensor or parameter estimation method, better understand and process complex coating processes, and make adaptive adjustments according to the characteristics of different slurries, with high flexibility.

[0049] Figure 6 1 is a system block diagram of an automatic control system of a coating machine according to an embodiment of the present application. Figure 6As shown, according to the automatic control system 100 of the coating machine of the embodiment of the present application, it includes: a slurry coating state image acquisition module 110, which is used to acquire the slurry coating state image of the target detection area collected by the camera when the coating machine performs a slurry coating operation; a slurry characteristic text description module 120, which is used to extract the characteristic text description of the slurry from the slurry characteristic database; a slurry coating state-slurry attribute cross-modal significant interaction module 130, which is used to obtain the slurry coating state-slurry attribute cross-modal significant interaction feature based on the slurry coating state image and the characteristic text description of the slurry; a slurry detection surface density acquisition module 140, which is used to obtain the detection surface density corresponding to the slurry in the target detection area based on the slurry coating state-slurry attribute cross-modal significant interaction feature; a coating machine adjustment module 150, which is used to adjust the coating machine based on the detection surface density corresponding to the slurry in the target detection area and the target surface density.

[0050] Here, those skilled in the art will appreciate that the specific operations of each step in the above-mentioned coating machine automation control system have been described in the above reference. Figures 1 to 5 The automatic control method of the coating machine has been described in detail, and therefore, its repeated description will be omitted.

[0051] As described above, the automatic control system 100 of the coating machine according to the embodiment of the present application can be implemented in various terminal devices. In one example, the automatic control system 100 of the coating machine can be integrated into the terminal device as a software module and / or a hardware module. For example, the automatic control system 100 of the coating machine can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the automatic control system 100 of the coating machine can also be one of the many hardware modules of the terminal device.

[0052] Alternatively, in another example, the coating machine's automated control system 100 and the terminal device may be separate devices, and the coating machine's automated control system 100 may be connected to the terminal device via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0053] In summary, the automatic control system of the coating machine based on the embodiment of the present application is explained, which first obtains the slurry coating state image and the slurry characteristic text description of the target detection area, and then uses machine learning and natural language analysis technology to perform image encoding on the slurry coating state image, and semantic encoding on the slurry characteristic text description, and generates the detection surface density corresponding to the slurry in the target detection area based on the cross-modal significant interaction of the core clues of the two, so as to adjust the coating machine in combination with the target surface density corresponding to the slurry. In this way, by fusing multimodal data, it is possible to reduce the dependence on environmental factors of a single sensor or parameter estimation method, better understand and process complex coating processes, and make adaptive adjustments according to the characteristics of different slurries, with high flexibility.

[0054] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0055] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0056] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0057] It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0058] Therefore, no matter from which point of view, the embodiments should be regarded as illustrative and non-restrictive, and the scope of the present invention is limited by the appended claims rather than the above description, so it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any attached figure mark in the claims should not be regarded as limiting the claims involved.

[0059] In addition, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices stated in the system claim can also be implemented by one unit or device through software or hardware. The second and other words are used to indicate names, but not to indicate any particular order.

[0060] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit of the technical solution of the present invention.

Claims

1. An automatic control method for a coating machine, characterized in that: include: When the coating machine performs a slurry coating operation, obtaining a slurry coating state image of a target detection area captured by a camera; Extracting a text description of characteristics of the slurry from a slurry characteristics database; Based on the slurry coating state image and the characteristic text description of the slurry, a cross-modal significant interaction feature of the slurry coating state-slurry attribute is obtained, including: encoding the slurry coating state image and the characteristic text description of the slurry and cross-domain combining the slurry coating state-slurry attribute to obtain the cross-modal significant interaction feature of the slurry coating state-slurry attribute; based on the cross-modal significant interaction feature of the slurry coating state-slurry attribute, a detection surface density corresponding to the slurry in the target detection area is obtained; based on the detection surface density corresponding to the slurry in the target detection area and the target surface density, the coater is adjusted.

2. The automatic control method of the coating machine according to claim 1, characterized in that: The slurry coating state image and the characteristic text description of the slurry are encoded and the slurry coating state-slurry attribute are cross-domain combined to obtain the slurry coating state-slurry attribute cross-modal significant interaction features, including: extracting the slurry coating state features from the slurry coating state image to obtain a slurry coating state image encoding feature map; semantically encoding the characteristic text description of the slurry to obtain a slurry characteristic text semantic encoding vector; cross-domain joint encoding of the slurry coating state image encoding feature map and the slurry characteristic text semantic encoding vector of the slurry coating state-slurry attribute core clues to obtain a slurry coating state-slurry attribute cross-modal significant interaction encoding vector as the slurry coating state-slurry attribute cross-modal significant interaction feature.

3. The automatic control method of the coating machine according to claim 2, characterized in that: Extracting slurry coating state features from the slurry coating state image to obtain a slurry coating state image encoding feature map includes: using a slurry coating state feature extractor based on DenseNet to perform feature extraction on the slurry coating state image to obtain the slurry coating state image encoding feature map.

4. The automatic control method of the coating machine according to claim 3, characterized in that: The method comprises: semantically encoding the text description of the characteristics of the slurry to obtain the semantic encoding vector of the slurry characteristics text, including: using a semantic encoder based on a bidirectional recurrent neural network model to semantically encode the text description of the characteristics of the slurry to obtain the semantic encoding vector of the slurry characteristics text.

5. The automatic control method of the coating machine according to claim 4, characterized in that: The slurry coating state image coding feature map and the slurry property text semantic coding vector are cross-domain jointly encoded with the slurry coating state-slurry property core clues to obtain the slurry coating state-slurry property cross-modal significant interaction coding vector, including: calculating the slurry coating state-slurry property core clue weaving template matrix between the slurry coating state image coding feature map and the slurry property text semantic coding vector; based on the slurry coating state-slurry property core clue weaving template matrix, the slurry coating state image coding feature map and the slurry property text semantic coding vector are cross-modally fine-grained interacted to obtain the slurry coating state-slurry property cross-modal significant interaction coding vector.

6. The automatic control method of the coating machine according to claim 5, characterized in that: The method comprises: performing core clue point convolution extraction on the slurry property text semantic coding vector to obtain the slurry property text semantic core clue coding vector; extracting the slurry coating state image feature core clue coding vector from the slurry coating state image coding feature map; and constructing the slurry coating state-slurry property core clue weaving template matrix between the slurry property text semantic core clue coding vector and the slurry coating state image feature core clue coding vector.

7. The automatic control method of the coating machine according to claim 6, characterized in that: The method comprises: constructing the slurry coating state-slurry attribute core clue weaving template matrix between the slurry characteristic text semantic core clue coding vector and the slurry coating state image feature core clue coding vector, comprising: using a feature mapping function to process the slurry characteristic text semantic core clue coding vector and the slurry coating state image feature core clue coding vector to obtain a slurry coating state-slurry attribute core association matrix; performing a dissimilarity model on the slurry characteristic text semantic core clue coding vector and the slurry coating state image feature core clue coding vector; The state mapping residuals are semantically aligned to obtain the slurry coating state-slurry attribute semantic alignment vector and the slurry attribute-slurry coating state semantic alignment vector; the slurry coating state-slurry attribute significance difference matrix between the slurry coating state-slurry attribute semantic alignment vector and the slurry attribute-slurry coating state semantic alignment vector is calculated; based on the slurry coating state-slurry attribute significance difference matrix, the slurry coating state-slurry attribute core association matrix is ​​fine-grained significance compensation is performed to obtain the slurry coating state-slurry attribute core clue weaving template matrix.

8. The automatic control method of the coating machine according to claim 7, characterized in that: Based on the slurry coating state-slurry attribute core clue weaving template matrix, the slurry coating state image coding feature map and the slurry property text semantic coding vector are cross-modally fine-grained interactively performed to obtain the slurry coating state-slurry attribute cross-modal significant interactive coding vector, including: performing feature decoupling along the channel dimension on the slurry coating state image coding feature map to obtain a set of slurry coating state image local feature matrices; using the slurry property text semantic coding vector as a query vector, using each slurry coating state image local feature matrix in the set of slurry coating state image local feature matrices as a key matrix, and using the slurry coating state-slurry attribute core clue weaving template matrix as a priori information constraint matrix, performing cross-modal heterogeneous transformation constraint encoding on them to obtain a set of slurry coating state-slurry attribute cross-modal fine-grained interactive coding vectors; calculating the positional mean vector of the set of slurry coating state-slurry attribute cross-modal fine-grained interactive coding vectors to obtain the slurry coating state-slurry attribute cross-modal significant interactive coding vector.

9. The automatic control method of the coating machine according to claim 8, characterized in that: Based on the cross-modal significant interaction characteristics of the slurry coating state and the slurry properties, the detection surface density corresponding to the slurry in the target detection area is obtained, including: inputting the slurry coating state and the slurry properties cross-modal significant interaction encoding vector into a decoder-based detection surface density estimator to obtain the detection surface density corresponding to the slurry in the target detection area.

10. An automatic control system for a coating machine, characterized in that: include: A slurry coating state image acquisition module is used to acquire a slurry coating state image of a target detection area captured by a camera when the coating machine performs a slurry coating operation; a slurry property text description module is used to extract a property text description of the slurry from a slurry property database; a slurry coating state-slurry property cross-modal significant interaction module is used to obtain a slurry coating state-slurry property cross-modal significant interaction feature based on the slurry coating state image and the property text description of the slurry; A slurry detection surface density acquisition module is used to obtain the detection surface density corresponding to the slurry in the target detection area based on the cross-modal significant interaction characteristics of the slurry coating state-slurry properties; a coating machine adjustment module is used to adjust the coating machine based on the detection surface density corresponding to the slurry in the target detection area and the target surface density.

Images