Inorganic plate quality classification system and method

Abstract

The invention discloses an inorganic plate quality classification system and method, and relates to the field of intelligent detection, and the method comprises the steps: firstly obtaining the surface image and depth information of a to-be-detected inorganic plate, calculating the basic score of the structural strength of the inorganic plate through the information, calculating the secondary score of the surface flatness of the inorganic plate according to the surface depth information, and finally, carrying out the quality classification of the inorganic plate. And carrying out quality classification on the to-be-detected inorganic plate by combining the structural strength basic score and the surface flatness secondary score. In this way, more reliable inorganic plate quality control can be achieved.

Description

Technical Field

[0001] The present application relates to the field of intelligent detection, and more specifically, to an inorganic panel quality classification system and method. Background Art

[0002] Inorganic panels usually refer to panels made of inorganic materials, such as cement boards, fiberglass boards, asbestos boards, etc. These materials are widely used in construction, decoration, industrial manufacturing and other fields due to their excellent physical and chemical properties, such as high temperature resistance and fire resistance. The quality of inorganic panels directly affects the safety and service life of the product, so strict quality control is essential.

[0003] Patent CN118245855B discloses an artificial intelligence-based building board quality classification method, which uses a machine vision unit to perform basic scoring, secondary scoring, and comprehensive scoring on the slight deformation of building boards, and uses a human terminal to manually screen to determine the smoothness of the building board surface. Subsequently, by comparing the comprehensive score of the human and the comprehensive score of the machine, the scoring weight is adjusted, and finally a comprehensive score is obtained, and the building boards are divided into three grades: excellent, good, and poor.

[0004] When the above-mentioned patents perform basic scoring on the boards, they mainly evaluate the structural strength by combining the image and depth information to calculate the size and surface defect location information of the building boards. However, this method mainly focuses on explicit features such as surface defects, while ignoring other potential and complex features. For example, subtle features such as surface texture, color changes, and tiny cracks have an important impact on structural strength, but traditional methods have difficulty in effectively extracting and quantifying these features, resulting in incomplete and inaccurate evaluation results. In addition, there is a contextual interaction between the surface image and depth information of the board, that is, there is a mutual influence between the basic image information such as surface texture and color changes and the depth information. For example, tiny cracks may not be obvious in the image, but the depth information can reveal their existence. Traditional scoring methods have difficulty capturing this interaction, which further affects the comprehensiveness and accuracy of the evaluation results.

[0005] Therefore, an optimized quality classification scheme for inorganic panels is desired. Summary of the invention

[0006] In view of the shortcomings of the prior art, the present application provides an inorganic plate quality classification system and method.

[0007] According to one aspect of the present application, a method for quality classification of inorganic panels is provided, comprising: Acquire surface image information and surface depth information of the inorganic plate to be inspected; Based on the surface image information and surface depth information of the inorganic board to be detected, a basic score of the structural strength of the inorganic board to be detected is obtained, including: performing semantic feature extraction on the surface image information and surface depth information of the inorganic board to be detected respectively to obtain the semantic coding features of the surface state of the inorganic board and the depth semantic coding features of the structural morphology of the inorganic board; performing cross-domain fine-grained semantic joint interaction on the semantic coding features of the surface state of the inorganic board and the depth semantic coding features of the structural morphology of the inorganic board to obtain the image-depth fine-grained joint semantic significant fusion coding features of the inorganic board; based on the inorganic board image-depth fine-grained joint semantic significant fusion coding features, the basic score of the structural strength is obtained; Based on the surface depth information of the inorganic plate to be detected, obtaining a secondary score of the surface flatness of the inorganic plate to be detected; The inorganic plate to be inspected is quality classified based on the basic structural strength score and the secondary surface flatness score to obtain a quality classification result.

[0008] In the above-mentioned inorganic panel quality classification method, semantic feature extraction is performed on the surface image information and surface depth information of the inorganic panel to be inspected respectively to obtain the inorganic panel surface state semantic coding features and the inorganic panel structural morphology depth semantic coding features, including: inputting the surface image information of the inorganic panel to be inspected into an inorganic panel surface state feature extractor based on a void convolutional neural network model to obtain an inorganic panel surface state semantic coding feature map as the inorganic panel surface state semantic coding features; inputting the surface depth information of the inorganic panel to be inspected into an inorganic panel surface structure morphology feature extractor based on a depth-separable convolutional neural model to obtain an inorganic panel structural morphology depth semantic coding feature map as the inorganic panel structural morphology depth semantic coding features.

[0009] In the above-mentioned inorganic panel quality classification method, the inorganic panel surface state semantic coding features and the inorganic panel structural morphology depth semantic coding features are subjected to cross-domain fine-grained semantic joint interaction to obtain inorganic panel image-depth fine-grained joint semantic significant fusion coding features, including: feature fine-grained decoupling of the inorganic panel surface state semantic coding feature map and the inorganic panel structural morphology depth semantic coding feature map to obtain a set of inorganic panel surface state semantic local fine-grained feature vectors and a set of inorganic panel structural morphology depth semantic local fine-grained feature vectors; cross-domain fine-grained feature significant fusion of the set of inorganic panel surface state semantic local fine-grained feature vectors and the set of inorganic panel structural morphology depth semantic local fine-grained feature vectors is performed to obtain the inorganic panel image-depth fine-grained joint semantic significant fusion coding features.

[0010] In the above-mentioned inorganic panel quality classification method, the inorganic panel surface state semantic coding feature map and the inorganic panel structural morphology depth semantic coding feature map are feature fine-grained decoupling to obtain a set of inorganic panel surface state semantic local fine-grained feature vectors and a set of inorganic panel structural morphology depth semantic local fine-grained feature vectors, including: expanding each feature matrix of the inorganic panel surface state semantic coding feature map along the channel dimension into a feature vector to obtain the set of inorganic panel surface state semantic local fine-grained feature vectors; expanding each feature matrix of the inorganic panel structural morphology depth semantic coding feature map along the channel dimension into a feature vector to obtain the set of inorganic panel structural morphology depth semantic local fine-grained feature vectors.

[0011] In the above-mentioned inorganic panel quality classification method, a cross-domain fine-grained feature significant fusion is performed on the set of inorganic panel surface state semantic local fine-grained feature vectors and the set of inorganic panel structural morphology depth semantic local fine-grained feature vectors to obtain the inorganic panel image-depth fine-grained joint semantic significant fusion coding features, including: each group of corresponding inorganic panel surface state semantic local fine-grained feature vectors in the set of inorganic panel surface state semantic local fine-grained feature vectors and the set of inorganic panel structural morphology depth semantic local fine-grained feature vectors are fused. The depth feature vector and the local fine-grained feature vector of the inorganic plate structural morphology depth semantics are input into the fine-grained common feature extraction network to obtain a set of inorganic plate surface state-structural morphology joint fine-grained core feature vectors; the set of inorganic plate surface state-structural morphology joint fine-grained core feature vectors is input into the clustering network to obtain the inorganic plate surface state-structural morphology joint fine-grained core feature semantic clustering center vector; based on the inorganic plate surface state-structural morphology joint fine-grained core feature vectors, each inorganic plate surface state-structural morphology in the set of inorganic plate surface state-structural morphology fine-grained core feature vectors is clustered. The inorganic board surface state-structure morphology joint fine-grained core feature vector is used to construct an inorganic board surface state-structure morphology query vector and an inorganic board surface state-structure morphology value vector, and the inorganic board surface state-structure morphology key vector is constructed based on the inorganic board surface state-structure morphology joint fine-grained core feature semantic clustering center vector, and the inorganic board surface state-structure morphology query vector, the inorganic board surface state-structure morphology value vector and the inorganic board surface state-structure morphology of each inorganic board surface state-structure morphology joint fine-grained core feature vector are combined. The appearance key vector is input into a global semantic interaction encoder based on a converter structure to obtain a set of inorganic panel surface state-structural morphology context joint fine-grained core feature vectors; each inorganic panel surface state-structural morphology context joint fine-grained core feature vector in the set of inorganic panel surface state-structural morphology context joint fine-grained core feature vectors is feature-reshaped to obtain an inorganic panel image-depth fine-grained joint semantically significant fusion coding feature map as the inorganic panel image-depth fine-grained joint semantically significant fusion coding feature.

[0012] In the above-mentioned inorganic panel quality classification method, each group of corresponding inorganic panel surface state semantic local fine-grained feature vectors and inorganic panel structural morphology depth semantic local fine-grained feature vectors in the set of inorganic panel surface state semantic local fine-grained feature vectors and the set of inorganic panel structural morphology depth semantic local fine-grained feature vectors are input into a fine-grained common feature extraction network to obtain a set of inorganic panel surface state-structural morphology joint fine-grained core feature vectors, including: respectively calculating the position point addition, position point multiplication and position point subtraction between the inorganic panel surface state semantic local fine-grained feature vectors and the inorganic panel structural morphology depth semantic local fine-grained feature vectors to obtain the inorganic panel surface state-structural morphology point addition vector, the inorganic panel surface state-structural morphology point addition vector, the inorganic panel surface state-structural morphology point addition vector and the inorganic panel surface state-structural morphology point addition vector. The inorganic board surface state-structural morphology point product vector and the inorganic board surface state-structural morphology point subtraction vector are cascaded to obtain the inorganic board surface state-structural morphology cascade vector; the weight matrix and the inorganic board surface state-structural morphology cascade vector are matrix-vector multiplied, and the obtained inorganic board surface state-structural morphology weight vector and the bias vector are added by position to obtain the inorganic board surface state-structural morphology joint vector; the inorganic board surface state-structural morphology joint vector is processed using the tanh function to obtain the inorganic board surface state-structural morphology joint fine-grained core feature vector.

[0013] In the above-mentioned inorganic panel quality classification method, the set of inorganic panel surface state-structural morphology joint fine-grained core feature vectors is input into the clustering network to obtain the inorganic panel surface state-structural morphology joint fine-grained core feature semantic clustering center vector, including: calculating the positional mean vector of the set of inorganic panel surface state-structural morphology joint fine-grained core feature vectors to obtain the inorganic panel surface state-structural morphology joint fine-grained core feature semantic clustering center vector.

[0014] In the above-mentioned inorganic panel quality classification method, an inorganic panel surface state-structural morphology query vector and an inorganic panel surface state-structural morphology value vector are constructed based on each inorganic panel surface state-structural morphology joint fine-grained core feature vector in the set of the inorganic panel surface state-structural morphology joint fine-grained core feature vectors, and an inorganic panel surface state-structural morphology key vector is constructed based on the inorganic panel surface state-structural morphology joint fine-grained core feature semantic clustering center vector, and the inorganic panel surface state-structural morphology query vector, the inorganic panel surface state-structural morphology value vector and the inorganic panel surface state-structural morphology key vector of each inorganic panel surface state-structural morphology joint fine-grained core feature vector are input into a global semantic interactive encoder based on a converter structure to obtain a set of inorganic panel surface state-structural morphology context joint fine-grained core feature vectors. The method comprises: multiplying a query embedding matrix with the inorganic board surface state-structural morphology joint fine-grained core feature vector to obtain the inorganic board surface state-structural morphology query vector; multiplying a value embedding matrix with the inorganic board surface state-structural morphology joint fine-grained core feature vector to obtain the inorganic board surface state-structural morphology value vector; multiplying a key embedding matrix with the inorganic board surface state-structural morphology joint fine-grained core feature semantic clustering center vector to obtain the inorganic board surface state-structural morphology key vector; inputting the inorganic board surface state-structural morphology query vector, the inorganic board surface state-structural morphology value vector and the inorganic board surface state-structural morphology key vector into the converter structure-based global semantic interactive encoder to obtain the inorganic board surface state-structural morphology context joint fine-grained core feature vector.

[0015] In the above-mentioned inorganic panel quality classification method, the basic structural strength score is obtained based on the inorganic panel image-depth fine-grained joint semantically significant fusion coding features, including: inputting the inorganic panel image-depth fine-grained joint semantically significant fusion coding feature map into a decoder-based basic scoring module to obtain the basic structural strength score.

[0016] According to another aspect of the present application, there is provided an inorganic plate quality classification system, comprising: The inorganic board surface information acquisition module is used to obtain the surface image information and surface depth information of the inorganic board to be detected; An inorganic panel structural strength basic score calculation module is used to obtain the structural strength basic score of the inorganic panel to be detected based on the surface image information and surface depth information of the inorganic panel to be detected, wherein the inorganic panel structural strength basic score calculation module includes: an inorganic panel surface information semantic feature extraction unit, which is used to perform semantic feature extraction on the surface image information and surface depth information of the inorganic panel to be detected respectively to obtain the inorganic panel surface state semantic coding feature and the inorganic panel structural morphology depth semantic coding feature; an inorganic panel image-depth fine-grained joint semantically significant fusion coding feature generation unit, which is used to perform cross-domain fine-grained semantic joint interaction on the inorganic panel surface state semantic coding feature and the inorganic panel structural morphology depth semantic coding feature to obtain the inorganic panel image-depth fine-grained joint semantically significant fusion coding feature; a structural strength basic score generation unit, which is used to obtain the structural strength basic score based on the inorganic panel image-depth fine-grained joint semantically significant fusion coding feature; An inorganic plate surface flatness secondary score calculation module, used to obtain a surface flatness secondary score of the inorganic plate to be detected based on the surface depth information of the inorganic plate to be detected; The inorganic plate quality classification result generating module is used to perform quality classification on the inorganic plate to be inspected based on the basic structural strength score and the secondary surface flatness score to obtain a quality classification result.

[0017] This application has significant technical effects due to the adoption of the above technical solutions: The inorganic panel quality classification system and method provided in the present application adopts artificial intelligence-based image analysis and information processing technology to extract surface state semantic features from the surface image information of the inorganic panel to be inspected, and extract surface structural morphology features from the surface depth information, thereby automatically obtaining the basic structural strength score based on the cross-domain fine-grained interaction features between the semantic coding features of the surface state of the inorganic panel and the semantic coding features of the depth of the structural morphology of the inorganic panel. In this way, the subtle differences and complex features of the surface state and structural morphology of the inorganic panel can be accurately identified and analyzed, avoiding the defects of the inorganic panel quality inspection classification method in the existing patents. At the same time, the mutual influence between the surface image information and the surface depth information can be perceived, which helps to enhance the comprehensiveness of the evaluation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] 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.

[0019] Figure 1 Flow chart of a method for classifying inorganic panel quality according to an embodiment of the present application.

[0020] Figure 2 The present invention is a flow chart of obtaining a basic score of the structural strength of the inorganic board to be inspected based on the surface image information and surface depth information of the inorganic board to be inspected in the inorganic board quality classification method according to an embodiment of the present application.

[0021] Figure 3 This is a schematic diagram of data flow for obtaining a basic score of structural strength of the inorganic panel to be inspected based on surface image information and surface depth information of the inorganic panel to be inspected in a quality classification method of the embodiment of the present application.

[0022] Figure 4 4 is a system block diagram of an inorganic panel quality classification system according to an embodiment of the present application. DETAILED DESCRIPTION

[0023] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0024] Inorganic panels, such as cement panels and fiberglass panels, are widely used in many fields due to their high temperature resistance and fire resistance. The quality of inorganic panels is directly related to the safety and service life of the product, so strict quality control is essential.

[0025] Patent CN118245855B proposes an artificial intelligence-based building board quality classification method, which uses machine vision units to perform basic scoring, secondary scoring, and comprehensive scoring on the slight deformation of building boards, and combines manual screening to evaluate the smoothness of the surface of building boards. During the scoring process, the weights are adjusted by comparing the comprehensive scores of manual and machine scores, and finally the building boards are divided into three grades: excellent, good, and poor.

[0026] In the above patent, the structural strength of building boards is evaluated mainly based on image and depth information when conducting basic scoring. However, this method focuses on explicit features such as size and surface defects, while ignoring potential complex features such as surface texture, color changes, and tiny cracks that have an important impact on structural strength. These subtle features are difficult to effectively extract and quantify by traditional methods, resulting in incomplete and inaccurate evaluation results. In addition, there is a contextual interaction between the surface image and depth information of the board, which has an important impact on the comprehensiveness and accuracy of the evaluation results, but it is difficult for traditional scoring methods to capture this interaction.

[0027] Based on this, the present application proposes a quality classification method for inorganic panels. Figure 1 FIG. 1 is a flow chart of a method for classifying inorganic plate quality according to an embodiment of the present application. Figure 1 As shown, the inorganic panel quality classification method according to the embodiment of the present application includes: S110, acquiring surface image information and surface depth information of the inorganic panel to be inspected; S120, obtaining a basic structural strength score of the inorganic panel to be inspected based on the surface image information and surface depth information of the inorganic panel to be inspected; S130, obtaining a secondary surface flatness score of the inorganic panel to be inspected based on the surface depth information of the inorganic panel to be inspected; S140, performing quality classification on the inorganic panel to be inspected based on the basic structural strength score and the secondary surface flatness score to obtain a quality classification result.

[0028] In step S110, the surface image information and surface depth information of the inorganic board to be inspected are obtained. It should be understood that the surface image information of the inorganic board includes the surface texture information, color information, specific defect information, etc. of the inorganic board, and the surface depth information of the inorganic board specifically refers to the three-dimensional data information of the inorganic board (such as the depth map), which reflects the concave, convex and overall contour of the surface of the inorganic board. In general, the surface image information of the inorganic board provides the visual characteristics of the material, while the depth information provides the three-dimensional shape and structural characteristics. The quality of the inorganic board can be comprehensively evaluated by combining and analyzing these two types of information.

[0029] In step S120, a basic score of the structural strength of the inorganic board to be inspected is obtained based on the surface image information and the surface depth information of the inorganic board to be inspected.

[0030] Accordingly, in obtaining the basic score of the structural strength of the inorganic board to be detected based on the surface image information and surface depth information of the inorganic board to be detected, the technical concept of the present application is to use artificial intelligence-based image analysis and information processing technology to extract surface state semantic features from the surface image information of the inorganic board to be detected, and to extract surface structural morphology features from the surface depth information, so as to automatically obtain the basic score of structural strength based on the cross-domain fine-grained interaction features between the semantic coding features of the surface state of the inorganic board and the semantic coding features of the depth of the structural morphology of the inorganic board. In this way, the subtle differences and complex features of the surface state and structural morphology of the inorganic board can be accurately identified and analyzed, avoiding the defects of the inorganic board quality inspection classification method in the existing patents. At the same time, the mutual influence between the surface image information and the surface depth information can be perceived, which helps to enhance the comprehensiveness of the evaluation results.

[0031] Figure 2 The present invention is a flow chart of obtaining a basic score of the structural strength of the inorganic board to be inspected based on the surface image information and surface depth information of the inorganic board to be inspected in the inorganic board quality classification method according to an embodiment of the present application. Figure 3 The present invention is a data flow diagram for obtaining a basic score of the structural strength of the inorganic board to be inspected based on the surface image information and surface depth information of the inorganic board to be inspected in the inorganic board quality classification method according to the embodiment of the present application. Figure 2 and Figure 3 As shown, based on the surface image information and surface depth information of the inorganic panel to be detected, a basic score of the structural strength of the inorganic panel to be detected is obtained, including: S121, respectively performing semantic feature extraction on the surface image information and surface depth information of the inorganic panel to be detected to obtain the surface state semantic coding features of the inorganic panel and the depth semantic coding features of the inorganic panel structure; S122, performing cross-domain fine-grained semantic joint interaction on the surface state semantic coding features of the inorganic panel and the depth semantic coding features of the inorganic panel structure to obtain the image-depth fine-grained joint semantic significant fusion coding features of the inorganic panel; S123, obtaining the basic score of the structural strength based on the image-depth fine-grained joint semantic significant fusion coding features of the inorganic panel.

[0032] In step S121, semantic feature extraction is performed on the surface image information and surface depth information of the inorganic board to be detected, respectively, to obtain semantic coding features of the surface state of the inorganic board and semantic coding features of the depth of the structural morphology of the inorganic board. Specifically, in an embodiment of the present application, semantic feature extraction is performed on the surface image information and surface depth information of the inorganic board to be detected, respectively, to obtain semantic coding features of the surface state of the inorganic board and semantic coding features of the depth of the structural morphology of the inorganic board, including: inputting the surface image information of the inorganic board to be detected into an inorganic board surface state feature extractor based on a void convolutional neural network model to obtain a semantic coding feature map of the surface state of the inorganic board as the semantic coding feature of the surface state of the inorganic board; inputting the surface depth information of the inorganic board to be detected into an inorganic board surface structure morphology feature extractor based on a depth-separable convolutional neural model to obtain a semantic coding feature map of the depth of the structural morphology of the inorganic board as the depth semantic coding feature of the structural morphology of the inorganic board.

[0033] It should be understood that the surface image of the inorganic board to be inspected contains information about the surface state of the board, such as whether there are defects on the surface of the inorganic board, such as cracks, holes, stains, etc., as well as some texture information such as smoothness. Considering that the hole convolutional neural network model inserts holes between the elements of the convolution kernel, it can capture a wider range of contextual information while keeping the size of the convolution kernel unchanged, which helps to identify the complex surface features of the inorganic board. Based on this, in order to capture and mine the complex semantic information and detail features in the surface state of the inorganic board, in the technical solution of the present application, the surface image information of the inorganic board to be inspected is input into the inorganic board surface state feature extractor based on the hole convolutional neural network model to obtain the inorganic board surface state semantic coding feature map.

[0034] Accordingly, considering that the surface depth information of the inorganic board to be detected reflects the three-dimensional morphology of the surface of the inorganic board, including depth geometric features such as bumps and undulations, and this information can help identify subtle changes in the surface, such as subtle depressions, protrusions, etc. Therefore, in the technical solution of the present application, the surface depth information of the inorganic board to be detected is input into the inorganic board surface structure morphology feature extractor based on the deep separable convolutional neural model to refine and mine the structural semantic feature information of the inorganic board, and obtain the inorganic board structural morphology depth semantic coding feature map. That is, the deep separable convolutional neural model can effectively extract features from the depth information while reducing the amount of calculation to achieve efficient feature extraction through its unique convolutional coding method - the combination of deep convolution and point-by-point convolution, so as to obtain the inorganic board structural morphology depth semantic coding feature map containing rich inorganic board surface depth semantic information.

[0035] In step S122, the inorganic board surface state semantic coding features and the inorganic board structure morphology depth semantic coding features are cross-domain fine-grained semantic joint interaction to obtain inorganic board image-depth fine-grained joint semantic salient fusion coding features. It should be understood that the inorganic board surface state semantic coding feature map mainly contains information based on visual images, such as surface defects, textures, etc. The inorganic board structure morphology depth semantic coding feature map mainly contains structural features based on depth information, such as concave and convex, undulations, etc. At the same time, there is an associated interactive relationship between different regions between the two features. In order to integrate the information from both aspects, improve the understanding and analysis capabilities of the tiny defects and structural changes on the surface of the inorganic board, and form a richer and more comprehensive feature representation, in the technical solution of the present application, the inorganic board surface state semantic coding features and the inorganic board structure morphology depth semantic coding features are cross-domain fine-grained semantic joint interaction to obtain the inorganic board image-depth fine-grained joint semantic salient fusion coding feature map as the inorganic board image-depth fine-grained joint semantic salient fusion coding feature.

[0036] In detail, first, in order to capture the features of each local area more carefully to identify the tiny defects and structural changes on the surface of the inorganic board, the inorganic board surface state semantic encoding feature map and the inorganic board structural morphology deep semantic encoding feature map are fine-grained feature decoupling to more accurately separate the local state and subtle structural morphology of the inorganic board surface, and obtain a set of inorganic board surface state semantic local fine-grained feature vectors and a set of inorganic board structural morphology deep semantic local fine-grained feature vectors, which provide support for subsequent fine-grained cross-domain interactive fusion. Then, these two sets of sets are input into the fine-grained common feature extraction network to capture and extract the common and representative features between each set of corresponding inorganic board surface state local features and structural morphology, and obtain a set of inorganic board surface state-structural morphology joint fine-grained core feature vectors. The obtained joint fine-grained core feature vector not only covers the data of the original feature map, but also reveals the potential association between cross-domain data, thereby capturing the deep connection between cross-domain data. Then, the set of inorganic board surface state-structural morphology joint fine-grained core feature vectors is input into the clustering network to obtain the inorganic board surface state-structural morphology joint fine-grained core feature semantic clustering center vector. That is, the clustering network can reveal and reflect the global common semantic coding features between the inorganic board surface state semantic coding feature map and the inorganic board structural morphology deep semantic coding feature map to express the main mode of the inorganic board surface state and structural morphology. Subsequently, a query vector and a value vector are constructed based on each joint fine-grained core feature vector, and a key vector is constructed based on the semantic clustering center vector, which is input into the global semantic interaction encoder based on the converter structure to capture the long-distance contextual dependency between feature vectors using the self-attention mechanism, realize global feature interaction, and obtain the set of inorganic board surface state-structural morphology context joint fine-grained core feature vectors. That is, the query vector represents the information that the current feature vector needs to pay attention to, the key vector is used to calculate the semantic association between different features, and the value vector represents the specific content of the feature, so that each feature vector can represent its position and role in the whole based on the global context. Finally, the surface state-structure morphology context joint fine-grained core feature vectors of each inorganic board are reshaped to obtain the final inorganic board image-depth fine-grained joint semantic saliency fusion encoding feature map. That is, through reshaping, the spatial information of the feature vector can be restored, so that each feature vector corresponds to a specific position in the image, thereby better representing and utilizing the spatial relationship and contextual information between the surface features of the inorganic board.

[0037] Specifically, in an embodiment of the present application, the inorganic panel surface state semantic coding features and the inorganic panel structural morphology depth semantic coding features are subjected to cross-domain fine-grained semantic joint interaction to obtain inorganic panel image-depth fine-grained joint semantic significant fusion coding features, including: performing feature fine-grained decoupling on the inorganic panel surface state semantic coding feature map and the inorganic panel structural morphology depth semantic coding feature map to obtain a set of inorganic panel surface state semantic local fine-grained feature vectors and a set of inorganic panel structural morphology depth semantic local fine-grained feature vectors; performing cross-domain fine-grained feature significant fusion on the set of inorganic panel surface state semantic local fine-grained feature vectors and the set of inorganic panel structural morphology depth semantic local fine-grained feature vectors to obtain the inorganic panel image-depth fine-grained joint semantic significant fusion coding features.

[0038] Specifically, in an embodiment of the present application, the inorganic board surface state semantic coding feature map and the inorganic board structural morphology depth semantic coding feature map are feature fine-grained decoupling to obtain a set of inorganic board surface state semantic local fine-grained feature vectors and a set of inorganic board structural morphology depth semantic local fine-grained feature vectors, including: expanding each feature matrix of the inorganic board surface state semantic coding feature map along the channel dimension into a feature vector to obtain the set of inorganic board surface state semantic local fine-grained feature vectors; expanding each feature matrix of the inorganic board structural morphology depth semantic coding feature map along the channel dimension into a feature vector to obtain the set of inorganic board structural morphology depth semantic local fine-grained feature vectors.

[0039] Specifically, in an embodiment of the present application, a cross-domain fine-grained feature significant fusion is performed on the set of inorganic board surface state semantic local fine-grained feature vectors and the set of inorganic board structural morphology depth semantic local fine-grained feature vectors to obtain the inorganic board image-depth fine-grained joint semantic significant fusion coding features, including: each group of corresponding inorganic board surface state semantic local fine-grained feature vectors in the set of inorganic board surface state semantic local fine-grained feature vectors and the set of inorganic board structural morphology depth semantic local fine-grained feature vectors are fused. The feature vector and the deep semantic local fine-grained feature vector of the inorganic board structural morphology are input into the fine-grained common feature extraction network to obtain a set of fine-grained core feature vectors of the surface state-structural morphology of the inorganic board; the set of fine-grained core feature vectors of the surface state-structural morphology of the inorganic board is input into the clustering network to obtain the semantic clustering center vector of the fine-grained core feature of the surface state-structural morphology of the inorganic board; based on the surface state-structural morphology of each inorganic board in the set of fine-grained core feature vectors of the inorganic board The morphology joint fine-grained core feature vector constructs an inorganic board surface state-structure morphology query vector and an inorganic board surface state-structure morphology value vector, and constructs an inorganic board surface state-structure morphology key vector based on the inorganic board surface state-structure morphology joint fine-grained core feature semantic clustering center vector, and the inorganic board surface state-structure morphology query vector, the inorganic board surface state-structure morphology value vector and the inorganic board surface state-structure morphology of each inorganic board surface state-structure morphology joint fine-grained core feature vector are combined. The appearance key vector is input into a global semantic interaction encoder based on a converter structure to obtain a set of inorganic panel surface state-structural morphology context joint fine-grained core feature vectors; each inorganic panel surface state-structural morphology context joint fine-grained core feature vector in the set of inorganic panel surface state-structural morphology context joint fine-grained core feature vectors is feature-reshaped to obtain an inorganic panel image-depth fine-grained joint semantically significant fusion coding feature map as the inorganic panel image-depth fine-grained joint semantically significant fusion coding feature.

[0040] More specifically, in an embodiment of the present application, each group of corresponding inorganic board surface state semantic local fine-grained feature vectors and inorganic board structural morphology depth semantic local fine-grained feature vectors in the set of inorganic board surface state semantic local fine-grained feature vectors and the set of inorganic board structural morphology depth semantic local fine-grained feature vectors are input into a fine-grained common feature extraction network to obtain a set of inorganic board surface state-structural morphology joint fine-grained core feature vectors, including: respectively calculating the position point addition, position point multiplication and position point subtraction between the inorganic board surface state semantic local fine-grained feature vectors and the inorganic board structural morphology depth semantic local fine-grained feature vectors to obtain the inorganic board surface state-structural morphology point addition vector, ... morphology point product vector and inorganic board surface state - structural morphology point subtraction vector; cascade the inorganic board surface state - structural morphology point addition vector, the inorganic board surface state - structural morphology point product vector and the inorganic board surface state - structural morphology point subtraction vector to obtain an inorganic board surface state - structural morphology cascade vector; matrix-vector multiplication of the weight matrix and the inorganic board surface state - structural morphology cascade vector, and positionally add the obtained inorganic board surface state - structural morphology weight vector and bias vector to obtain an inorganic board surface state - structural morphology joint vector; use tanh function to process the inorganic board surface state - structural morphology joint vector to obtain the inorganic board surface state - structural morphology joint fine-grained core feature vector.

[0041] More specifically, in an embodiment of the present application, the set of fine-grained core feature vectors of the surface state-structural morphology of the inorganic panel is input into a clustering network to obtain a semantic clustering center vector of the fine-grained core features of the surface state-structural morphology of the inorganic panel, including: calculating the positional mean vector of the set of fine-grained core feature vectors of the surface state-structural morphology of the inorganic panel to obtain a semantic clustering center vector of the fine-grained core features of the surface state-structural morphology of the inorganic panel.

[0042] More specifically, in an embodiment of the present application, an inorganic panel surface state-structural morphology query vector and an inorganic panel surface state-structural morphology value vector are constructed based on each inorganic panel surface state-structural morphology joint fine-grained core feature vector in the set of the inorganic panel surface state-structural morphology joint fine-grained core feature vectors, and an inorganic panel surface state-structural morphology key vector is constructed based on the inorganic panel surface state-structural morphology joint fine-grained core feature semantic clustering center vector, and the inorganic panel surface state-structural morphology query vector, the inorganic panel surface state-structural morphology value vector and the inorganic panel surface state-structural morphology key vector of each inorganic panel surface state-structural morphology joint fine-grained core feature vector are input into a global semantic interactive encoder based on a converter structure to obtain a set of inorganic panel surface state-structural morphology context joint fine-grained core feature vectors. , including: multiplying the query embedding matrix with the inorganic board surface state-structural morphology joint fine-grained core feature vector to obtain the inorganic board surface state-structural morphology query vector; multiplying the value embedding matrix with the inorganic board surface state-structural morphology joint fine-grained core feature vector to obtain the inorganic board surface state-structural morphology value vector; multiplying the key embedding matrix with the inorganic board surface state-structural morphology joint fine-grained core feature semantic clustering center vector to obtain the inorganic board surface state-structural morphology key vector; inputting the inorganic board surface state-structural morphology query vector, the inorganic board surface state-structural morphology value vector and the inorganic board surface state-structural morphology key vector into the converter structure-based global semantic interactive encoder to obtain the inorganic board surface state-structural morphology context joint fine-grained core feature vector.

[0043] More specifically, in an embodiment of the present application, the inorganic board surface state-structural morphology query vector, the inorganic board surface state-structural morphology value vector and the inorganic board surface state-structural morphology key vector are input into the global semantic interactive encoder based on the converter structure to obtain the inorganic board surface state-structural morphology context joint fine-grained core feature vector, including: multiplying the inorganic board surface state-structural morphology query vector with the transposed vector of the inorganic board surface state-structural morphology key vector, and obtaining the inorganic board surface state-structural morphology key vector. The state-structure morphology association matrix and the square root of the length of the inorganic board surface state-structure morphology key vector are divided by position to obtain the inorganic board surface state-structure morphology association modulation matrix; the inorganic board surface state-structure morphology association modulation matrix is ​​processed using a softmax function, and the obtained inorganic board surface state-structure morphology association activation matrix is ​​matrix-vector multiplied with the inorganic board surface state-structure morphology value vector to obtain the inorganic board surface state-structure morphology context joint fine-grained core feature vector.

[0044] In the embodiment of the present application, specifically, the inorganic board surface state semantic coding feature map and the inorganic board structure morphology depth semantic coding feature map are subjected to cross-domain fine-grained global semantic interaction, which can be expressed by the formula: ; ; ; ; ; ; ; ; ; ;in, and They are respectively the inorganic board surface state semantic coding feature map and the inorganic board structure morphology depth semantic coding feature map, In order to perform fine-grained feature decoupling operations on feature maps, , , and They are the first, second, and third semantic local fine-grained feature vectors of the surface state of inorganic panels. and The semantic local fine-grained feature vector of the surface state of inorganic panels, , , and They are the first, second, and third in the set of local fine-grained feature vectors of the depth semantics of the inorganic plate structure morphology. and The local fine-grained feature vector of the inorganic plate structure morphology deep semantics, , and They are point-by-point addition, point-by-point multiplication, and point-by-point subtraction. For cascade processing, and are the weight matrix and bias vector respectively, for function, It is the first in the set of fine-grained core feature vectors of the surface state and structural morphology of inorganic plates. The surface state-structure morphology of inorganic panels is combined with fine-grained core feature vectors. is the number of vectors in the set of fine-grained core feature vectors of the inorganic plate surface state-structure morphology joint, is the center vector of the semantic clustering of the fine-grained core features of the surface state and structural morphology of inorganic panels. is the query embedding matrix, yes The corresponding inorganic plate surface state-structure morphology query vector, is the value embedding matrix, yes The corresponding inorganic plate surface state-structure morphology value vector, is the key embedding matrix, is the surface state-structure morphology bond vector of the inorganic plate, for The transposed vector of for Length, yes function, is the matrix-vector multiplication, , , and They are the first, second, and third in the set of fine-grained core feature vectors of the inorganic plate surface state-structure morphology context joint. and The inorganic plate surface state-structure morphology context joint fine-grained core feature vector, is a set of fine-grained core feature vectors of the inorganic plate surface state-structure morphology context joint, To perform a feature reshape operation, It is the inorganic board image-depth fine-grained joint semantically significant fusion encoding feature map.

[0045] In step S123, the basic score of structural strength is obtained based on the inorganic board image-depth fine-grained joint semantically significant fusion coding features. Specifically, in the implementation of the present application, the basic score of structural strength is obtained based on the inorganic board image-depth fine-grained joint semantically significant fusion coding features, including: inputting the inorganic board image-depth fine-grained joint semantically significant fusion coding feature map into the basic scoring module based on the decoder to obtain the basic score of structural strength. That is, the inorganic board image-depth fine-grained joint semantically significant fusion coding features obtained by cross-domain fine-grained interaction using the inorganic board surface state semantic coding feature map and the inorganic board structural morphology depth semantic coding feature map are decoded to automatically obtain the basic score of structural strength. In this way, the subtle differences and complex features of the surface state and structural morphology of inorganic boards can be accurately identified and analyzed. At the same time, the mutual influence between the surface image information and the surface depth information can be perceived, thereby enhancing the comprehensiveness and accuracy of the evaluation results. In particular, in a specific embodiment of the present application, the inorganic board image-depth fine-grained joint semantically significant fusion coding feature map is input into a decoder-based basic scoring module to obtain the structural strength basic score, including: using the decoder to decode and regress the inorganic board image-depth fine-grained joint semantically significant fusion coding feature map using the following decoding formula to obtain the structural strength basic score; wherein the decoding formula is: ;in, is a vector obtained by expanding and concatenating the feature matrices on each channel dimension in the inorganic board image-depth fine-grained joint semantically significant fusion coding feature map, is the structural strength base score, is the weight matrix, Represents matrix-vector multiplication.

[0046] In particular, when the inorganic board surface state semantic coding feature map and the inorganic board structural morphology depth semantic coding feature map respectively represent the surface image and surface depth coding features of the surface image information and surface depth information of the inorganic board to be detected, when performing cross-domain fine-grained core joint feature semantic interaction coding, the inorganic board image-depth fine-grained joint semantic significant fusion coding feature map will also have the diversity of joint interaction distribution of heterogeneous image information based on their respective feature fine-grained cores under the cross-feature domain feature distribution. Therefore, when the inorganic board image-depth fine-grained joint semantic significant fusion coding feature map is decoded by a decoder, it will affect the accuracy of the decoding result.

[0047] Preferably, inputting the inorganic board image-depth fine-grained joint semantically significant fusion coding feature map into a decoder-based basic scoring module to obtain a structural strength basic score includes: determining the feature mean and feature variance of the inorganic board image-depth fine-grained joint semantically significant fusion coding feature map, and dividing the feature mean by the feature variance to obtain an inorganic board image-depth fine-grained joint semantically significant fusion coding probability statistic value: wherein the process can be expressed as: ;in, is the feature mean of the inorganic panel image-depth fine-grained joint semantically significant fusion encoding feature map, is the feature variance of the inorganic panel image-depth fine-grained joint semantically significant fusion encoding feature map, A probability statistic value of the inorganic plate image-depth fine-grained joint semantic saliency fusion encoding; The inorganic board image-depth fine-grained joint semantically significant fusion coding feature map is point-multiplied with the inverse of the maximum eigenvalue in the inorganic board image-depth fine-grained joint semantically significant fusion coding feature map to obtain the inorganic board image-depth fine-grained joint semantically significant fusion coding probability constraint map: wherein the process can be expressed as: ;in, The inorganic board image-depth fine-grained joint semantically significant fusion encoding feature map, represents the inverse of the maximum eigenvalue in the inorganic board image-depth fine-grained joint semantically significant fusion encoding feature map, It means point multiplication by position. A probability constraint map for encoding the inorganic panel image-depth fine-grained joint semantic saliency fusion; After dot-adding the inorganic board image-depth fine-grained joint semantically significant fusion coding probability constraint graph and the inorganic board image-depth fine-grained joint semantically significant fusion coding probability statistical value, the logarithmic value with base 2 is calculated to obtain the inorganic board image-depth fine-grained joint semantically significant fusion coding information interaction graph: wherein the process can be expressed as: ;in, is the inorganic panel image-depth fine-grained joint semantic saliency fusion encoding probability constraint map, Indicates adding by position point. The probability statistics of the inorganic plate image-depth fine-grained joint semantic saliency fusion encoding, The inorganic panel image-depth fine-grained joint semantically significant fusion coding information interaction diagram; After point-subtracting the inorganic board image-depth fine-grained joint semantically significant fusion coding probability constraint map with the inorganic board image-depth fine-grained joint semantically significant fusion coding probability statistics, the inverse of each eigenvalue is calculated to obtain the inorganic board image-depth fine-grained joint semantically significant fusion coding sequence constraint map: wherein the process can be expressed as: ;in, The probability statistics of the inorganic plate image-depth fine-grained joint semantic saliency fusion encoding, is the inorganic panel image-depth fine-grained joint semantic saliency fusion encoding probability constraint map, It means to subtract by position point. It means calculating the inverse of each eigenvalue in the feature map. The inorganic panel image-depth fine-grained joint semantic saliency fusion coding sequence constraint graph; The inorganic panel image-depth fine-grained joint semantically significant fusion coding information interaction graph and the inorganic panel image-depth fine-grained joint semantically significant fusion coding sequence constraint graph are interpolated to obtain an optimized inorganic panel image-depth fine-grained joint semantically significant fusion coding feature graph; the optimized inorganic panel image-depth fine-grained joint semantically significant fusion coding feature graph is input into the decoder-based basic scoring module to obtain the structural strength basic score.

[0048] That is, the present application takes into account that when the weight matrix of the decoder acts on the inorganic board image-depth fine-grained joint semantically significant fusion coding feature vector after the inorganic board image-depth fine-grained joint semantically significant fusion coding feature map is expanded, the distribution diversity of the inorganic board image-depth fine-grained joint semantically significant fusion coding feature map causes the uncertainty of the weight stimulus parameters of the weight matrix of the decoder, resulting in the lack of a posteriori probability density inference of the inorganic board image-depth fine-grained joint semantically significant fusion coding feature map through the weight matrix, which will affect the accuracy of the decoding result.

[0049] In this case, the probability statistical characteristics of the inorganic board image-depth fine-grained joint semantically significant fusion coding feature map are used to simulate the mesoscale interactive structure under probability constraints between the feature value scale and the feature map scale of the inorganic board image-depth fine-grained joint semantically significant fusion coding feature map, so as to construct a bidirectional latent variable motif based on the mesoscale short sequence relative to the probability statistical value to perform mesoscale bidirectional migration, and perform posterior recovery based on the sequence constraints on the interactive information, thereby improving the convergence effect in the probability density domain and improving the accuracy of the basic structural strength score obtained by the basic scoring module based on the decoder of the inorganic board image-depth fine-grained joint semantically significant fusion coding feature map. In this way, the subtle differences and complex features of the surface state and structural morphology of inorganic boards can be accurately identified and analyzed. At the same time, the mutual influence between the surface image information and the surface depth information can be perceived, thereby enhancing the comprehensiveness and accuracy of the evaluation results.

[0050] Therefore, based on the surface image information and surface depth information of the inorganic board to be detected, the basic score of the structural strength of the inorganic board to be detected is clearly explained, which uses artificial intelligence-based image analysis and information processing technology to extract surface state semantic features from the surface image information of the inorganic board to be detected, and extract surface structural morphology features from the surface depth information, so as to automatically obtain the basic score of structural strength based on the cross-domain fine-grained interaction features between the semantic coding features of the surface state of the inorganic board and the semantic coding features of the depth of the structural morphology of the inorganic board. In this way, the subtle differences and complex features of the surface state and structural morphology of the inorganic board can be accurately identified and analyzed, avoiding the defects of the inorganic board quality inspection classification method in the existing patents. At the same time, the mutual influence between the surface image information and the surface depth information can be perceived, which helps to enhance the comprehensiveness of the evaluation results.

[0051] In step S130, based on the surface depth information of the inorganic plate to be detected, a secondary score of the surface flatness of the inorganic plate to be detected is obtained. It should be understood that the surface depth information of the inorganic plate provides detailed data on the surface morphology, including features such as unevenness and corrugation. These features directly affect the flatness and overall quality of the plate. Considering that there may be subtle defects or irregularity information that cannot be captured during the initial scoring of the inorganic plate, and it is necessary to ensure that no important information is ignored in actual production, based on this, in the technical solution of the present application, it is necessary to obtain a secondary score of the surface flatness of the inorganic plate to be detected based on the surface depth information of the inorganic plate to be detected. By calculating the secondary score of surface flatness, it is beneficial to improve the reliability of the quality classification results of the inorganic plate. In particular, in a specific embodiment of the present application, based on the surface depth information of the inorganic board to be detected, a secondary score of the surface flatness of the inorganic board to be detected is obtained, including: a. Depth information preprocessing: the surface depth information (i.e., depth map) of the inorganic board to be detected collected by a depth camera is subjected to noise reduction processing to reduce the impact of noise on subsequent analysis; b. Flatness-related feature extraction: the height difference between each divided area and its neighboring area in the depth map is calculated to obtain specific data on the height change distribution of the inorganic board surface, and the surface roughness of each area is calculated using a standard roughness index to obtain surface roughness data of each area; c. Score calculation: the score corresponding to each area is obtained based on the result of comparing the height change and the roughness index with a preset threshold, and the secondary score of the surface flatness of the inorganic board is obtained by comprehensive calculation based on the scores of each area.

[0052] In step S140, the inorganic panel to be inspected is quality classified based on the basic structural strength score and the secondary surface flatness score to obtain a quality classification result. It should be understood that the quality of the inorganic panel not only depends on its surface flatness, but is also closely related to its structural strength. Relying on a certain score alone may not fully reflect the overall performance of the panel. For example, a panel with a seemingly flat surface may still break or deform during use if its structural strength is insufficient. Therefore, combining the basic structural strength score and the secondary surface flatness score can more comprehensively evaluate the quality of the inorganic panel, and this comprehensive evaluation can effectively improve the quality control level of the inorganic panel. In particular, in a specific embodiment of the present application, the inorganic panel to be tested is quality classified based on the basic structural strength score and the secondary surface flatness score to obtain a quality classification result, including: a. Setting classification standards: the quality grades of inorganic panels are divided into three grades: A, B, and C, among which grade A is the highest quality grade, requiring both the basic structural strength score and the secondary surface flatness score to reach a higher standard; grade B is a medium quality grade, requiring both scores to reach a qualified standard; grade C is the lowest quality grade, as long as the basic usage requirements are met; b. According to the set weights, the basic structural strength score and the secondary surface flatness score are weighted and summed to obtain a comprehensive score of the inorganic panel, where the weight setting can be adjusted according to the actual application scenario requirements; c. Classification judgment: compare the comprehensive score of the inorganic panel with the set classification standard (actually the set threshold) to determine its quality grade.

[0053] In summary, the inorganic panel quality classification method based on the embodiment of the present application is explained, which first obtains the surface image and depth information of the inorganic panel to be tested, calculates the basic structural strength score of the inorganic panel through this information, and calculates the secondary surface flatness score according to the surface depth information, and finally, combines the basic structural strength score and the secondary surface flatness score to classify the inorganic panel to be tested. In this way, more reliable inorganic panel quality control can be achieved.

[0054] Figure 4 FIG. 1 is a system block diagram of an inorganic panel quality classification system according to an embodiment of the present application. Figure 4As shown, according to the inorganic panel quality classification system 100 of the embodiment of the present application, it includes: an inorganic panel surface information acquisition module 110, which is used to obtain the surface image information and surface depth information of the inorganic panel to be detected; an inorganic panel structural strength basic score calculation module 120, which is used to obtain the structural strength basic score of the inorganic panel to be detected based on the surface image information and surface depth information of the inorganic panel to be detected; an inorganic panel surface flatness secondary score calculation module 130, which is used to obtain the surface flatness secondary score of the inorganic panel to be detected based on the surface depth information of the inorganic panel to be detected; an inorganic panel quality classification result generation module 140, which is used to perform quality classification on the inorganic panel to be detected based on the structural strength basic score and the surface flatness secondary score to obtain a quality classification result, wherein, The inorganic panel structural strength basic score calculation module 120 includes: an inorganic panel surface information semantic feature extraction unit, which is used to extract semantic features from the surface image information and surface depth information of the inorganic panel to be detected respectively to obtain the inorganic panel surface state semantic coding features and the inorganic panel structural morphology depth semantic coding features; an inorganic panel image-depth fine-grained joint semantically significant fusion coding feature generation unit, which is used to perform cross-domain fine-grained semantic joint interaction on the inorganic panel surface state semantic coding features and the inorganic panel structural morphology depth semantic coding features to obtain the inorganic panel image-depth fine-grained joint semantically significant fusion coding features; a structural strength basic score generation unit, which is used to obtain the structural strength basic score based on the inorganic panel image-depth fine-grained joint semantically significant fusion coding features.

[0055] Here, those skilled in the art will appreciate that the specific functions and operations of the various units and modules in the inorganic panel quality classification system 100 have been described in detail above. Figures 1 to 3 The inorganic plate quality classification method has been described in detail, and therefore, its repeated description will be omitted.

[0056] In summary, the inorganic panel quality classification system 100 based on the embodiment of the present application is explained, which first obtains the surface image and depth information of the inorganic panel to be tested, calculates the basic structural strength score of the inorganic panel through this information, and calculates the secondary surface flatness score according to the surface depth information, and finally, combines the basic structural strength score and the secondary surface flatness score to classify the inorganic panel to be tested. In this way, more reliable inorganic panel quality control can be achieved.

Claims

1. A method for quality classification of inorganic panels, characterized in that: include: Acquire surface image information and surface depth information of the inorganic plate to be inspected; Based on the surface image information and surface depth information of the inorganic board to be detected, a basic score of the structural strength of the inorganic board to be detected is obtained, including: performing semantic feature extraction on the surface image information and surface depth information of the inorganic board to be detected respectively to obtain the semantic coding features of the surface state of the inorganic board and the depth semantic coding features of the structural morphology of the inorganic board; performing cross-domain fine-grained semantic joint interaction on the semantic coding features of the surface state of the inorganic board and the depth semantic coding features of the structural morphology of the inorganic board to obtain the image-depth fine-grained joint semantic significant fusion coding features of the inorganic board; based on the inorganic board image-depth fine-grained joint semantic significant fusion coding features, the basic score of the structural strength is obtained; Based on the surface depth information of the inorganic plate to be detected, obtaining a secondary score of the surface flatness of the inorganic plate to be detected; The inorganic plate to be inspected is quality classified based on the basic structural strength score and the secondary surface flatness score to obtain a quality classification result.

2. The inorganic plate quality classification method according to claim 1, characterized in that: Semantic feature extraction is performed on the surface image information and the surface depth information of the inorganic board to be detected to obtain the semantic coding features of the surface state of the inorganic board and the semantic coding features of the depth of the structure morphology of the inorganic board, including: Inputting the surface image information of the inorganic board to be detected into an inorganic board surface state feature extractor based on a hole convolutional neural network model to obtain an inorganic board surface state semantic coding feature map as the inorganic board surface state semantic coding feature; The surface depth information of the inorganic board to be detected is input into the inorganic board surface structure morphology feature extractor based on the depth separable convolutional neural model to obtain the inorganic board structure morphology depth semantic coding feature map as the inorganic board structure morphology depth semantic coding feature.

3. The inorganic plate quality classification method according to claim 2, characterized in that: The inorganic board surface state semantic coding features and the inorganic board structure morphology depth semantic coding features are subjected to cross-domain fine-grained semantic joint interaction to obtain inorganic board image-depth fine-grained joint semantic significant fusion coding features, including: Performing feature fine-grained decoupling on the inorganic board surface state semantic coding feature map and the inorganic board structure morphology depth semantic coding feature map to obtain a set of inorganic board surface state semantic local fine-grained feature vectors and a set of inorganic board structure morphology depth semantic local fine-grained feature vectors; The set of semantic local fine-grained feature vectors of the inorganic board surface state and the set of semantic local fine-grained feature vectors of the inorganic board structural morphology depth are significantly fused across domains to obtain the inorganic board image-depth fine-grained joint semantically significant fused coding features.

4. The inorganic plate quality classification method according to claim 3, characterized in that: The inorganic board surface state semantic coding feature map and the inorganic board structure morphology depth semantic coding feature map are subjected to feature fine-grained decoupling to obtain a set of inorganic board surface state semantic local fine-grained feature vectors and a set of inorganic board structure morphology depth semantic local fine-grained feature vectors, including: Expanding each feature matrix along the channel dimension of the inorganic board surface state semantic encoding feature map into a feature vector to obtain a set of inorganic board surface state semantic local fine-grained feature vectors; Each feature matrix along the channel dimension of the inorganic board structure morphology depth semantic encoding feature map is expanded into a feature vector to obtain a set of local fine-grained feature vectors of the inorganic board structure morphology depth semantics.

5. The inorganic plate quality classification method according to claim 4, characterized in that: The set of the inorganic board surface state semantic local fine-grained feature vectors and the set of the inorganic board structure morphology depth semantic local fine-grained feature vectors are subjected to cross-domain fine-grained feature saliency fusion to obtain the inorganic board image-depth fine-grained joint semantic saliency fusion coding features, including: Inputting each corresponding group of inorganic board surface state semantic local fine-grained feature vectors and inorganic board structural morphology depth semantic local fine-grained feature vectors in the set of inorganic board surface state semantic local fine-grained feature vectors and inorganic board structural morphology depth semantic local fine-grained feature vectors into a fine-grained common feature extraction network to obtain a set of inorganic board surface state-structural morphology joint fine-grained core feature vectors; Inputting the set of inorganic plate surface state-structure morphology joint fine-grained core feature vectors into a clustering network to obtain the inorganic plate surface state-structure morphology joint fine-grained core feature semantic clustering center vector; Based on each inorganic board surface state-structural morphology joint fine-grained core feature vector in the set of inorganic board surface state-structural morphology joint fine-grained core feature vectors, an inorganic board surface state-structural morphology query vector and an inorganic board surface state-structural morphology value vector are constructed, and based on the inorganic board surface state-structural morphology joint fine-grained core feature semantic clustering center vector, an inorganic board surface state-structural morphology key vector is constructed, and the inorganic board surface state-structural morphology query vector, the inorganic board surface state-structural morphology value vector and the inorganic board surface state-structural morphology key vector of each inorganic board surface state-structural morphology joint fine-grained core feature vector are input into a global semantic interactive encoder based on a converter structure to obtain a set of inorganic board surface state-structural morphology context joint fine-grained core feature vectors; The feature shape of each inorganic board surface state-structural morphology context joint fine-grained core feature vector in the set of the inorganic board surface state-structural morphology context joint fine-grained core feature vector is reshaped to obtain an inorganic board image-depth fine-grained joint semantically significant fusion coding feature map as the inorganic board image-depth fine-grained joint semantically significant fusion coding feature.

6. The inorganic plate quality classification method according to claim 5, characterized in that: Inputting each corresponding group of inorganic board surface state semantic local fine-grained feature vectors and inorganic board structural morphology depth semantic local fine-grained feature vectors in the set of inorganic board surface state semantic local fine-grained feature vectors and inorganic board structural morphology depth semantic local fine-grained feature vectors into a fine-grained common feature extraction network to obtain a set of inorganic board surface state-structural morphology joint fine-grained core feature vectors, including: Respectively calculating the position point addition, position point multiplication and position point subtraction between the inorganic board surface state semantic local fine-grained feature vector and the inorganic board structural morphology depth semantic local fine-grained feature vector to obtain the inorganic board surface state-structural morphology point addition vector, the inorganic board surface state-structural morphology point multiplication vector and the inorganic board surface state-structural morphology point subtraction vector; Cascading the inorganic plate surface state-structure morphology point addition vector, the inorganic plate surface state-structure morphology point multiplication vector and the inorganic plate surface state-structure morphology point subtraction vector to obtain an inorganic plate surface state-structure morphology cascade vector; Performing matrix-vector multiplication on the weight matrix and the inorganic plate surface state-structure morphology cascade vector, adding the obtained inorganic plate surface state-structure morphology weight vector and the bias vector by position to obtain the inorganic plate surface state-structure morphology joint vector; The inorganic plate surface state-structure morphology joint vector is processed using a tanh function to obtain the inorganic plate surface state-structure morphology joint fine-grained core feature vector.

7. The inorganic plate quality classification method according to claim 6, characterized in that: The set of inorganic plate surface state-structure morphology joint fine-grained core feature vectors is input into a clustering network to obtain the inorganic plate surface state-structure morphology joint fine-grained core feature semantic clustering center vector, including: The positional mean vector of the set of the inorganic board surface state-structure morphology joint fine-grained core feature vectors is calculated to obtain the inorganic board surface state-structure morphology joint fine-grained core feature semantic clustering center vector.

8. The inorganic plate quality classification method according to claim 7, characterized in that: Based on each inorganic board surface state-structure morphology joint fine-grained core feature vector in the set of inorganic board surface state-structure morphology joint fine-grained core feature vectors, an inorganic board surface state-structure morphology query vector and an inorganic board surface state-structure morphology value vector are constructed, and based on the inorganic board surface state-structure morphology joint fine-grained core feature semantic clustering center vector, an inorganic board surface state-structure morphology key vector is constructed, and the inorganic board surface state-structure morphology query vector, the inorganic board surface state-structure morphology value vector and the inorganic board surface state-structure morphology key vector of each inorganic board surface state-structure morphology joint fine-grained core feature vector are input into a global semantic interactive encoder based on a converter structure to obtain a set of inorganic board surface state-structure morphology context joint fine-grained core feature vectors, including: Multiplying the query embedding matrix with the inorganic plate surface state-structure morphology joint fine-grained core feature vector to obtain the inorganic plate surface state-structure morphology query vector; Multiplying the value embedding matrix with the inorganic plate surface state-structure morphology joint fine-grained core feature vector to obtain the inorganic plate surface state-structure morphology value vector; Multiplying the bond embedding matrix with the inorganic board surface state-structure morphology joint fine-grained core feature semantic clustering center vector to obtain the inorganic board surface state-structure morphology bond vector; The inorganic panel surface state-structural morphology query vector, the inorganic panel surface state-structural morphology value vector and the inorganic panel surface state-structural morphology key vector are input into the converter structure-based global semantic interactive encoder to obtain the inorganic panel surface state-structural morphology context joint fine-grained core feature vector.

9. The inorganic plate quality classification method according to claim 8, characterized in that: Based on the inorganic plate image-depth fine-grained joint semantically significant fusion coding features, the structural strength basic score is obtained, including: The inorganic board image-depth fine-grained joint semantically significant fusion encoding feature map is input into a decoder-based basic scoring module to obtain the structural strength basic score.

10. An inorganic plate quality classification system, characterized in that: include: The inorganic board surface information acquisition module is used to obtain the surface image information and surface depth information of the inorganic board to be detected; An inorganic panel structural strength basic score calculation module is used to obtain the structural strength basic score of the inorganic panel to be detected based on the surface image information and surface depth information of the inorganic panel to be detected, wherein the inorganic panel structural strength basic score calculation module includes: an inorganic panel surface information semantic feature extraction unit, which is used to perform semantic feature extraction on the surface image information and surface depth information of the inorganic panel to be detected respectively to obtain the inorganic panel surface state semantic coding feature and the inorganic panel structural morphology depth semantic coding feature; an inorganic panel image-depth fine-grained joint semantically significant fusion coding feature generation unit, which is used to perform cross-domain fine-grained semantic joint interaction on the inorganic panel surface state semantic coding feature and the inorganic panel structural morphology depth semantic coding feature to obtain the inorganic panel image-depth fine-grained joint semantically significant fusion coding feature; a structural strength basic score generation unit, which is used to obtain the structural strength basic score based on the inorganic panel image-depth fine-grained joint semantically significant fusion coding feature; An inorganic plate surface flatness secondary score calculation module, used to obtain a surface flatness secondary score of the inorganic plate to be detected based on the surface depth information of the inorganic plate to be detected; The inorganic plate quality classification result generating module is used to perform quality classification on the inorganic plate to be inspected based on the basic structural strength score and the secondary surface flatness score to obtain a quality classification result.

Images