A lithology classification model construction method and lithology classification method

Abstract

The present invention provides a lithology classification model construction method and a lithology classification method, and relates to the technical field of lithology classification. The lithology classification model construction method comprises: extracting the lithology characteristics of each lithology in a lithology data set, and constructing a lithology adjacency matrix and a lithology adjacency submatrix according to the co-occurrence relationship of all lithologies; obtaining output lithology characteristics and output lithology sub-features according to the lithology adjacency matrix, the lithology adjacency submatrix and the lithology characteristics respectively; obtaining a graph representation and a node representation according to all output lithology characteristics, and obtaining a subgraph representation and a subnode representation according to all output lithology subfeatures; obtaining a model graph representation and a model subnode representation according to the graph representation and the subnode representation, and obtaining a model subgraph representation and a model node representation according to the subgraph representation and the node representation; and constructing a lithology classification model according to the model graph representation, the model node representation, the model subgraph representation and the model subnode representation. The present invention can effectively improve the efficiency of lithology classification of geological maps.

Description

Technical Field

[0001] The present invention relates to the technical field of lithology classification, and in particular to a lithology classification model construction method and a lithology classification method. Background Art

[0002] Remote sensing lithology classification is an important technology in the field of geology, and its application involves many fields such as mineral exploration, engineering geology, and environmental geology. The classification of lithology can help geologists and engineers better understand the geological structure and guide mineral exploration and engineering development.

[0003] Conventional classification methods usually rely on professionals to conduct field surveys and classify rock types through sampling and analysis. However, in order to obtain richer and more accurate geological lithology, a large amount of sampling and analysis is required, which results in a large amount of manpower and time required for manual rock type classification. Summary of the invention

[0004] The problem solved by the present invention is how to improve the efficiency of lithology classification.

[0005] In order to solve the above problems, the present invention provides a lithology classification model construction method and a lithology classification method.

[0006] In a first aspect, the present invention provides a method for constructing a lithology classification model, comprising:

[0007] Extracting the lithological characteristics of each lithology in the acquired lithological data set, and constructing a lithological adjacency matrix and a lithological adjacency submatrix according to the co-occurrence relationship of all the lithologies;

[0008] Based on the graph attention mechanism, a corresponding output lithology feature is obtained according to the lithology adjacency matrix and the lithology feature, and a corresponding output lithology sub-feature is obtained according to the lithology adjacency sub-matrix and the lithology feature;

[0009] Based on a preset multi-layer perceptron, a corresponding graph representation and a node representation are obtained according to all the output lithology features, and a corresponding subgraph representation and a subnode representation are obtained according to all the output lithology sub-features;

[0010] Compare the graph representation and the subnode representation to obtain a corresponding model graph representation and a model subnode representation, and compare the subgraph representation and the node representation to obtain a corresponding model subgraph representation and a model node representation;

[0011] The model graph representation, the model node representation, the model subgraph representation and the model subnode representation are aggregated to construct a lithology classification model.

[0012] Optionally, obtaining corresponding output lithological features according to the lithological adjacency matrix and the lithological features includes:

[0013] Determine, according to the lithology adjacency matrix and the lithology feature, a first adjacent lithology feature of the lithology feature in the lithology adjacency matrix;

[0014] Obtaining the corresponding output lithology feature according to the lithology feature and all the corresponding first adjacent lithology features;

[0015] The output lithology characteristics satisfy:

[0016]

[0017] in, is the x-th output lithological feature, is the xth lithological feature, is the i-th first adjacent lithological feature, N is the number of the first adjacent lithological features, W is a preset first linear transformation matrix, is the preset first initial weight vector, T is the transposition operation, LeakyReLU is the nonlinear operation, exp is the exponential operation, σ is the nonlinear activation function; || is the concatenation operation.

[0018] Optionally, obtaining corresponding output lithology sub-features according to the lithology adjacency sub-matrix and the lithology feature includes:

[0019] Determine, according to the lithology adjacency sub-matrix and the lithology feature, a second adjacent lithology feature of the lithology feature in the lithology adjacency sub-matrix;

[0020] According to the lithological feature and all the corresponding second adjacent lithological features, obtaining the corresponding output lithological sub-feature;

[0021] The output lithology sub-characteristics satisfy:

[0022]

[0023] in, is the x-th output lithology sub-feature, is the xth lithological feature, is the jth second adjacent lithological feature, M is the number of the second adjacent lithological features, V is a preset second linear transformation matrix, is the preset second initial weight vector, T is the transposition operation, LeakyReLU is the nonlinear operation, exp is the exponential operation, σ is the nonlinear activation function; || is the concatenation operation.

[0024] Optionally, the multilayer perceptron includes a first multilayer perceptron and a second multilayer perceptron; and obtaining corresponding graph representations and node representations according to all the output lithological characteristics includes:

[0025] Inputting all the output lithological features after pooling processing into the first multi-layer perceptron, and outputting the corresponding graph representation;

[0026] All the output lithology features are input into the second multi-layer perceptron, and the corresponding node representations are output.

[0027] Optionally, the multilayer perceptron includes a third multilayer perceptron and a fourth multilayer perceptron; the step of obtaining corresponding subgraph representations and subnode representations according to all the output lithology sub-features includes:

[0028] Inputting all the output lithology sub-features after pooling processing into the third multi-layer perceptron, and outputting the corresponding sub-graph representation;

[0029] All of the output lithology sub-features are input into the fourth multi-layer perceptron, and the corresponding sub-node representations are output.

[0030] Optionally, the comparing the graph representation and the subnode representation to obtain a corresponding model graph representation and model subnode representation includes:

[0031] Obtaining a first comparison result through a preset first discriminator according to the graph representation and the child node representation;

[0032] The weight of the graph representation is updated according to the first comparison result to obtain the corresponding model graph representation, and the sub-node representation is updated according to the first comparison result to obtain the corresponding model sub-node representation.

[0033] Optionally, the comparing the subgraph representation and the node representation to obtain a corresponding model subgraph representation and model node representation includes:

[0034] Obtaining a second comparison result through a preset second discriminator according to the subgraph representation and the node representation;

[0035] The weight of the subgraph representation is updated according to the second comparison result to obtain the corresponding model subgraph representation, and the node representation is updated according to the second comparison result to obtain the corresponding model node representation.

[0036] Optionally, constructing a lithology adjacency matrix according to the co-occurrence relationship of all the lithologies includes:

[0037] constructing an initial lithology adjacency matrix according to the co-occurrence relationship of all the lithologies in the lithology data set;

[0038] Obtaining the lithology adjacency matrix through probability conversion according to the initial lithology adjacency matrix;

[0039] The lithology adjacency submatrix is ​​obtained by diffusion processing according to the lithology adjacency matrix.

[0040] Optionally, the adjacency submatrix satisfies:

[0041] S = exp(tAD --1 -t);

[0042] Where S is the lithology adjacency matrix, t is the preset diffusion time, A is the lithology adjacency matrix, and D -1 is the inverse matrix of the preset diagonal matrix D.

[0043] In a second aspect, the present invention provides a lithology classification method, comprising:

[0044] Acquiring an optical image to be processed;

[0045] The optical image to be processed is input into a lithology classification model, and a lithology classification result corresponding to the optical image to be processed is output; wherein the lithology classification model is obtained by using the lithology classification model construction method as described in any one of the first aspects.

[0046] The beneficial effects of the lithology classification model construction method and the lithology classification method of the present invention are as follows: the lithology adjacency matrix and the lithology adjacency submatrix constructed by all lithologies extracted from the lithology data set can reflect the structural relationship of the complete lithology characteristics of the lithology data set through the lithology adjacency matrix, and the local relationship in the lithology structure can be further reflected through the lithology adjacency submatrix, and all the complete lithology feature information in the lithology data set can be more comprehensively reflected through the lithology adjacency matrix and the lithology adjacency submatrix, so that the co-occurrence relationship and combination law between different lithology categories in the lithology data set can be more clearly displayed, and the connection and interaction between different lithologies can be discovered; the output lithology characteristics and the output lithology sub-features are obtained respectively through the extracted lithology characteristics, the lithology adjacency matrix and the lithology adjacency submatrix, and the corresponding graph representation, subgraph representation, node representation and subnode representation are obtained according to the output lithology characteristics and the output lithology sub-features through the multi-layer perceptron, so that the correlation relationship and the overall structure of different lithology characteristics can be more intuitively presented, so that the relationship and characteristics between the lithology characteristics can be deeply analyzed and compared, while avoiding the large number of lithology characteristics. The separate processing of feature data effectively reduces the computational cost of the model and improves the generalization ability and efficiency; the graph representation and the sub-node representation are compared to obtain the model icon and model node representation that meet the requirements of consistency, and the sub-graph representation and the node representation are compared to obtain the model sub-graph representation and model node representation that meet the requirements of consistency. Finally, the lithology classification model is constructed by aggregation through the model graph representation, the model sub-graph representation, the model node representation and the model sub-node representation, so that the lithology classification model can comprehensively utilize the node characteristics and graph structure information, better understand the characteristics and relationships in the lithology data set, improve the accuracy and efficiency of lithology classification, and can effectively handle complex lithology categories, provide fast and reliable classification results, avoid the large amount of manpower and time consumed by manual field sampling, thereby effectively improving the efficiency of lithology classification and saving human resources. At the same time, through the graph structure and node structure that meet the consistency requirements, the lithology classification model performs lithology classification according to the same classification standard, thereby improving the performance and effect of the lithology classification model, so that it performs better in the task of lithology classification of geological maps. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 A schematic diagram of a process of constructing a lithology classification model according to an embodiment of the present invention;

[0048] Figure 2 A schematic diagram of a process of generating a lithology classification model through aggregation according to an embodiment of the present invention;

[0049] Figure 3 Schematic diagram of an initial matrix according to an embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. Although certain embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be interpreted as being limited to the embodiments described herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present invention. It should be understood that the drawings and embodiments of the present invention are only for exemplary purposes and are not intended to limit the scope of protection of the present invention.

[0051] It should be understood that the various steps described in the method embodiments of the present invention may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this respect.

[0052] The term "including" and its variations used in this document are open inclusions, that is, "including but not limited to"; the term "based on" means "based at least in part on"; the term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one other embodiment"; the term "some embodiments" means "at least some embodiments"; the term "optionally" means "optional embodiments". The relevant definitions of other terms will be given in the following description. It should be noted that the concepts of "first", "second", etc. mentioned in the present invention are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0053] It should be noted that the modifications of "one" and "plurality" mentioned in the present invention are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0054] This embodiment provides a lithology classification model construction method and a lithology classification method.

[0055] like Figure 1 As shown, a method for constructing a lithology classification model provided by an embodiment of the present invention includes:

[0056] S100, extracting the lithology characteristics of each lithology in the acquired lithology data set, and constructing a lithology adjacency matrix and a lithology adjacency submatrix according to the co-occurrence relationship of all the lithologies.

[0057] Specifically, a lithology data set of an area requiring lithology classification is obtained, the lithology data set including a geological map and a high-resolution optical image of the area, the geological map and the high-resolution optical image are cropped according to a preset size, and a plurality of geological maps of preset sizes and corresponding high-resolution optical images are obtained, and different lithologies on the image are randomly selected according to the high-resolution optical image, for example, 2000 points representing intrusive rocks corresponding to the geological map are extracted, i.e., the lithology category of the lithology, and the spectral band information or texture and other characteristic information corresponding to the high-resolution optical influence can be extracted from each point, i.e., the lithology characteristics corresponding to the lithology, so as to obtain the lithology characteristics (spectrum, texture, etc.) of different lithologies, and at the same time, a lithology adjacency matrix and a lithology adjacency submatrix are constructed according to the co-occurrence relationship of all the extracted lithologies in the geological map. In general, due to The area where lithology analysis is required is relatively large. Usually, the overall geological image of the area where lithology analysis is required is divided into multiple relatively small geological maps, or a pair of geological maps are collected at different locations in the area, and then a complete geological map of the area is constructed through multiple geological maps. Therefore, the lithology category of the entire area can be obtained by analyzing the lithology categories of multiple segmented geological maps, and the corresponding lithology is extracted from each segmented geological map. Through the co-occurrence relationship of different lithologies in all geological maps, a lithology adjacency matrix is ​​constructed, wherein the co-occurrence relationship of lithology can be represented by the correlation between different lithologies in the geological map. For example, if sedimentary rock and volcanic rock appear in a segmented geological map at the same time, it is judged that the two lithologies are associated, and the co-occurrence relationship of the corresponding lithology is extracted through the geological map.

[0058] Furthermore, the constructed lithology adjacency matrix contains the relationship between all lithology categories in the lithology data set, so that by analyzing the relationship and characteristics between different lithology categories in the adjacency matrix, the correlation of different lithology characteristics can be obtained. According to the obtained lithology adjacency matrix, the corresponding lithology adjacency submatrix is ​​obtained by diffusion processing. The lithology adjacency submatrix can help us better understand the common characteristics and connections between lithologies in the lithology adjacency matrix, so as to discover the connection and interaction between different lithology characteristics through analysis and mining.

[0059] Furthermore, the lithology category of a geological map usually refers to the category or property of the rock in the geological map, which can be represented by the texture characteristics, color, structure, grain size and other characteristics of the rock. Common lithology categories include igneous rock, sedimentary rock, metamorphic rock and other major categories, among which each major category also includes several minor categories, such as 11 categories of sedimentary rocks: terrigenous clastic rock, carbonate rock, siliceous rock, phosphate rock, iron rock, manganese rock, aluminous rock, evaporite, acidic pyroclastic rock, neutral pyroclastic rock and base rock. There are three types of volcanic clastic rocks: acidic volcanic lava, neutral volcanic lava and basic volcanic lava; four types of intrusive rocks: acidic intrusive rocks, neutral intrusive rocks, basic intrusive rocks and ultrabasic intrusive rocks; 12 types of metamorphic rocks: mica schist and gneiss, metamorphic (shallow) granulite-quartzite, greenschist-amphibolite, marble, mixed rock, calcium-magnesium silicate rock, magnesian rock, magnetite, metamorphic phosphorite, granulite, blue schist and eclogite.

[0060] S200, based on the graph attention mechanism, obtain the corresponding output lithology feature according to the lithology adjacency matrix and the lithology feature, and obtain the corresponding output lithology sub-feature according to the lithology adjacency sub-matrix and the lithology feature.

[0061] Specifically, the lithology adjacency matrix and lithology features are input into a preset graph attention network (GAT) to output the corresponding output lithology features. Similarly, the lithology adjacency sub-matrix and lithology features are input into another preset graph attention network to output the corresponding output lithology sub-features.

[0062] S300, based on a preset multi-layer perceptron, obtain corresponding graph representations and node representations according to all the output lithology features, and obtain corresponding subgraph representations and subnode representations according to all the output lithology sub-features.

[0063] Specifically, first, the output lithological characteristics are respectively input into two different preset multilayer perceptrons (Multilayer Perceptron, MLP), and the corresponding graph representation and node representation are output respectively. Similarly, the output lithological sub-features are respectively input into two other preset multilayer perceptrons to obtain the corresponding subgraph representation and subnode representation. Among them, the multilayer perceptron is composed of an input layer, several hidden layers and an output layer. Each layer is composed of multiple neurons (neural network nodes), and the neurons between each layer are fully connected. In the multilayer perceptron, each neuron receives the weighted input of the neurons in the previous layer, and performs nonlinear transformation through the activation function, and then passes the output to the neurons in the next layer. By stacking multiple hidden layers, the multilayer perceptron can learn complex nonlinear relationships, thereby outputting the corresponding output results calculated through multiple neural network layers.

[0064] S400, comparing the graph representation and the sub-node representation to obtain corresponding model graph representation and model sub-node representation, and comparing the sub-graph representation and the node representation to obtain corresponding model sub-graph representation and model node representation.

[0065] Specifically, the graph representation is compared with the sub-node representation, and the consistency of the graph representation and the sub-node representation is determined by the comparison result. When the consistency does not meet the preset requirements, the weights in the graph representation and the sub-node representation are updated according to the corresponding comparison results, and finally the graph representation after the weights are updated, i.e., the model graph representation, and the sub-node representation after the weights are updated, i.e., the model sub-node representation, are obtained, so that the consistency of the final model graph representation and the model sub-node representation meets the preset requirements, i.e., the comparison result satisfies the preset conditions. Similarly, the corresponding sub-graph representation and the node representation are compared, and the weights in the sub-graph representation are updated according to the comparison result to obtain the updated sub-graph representation, i.e., the model sub-graph representation. The weights in the node representation are updated according to the comparison result to obtain the updated Node representation, that is, model node representation, so that the consistency of the final model subgraph representation and model node representation meets the preset requirements, that is, the comparison result meets the preset conditions, and the weights of the graph representation, node representation, subgraph representation and subnode representation are updated through consistency, so that the consistency of the final model graph representation, model node representation, model subgraph representation and model subnode representation meets the preset requirements, which can effectively improve the consistency and accuracy of the data processing of the subsequently constructed lithology classification model, better capture the global and local relationship of lithology characteristics in the geological map, reduce the risk of overfitting and improve the model interpretability, and at the same time make the constructed lithology classification model more representative, reliable and interpretable, and improve the performance and practicality of geological lithology classification tasks.

[0066] S500, constructing a lithology classification model by aggregating the model graph representation, the model node representation, the model subgraph representation and the model subnode representation.

[0067] Specifically, Figure 2As shown, according to the model graph representation and model sub-node representation that meet the consistency requirements after the weights are updated, as well as the model sub-graph representation and model node representation, the model graph representation and the model sub-graph representation are aggregated to obtain the aggregate graph representation, and the model node representation and the model sub-node representation are aggregated to obtain the aggregate node representation. The aggregation operation can be achieved through concatenation, addition, weighted average, element-wise multiplication and other methods. Finally, the aggregate graph representation and the aggregate node representation after aggregation are concatenated and integrated to obtain the lithology classification model.

[0068] In this embodiment, the lithology adjacency matrix and lithology adjacency submatrix constructed by all lithologies extracted from the lithology data set can reflect the structural relationship of the complete lithology characteristics of the lithology data set through the lithology adjacency matrix, and the local relationship in the lithology structure can be further reflected through the lithology adjacency submatrix. The lithology adjacency matrix and the lithology adjacency submatrix can more comprehensively reflect all the complete lithology feature information in the lithology data set, so that the co-occurrence relationship and combination law between different lithology categories in the lithology data set can be more clearly displayed, and the connection and interaction between different lithologies can be discovered; the output lithology characteristics and the output lithology sub-features are obtained respectively through the extracted lithology characteristics, the lithology adjacency matrix and the lithology adjacency submatrix, and the corresponding graph representation, subgraph representation, node representation and subnode representation are obtained according to the output lithology characteristics and the output lithology sub-features through a multi-layer perceptron, so that the correlation relationship and the overall structure of different lithology characteristics can be more intuitively presented, so that the relationship and characteristics between the lithology characteristics can be deeply analyzed and compared, while avoiding the separate processing of a large amount of lithology feature data, which is effective. The computational cost of the model is effectively reduced, and the generalization ability and efficiency are improved; the graph representation and the sub-node representation are compared to obtain the model icon and the model node representation that meet the requirements of consistency; the sub-graph representation and the node representation are compared to obtain the model sub-graph representation and the model node representation that meet the requirements of consistency; finally, the lithology classification model is constructed by aggregation through the model graph representation, the model sub-graph representation, the model node representation and the model sub-node representation, so that the lithology classification model can comprehensively utilize the node characteristics and the graph structure information, better understand the characteristics and relationships in the lithology data set, improve the accuracy and efficiency of lithology classification, and can effectively handle complex lithology categories, provide fast and reliable classification results, avoid the large amount of manpower and time consumed by manual field sampling, thereby effectively improving the efficiency of lithology classification and saving the consumption of human resources. At the same time, through the graph structure and the node structure that meet the consistency requirements, the lithology classification model performs lithology classification according to the same classification standard, thereby improving the performance and effect of the lithology classification model, so that it performs better in the lithology classification task of geological maps.

[0069] Optionally, obtaining corresponding output lithological features according to the lithological adjacency matrix and the lithological features includes:

[0070] Determine, according to the lithology adjacency matrix and the lithology feature, a first adjacent lithology feature of the lithology feature in the lithology adjacency matrix;

[0071] Obtaining the corresponding output lithology feature according to the lithology feature and all the corresponding first adjacent lithology features;

[0072] The output lithology characteristics satisfy:

[0073]

[0074] in, is the x-th output lithological feature, For the x-th lithological feature, each lithological feature can obtain a corresponding output lithological feature, is the i-th first adjacent lithological feature, N is the number of the first adjacent lithological features, W is a preset first linear transformation matrix, is the preset first initial weight vector, T is the transposition operation, LeakyReLU is the nonlinear operation, exp is the exponential operation, σ is the nonlinear activation function; || is the concatenation operation.

[0075] Specifically, according to the lithological characteristics of each geological map, the adjacent first adjacent lithological characteristics of the lithological characteristics in the lithological adjacency matrix are obtained, that is, the lithological characteristics associated with the lithological characteristics in the lithological adjacency matrix are taken as the first adjacent lithological characteristics, and the association relationship of each lithological characteristic in the lithological adjacency matrix is ​​determined, and then the correlation between different lithological characteristics is determined through the lithological adjacency matrix, so as to obtain the associated lithological characteristics corresponding to each lithological characteristic, and according to the current lithological characteristics and the corresponding associated first adjacent lithological characteristics, the first attention coefficient corresponding to the lithological characteristics and each first adjacent lithological characteristics in the lithological adjacency matrix is ​​calculated through the preset first attention coefficient relationship.

[0076] Exemplarily, the first attention coefficient relationship satisfies:

[0077]

[0078] Among them, e xi is the xth lithological feature and the adjacent ith lithological feature The corresponding first attention coefficient.

[0079] Furthermore, the first attention coefficients corresponding to all the associated first adjacent lithological features corresponding to the lithological features are calculated in the lithological adjacency matrix, and all the attention coefficients are normalized. Finally, the output lithological features are calculated through a nonlinear activation function based on all the normalized first attention coefficients and all the corresponding adjacent lithological features.

[0080] In this optional embodiment, the related lithological features corresponding to each lithological feature in the adjacency matrix can be obtained through the adjacency matrix and the lithological features, thereby obtaining the correlation between different lithological features. According to the lithological features and all the associated first adjacent lithological features, the output lithological features corresponding to each lithological feature are obtained. The output lithological features can more accurately reflect the correlation between different lithological features, thereby improving the ability to distinguish lithological categories, thereby improving the accuracy of the lithological classification model for lithological classification of geological maps.

[0081] Optionally, obtaining corresponding output lithology sub-features according to the lithology adjacency sub-matrix and the lithology feature includes:

[0082] Determine, according to the lithology adjacency sub-matrix and the lithology feature, a second adjacent lithology feature of the lithology feature in the lithology adjacency sub-matrix;

[0083] According to the lithological feature and all the corresponding second adjacent lithological features, obtaining the corresponding output lithological sub-feature;

[0084] The output lithology sub-characteristics satisfy:

[0085]

[0086] in, is the x-th output lithology sub-feature, is the x-th lithological feature, each lithological feature can obtain a corresponding output lithological sub-feature, so that each lithological feature can finally obtain a corresponding output lithological feature and output lithological sub-feature, respectively. is the jth second adjacent lithological feature, M is the number of the second adjacent lithological features, V is a preset second linear transformation matrix, is the preset second initial weight vector, T is the transposition operation, LeakyReLU is the nonlinear operation, exp is the exponential operation, σ is the nonlinear activation function; || is the concatenation operation.

[0087] Specifically, the relevant second adjacent lithological features are determined in the lithological adjacency submatrix according to each lithological feature, that is, in the lithological adjacency submatrix, the lithological features associated with the lithological feature are found as the second adjacent lithological features of the lithological feature, and then the association relationship of each lithological feature in the lithological adjacency submatrix can be determined, thereby determining the correlation between different lithological features relative to the lithological adjacency submatrix, and according to the corresponding lithological features and the second adjacent lithological features in the lithological adjacency submatrix, the second attention coefficient corresponding to the lithological feature and each second adjacent lithological feature is calculated through the second attention coefficient relationship.

[0088] Exemplarily, the second attention coefficient relationship satisfies:

[0089]

[0090] Among them, f cj is the xth lithological feature The second attention coefficient corresponding to the associated j-th second-neighboring lithological feature.

[0091] Furthermore, all the second attention coefficients corresponding to the lithological features are normalized, and finally the output lithological features are calculated by a nonlinear activation function based on all the normalized second attention coefficients and all the corresponding adjacent lithological features.

[0092] Optionally, the multilayer perceptron includes a first multilayer perceptron and a second multilayer perceptron; and obtaining corresponding graph representations and node representations according to all the output lithological characteristics includes:

[0093] Inputting all the output lithological features after pooling processing into the first multi-layer perceptron, and outputting the corresponding graph representation;

[0094] All the output lithology features are input into the second multi-layer perceptron, and the corresponding node representations are output.

[0095] Specifically, all output lithological features are pooled through a pre-set graph pooling function. The pooling operation can reduce the amount of calculation, extract the main lithological features, and improve the overall performance and generalization ability of the model. All output lithological features after pooling are input into a pre-set first multi-layer perceptron, and the corresponding graph representation is output. At the same time, all unprocessed output lithological features are input into a preset second multi-layer perceptron, and the corresponding node representation is output. Among them, the graph pooling functions include: Max Pooling, in which for each pooling window (usually a 2x2 or 3x3 window), the maximum value in the window is selected as the output value after pooling; Average Pooling, in which for each pooling window, the average value of all elements in the window is calculated as the output value after pooling; L2-Norm Pooling, the squares of the elements in each pooling window are summed and then squared as the output value after pooling; Stochastic Pooling, for each pooling window, an element is randomly selected in the window as the output value; Adaptive Pooling, a pooling function that automatically adjusts the size of the pooling window according to the size of the input feature map. Multilayer Perceptron (MLP) is a common artificial neural network model, which is composed of multiple neurons connected in layers. Each neuron is connected to all neurons in the previous layer, and each connection has a weight. MLP includes an input layer, multiple hidden layers and an output layer, where the hidden layer and the output layer usually contain activation functions. MLP is suitable for many machine learning tasks, such as classification, regression, etc., and can learn complex nonlinear relationships.

[0096] Exemplarily, the graph representation of the first multi-layer perceptron output can be expressed as:

[0097]

[0098] in, is a graph representation, n is the number of output lithological features, is the xth output lithological feature, and W is the preset first linear transformation matrix.

[0099] The node representation of the second multi-layer perceptron output can be expressed as:

[0100] H α =σ((A+I1)XO1);

[0101] Among them, H α is the node representation, A is the lithology adjacency matrix, I1 is the preset unit matrix, X is the set of all lithology characteristics, and O1 is the preset network parameter.

[0102] Optionally, the multilayer perceptron includes a third multilayer perceptron and a fourth multilayer perceptron; and obtaining corresponding subgraph representations and subnode representations according to all the output lithology sub-features includes:

[0103] Inputting all the output lithology sub-features after pooling processing into the third multi-layer perceptron, and outputting the corresponding sub-graph representation;

[0104] All of the output lithology sub-features are input into the fourth multi-layer perceptron, and the corresponding sub-node representations are output.

[0105] Specifically, all output lithology sub-features are pooled through another preset graph pooling function, and all output lithology sub-features after pooling are input into a preset third multi-layer perceptron to output the corresponding sub-graph representation, and all unprocessed output lithology sub-features are input into a preset fourth multi-layer perceptron to output the corresponding sub-node representation, wherein the first multi-layer perceptron shares parameters with the third multi-layer perceptron, and the second multi-layer perceptron shares parameters with the fourth multi-layer perceptron. Through parameter sharing, the multi-layer perceptron can obtain better performance and better generalization ability.

[0106] Exemplarily, the subgraph representation output by the third multi-layer perceptron can be expressed as:

[0107]

[0108] in, is the sub-graph representation, m is the number of output lithology sub-features, is the xth output lithology sub-feature, n is the number of lithology features, V is the preset second linear transformation matrix, and it should be noted that the number of lithology features, the number of output lithology features and the number of output lithology sub-features are all the same, that is, m=n.

[0109] The sub-node representation of the fourth multi-layer perceptron output can be expressed as:

[0110] H β =σ((S+I2)XO2);

[0111] Among them, H β is the child node representation, S is the lithology adjacency submatrix, I2 is the preset unit matrix, X is the set of all lithology characteristics, and O2 is the preset network parameter.

[0112] In this optional embodiment, according to the output lithological characteristics and the output lithological sub-characteristics, corresponding graph representations, node representations, subgraph representations and subnode representations are obtained through preset multiple multi-layer perceptrons. The graph representation and subgraph representation can capture the structural information, global attributes and relationships in the entire lithological data set, which is helpful for analyzing and predicting lithological categories. The node representation and subnode representation can capture the structural information, attribute information and contextual information of the lithological characteristics, which is helpful for better understanding and learning the relationship between lithological characteristics and improving the accuracy of lithological classification.

[0113] Optionally, the comparing the graph representation and the subnode representation to obtain a corresponding model graph representation and model subnode representation includes:

[0114] Obtaining a first comparison result through a preset first discriminator according to the graph representation and the child node representation;

[0115] The weight of the graph representation is updated according to the first comparison result to obtain the corresponding model graph representation, and the sub-node representation is updated according to the first comparison result to obtain the corresponding model sub-node representation.

[0116] Specifically, the graph representation and the sub-node representation are compared through a pre-set first discriminator, and the consistency of the graph representation and the node representation is scored, that is, the first comparison result. According to the score of the first discriminator, it is determined whether the consistency of the graph representation and the sub-node representation meets the preset consistency standard. If not, the respective weight data and preset parameters in the graph representation and the node representation are updated, and re-input into the first discriminator to obtain the first comparison result after the updated weights. The above process is iterated until the first comparison result output by the first discriminator meets the consistency standard. The graph representation corresponding to the consistency score that meets the requirements is used as the model graph representation, and the sub-node representation corresponding to the consistency score that meets the requirements is used as the model sub-node representation.

[0117] Exemplarily, the first discriminator can be expressed as:

[0118]

[0119] is the first comparison result, As shown in the figure, H β is the child node representation, and T is the transpose.

[0120] Optionally, the comparing the subgraph representation and the node representation to obtain a corresponding model subgraph representation and model node representation includes:

[0121] Obtaining a second comparison result through a preset second discriminator according to the subgraph representation and the node representation;

[0122] The weight of the subgraph representation is updated according to the second comparison result to obtain the corresponding model subgraph representation, and the node representation is updated according to the second comparison result to obtain the corresponding model node representation.

[0123] Specifically, the subgraph representation and the node representation are compared through a preset second discriminator, and a corresponding consistency score, i.e., a second comparison result, is obtained. According to the score of the second discriminator, it is determined whether the consistency of the corresponding subgraph representation and the node representation meets the preset consistency standard, i.e., whether the consistency of the subgraph representation and the node representation meets the requirements. If not, the weight information and preset parameter information in the subgraph representation and the node representation are updated respectively, and the consistency score is recalculated through the second discriminator. The above process is iterated until the second comparison result output by the second discriminator meets the requirements, and the current corresponding subgraph representation is determined as the model subgraph representation, and the current node representation is determined as the model node representation.

[0124] Exemplarily, the second discriminator can be expressed as:

[0125]

[0126] is the second comparison result, is a subgraph representation, H α is the node representation and T is the transpose.

[0127] In this optional embodiment, two preset discriminators are used to score the graph representation and sub-node representation, as well as the sub-graph representation and node representation in a consistent manner, thereby updating their respective weights according to the corresponding comparison results, so that the consistency of the final model graph representation, model node representation, model sub-graph representation and model sub-node representation meets the requirements, thereby improving the accuracy of the lithology classification of the final constructed lithology classification model.

[0128] Optionally, constructing a lithology adjacency matrix according to the co-occurrence relationship of all the lithologies includes:

[0129] constructing an initial lithology adjacency matrix based on all of the lithologies;

[0130] Obtaining the lithology adjacency matrix through probability conversion according to the initial lithology adjacency matrix;

[0131] The lithology adjacency submatrix is ​​obtained by diffusion processing according to the lithology adjacency matrix.

[0132] Optionally, the adjacency submatrix satisfies:

[0133] S = exp(tAD -1 -t);

[0134] Where S is the lithology adjacency matrix, t is the preset diffusion time, A is the lithology adjacency matrix, and D -1 is the inverse matrix of the preset diagonal matrix D, and exp is the exponential function.

[0135] Specifically, according to all lithologies extracted from the lithology data set and the co-occurrence relationship of all lithologies in the geological map, that is, the association relationship between different lithologies in the geological map, the data in the preset initial matrix is ​​updated according to the co-occurrence relationship of all lithologies to obtain the initial lithology adjacency matrix. For example, the overall geological map corresponding to the lithology classification area in the lithology data set is divided into three geological maps, namely the first geological map, the second geological map and the third geological map. The co-occurrence relationship between different lithologies is determined according to the category of the lithology corresponding to each geological map. The current lithology data set involves 5 different lithology categories, which are assumed to be C1, C2, C3, C4, and C5. For each geological map, a binary vector can be used to represent the category label of its lithology. The lithology feature of the first geological map is set to t1=(t 11 , t 21 , t 31 , t 41 , t 51 ), when the lithology of the first geological map includes C3 and C5, then t in t1 31 and t 51 All are set to 1, and other elements without corresponding lithology categories are set to 0, and finally the vector t1 of the first geological map is (0, 0, 1, 0, 1). The above method can be used to obtain the vector corresponding to each geological map. Figure 3 The preset initial matrix is ​​a 5×5 matrix, in which each row and column represents 5 different lithology categories, namely C1, C2, C3, C4, and C5. The node values ​​in the initial matrix are all 0. When a geological map meets the lithology category corresponding to the node, the node is updated to 1, and all lithologies in the lithology data set are traversed to obtain an updated initial lithology adjacency matrix. If multiple geological maps meet the lithology category of the same node, the value corresponding to the node is increased by the sum of the number of corresponding geological maps. For example, if there are 3 geological maps that meet the lithology category of a node at the same time, the value of the node is updated to 3. The values ​​of the nodes in the updated initial lithology adjacency matrix are probability converted, and the probability of the diagonal nodes after conversion is 1. According to the non-diagonal nodes, the statistical value of each node is divided by the sum of the values ​​of all non-diagonal nodes to obtain the probability value corresponding to each non-diagonal node. It should be noted that the lithology and lithology category both represent the lithology extracted from the lithology data set.

[0136] Exemplarily, the probability values ​​corresponding to the off-diagonal nodes satisfy:

[0137]

[0138] P ef is the probability value corresponding to the node corresponding to the e-th lithology and the f-th lithology in the updated initial lithology adjacency matrix, that is, the probability that the e-th lithology and the f-th lithology appear at the same time, CCM ef is the value of the node corresponding to the e-th lithology and the f-th lithology in the updated initial lithology adjacency matrix.

[0139] Furthermore, the initial lithology adjacency matrix is ​​updated according to the probability values ​​corresponding to all diagonal lines and off-diagonal lines, and finally the adjacency matrix corresponding to the lithology data set is obtained. The adjacency submatrix corresponding to the adjacency matrix is ​​obtained by performing diffusion processing on the adjacency matrix.

[0140] In this optional embodiment, by constructing an adjacency matrix, nodes with key roles can be identified, which helps to understand the correlation between different lithological characteristics, as well as the importance and influence of different lithological characteristics, and obtain an adjacency submatrix through diffusion of the adjacency matrix. The adjacency submatrix can help discover lithological characteristics that are similar in certain attributes or relationships, promote information exchange and learning between similar nodes, and improve the ability to distinguish related lithologies. The adjacency matrix helps to discover the structural relationship of key nodes, while the adjacency submatrix is ​​more suitable for discovering the similarity between nodes. The comprehensive use of the adjacency matrix and the adjacency submatrix can effectively process and analyze geological map data, and achieve in-depth exploration of the structure and relationship of lithological characteristics, thereby improving the accuracy of the final constructed lithological classification model and data processing efficiency.

[0141] An embodiment of the present invention provides a lithology classification method, comprising:

[0142] Acquiring an optical image to be processed;

[0143] The optical image to be processed is input into a lithology classification model, and a lithology classification result corresponding to the optical image to be processed is output; wherein the lithology classification model is obtained by using the lithology classification model construction method as described above.

[0144] Specifically, an optical image of the area to be lithologically classified, which may be a high-resolution optical image, is obtained and input into a lithological classification model constructed by a lithological classification model construction method, and the lithological category corresponding to the optical image to be processed is output. The lithological classification model is used to quickly perform lithological classification according to the optical image, thereby avoiding the consumption of manpower and resources due to manual lithological classification, effectively improving the accuracy of lithological classification, reducing the input of human resources, and improving the efficiency of lithological classification of geological maps.

[0145] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc. In the present application, the unit described as a separate component may or may not be physically separated, and the component displayed as a unit may or may not be a physical unit, that is, it may be located in one place, or it may be distributed on multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiment of the present invention. In addition, each functional unit in each embodiment of the present invention can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0146] Although the present invention is disclosed as above, the protection scope of the present invention is not limited thereto. Those skilled in the art may make various changes and modifications without departing from the spirit and scope of the present invention, and these changes and modifications will fall within the protection scope of the present invention.

Claims

1. A method for constructing a lithology classification model, characterized in that: include: Extracting the lithological characteristics of each lithology in the acquired lithological data set, and constructing a lithological adjacency matrix and a lithological adjacency submatrix according to the co-occurrence relationship of all the lithologies; Based on the graph attention mechanism, a corresponding output lithology feature is obtained according to the lithology adjacency matrix and the lithology feature, and a corresponding output lithology sub-feature is obtained according to the lithology adjacency sub-matrix and the lithology feature; The step of obtaining the corresponding output lithology characteristics according to the lithology adjacency matrix and the lithology characteristics includes: Determine, according to the lithology adjacency matrix and the lithology feature, a first adjacent lithology feature of the lithology feature in the lithology adjacency matrix; Obtaining the corresponding output lithology feature according to the lithology feature and all the corresponding first adjacent lithology features; The output lithology characteristics satisfy: in, is the x-th output lithological feature, is the xth lithological feature, is the i-th first adjacent lithological feature, N is the number of the first adjacent lithological features, W is a preset first linear transformation matrix, is the preset first initial weight vector, T is the transposition operation, LeakyReLU is the nonlinear operation, exp is the exponential operation, σ is the nonlinear activation function; || is the concatenation operation; Based on a preset multi-layer perceptron, a corresponding graph representation and a node representation are obtained according to all the output lithology features, and a corresponding subgraph representation and a subnode representation are obtained according to all the output lithology sub-features; Compare the graph representation and the subnode representation to obtain a corresponding model graph representation and a model subnode representation, and compare the subgraph representation and the node representation to obtain a corresponding model subgraph representation and a model node representation; The model graph representation, the model node representation, the model subgraph representation and the model subnode representation are aggregated to construct a lithology classification model.

2. The method for constructing a lithology classification model according to claim 1, characterized in that: The step of obtaining the corresponding output lithology sub-features according to the lithology adjacency sub-matrix and the lithology features includes: Determine, according to the lithology adjacency sub-matrix and the lithology feature, a second adjacent lithology feature of the lithology feature in the lithology adjacency sub-matrix; According to the lithological feature and all the corresponding second adjacent lithological features, obtaining the corresponding output lithological sub-feature; The output lithology sub-characteristics satisfy: in, is the x-th output lithology sub-feature, is the xth lithological feature, is the jth second adjacent lithological feature, M is the number of the second adjacent lithological features, V is a preset second linear transformation matrix, is the preset second initial weight vector, T is the transposition operation, LeakyReLU is the nonlinear operation, exp is the exponential operation, σ is the nonlinear activation function; || is the concatenation operation.

3. The method for constructing a lithology classification model according to claim 1, characterized in that: The multilayer perceptron includes a first multilayer perceptron and a second multilayer perceptron; the corresponding graph representation and node representation are obtained according to all the output lithological characteristics, including: Inputting all the output lithological features after pooling processing into the first multi-layer perceptron, and outputting the corresponding graph representation; All the output lithology features are input into the second multi-layer perceptron, and the corresponding node representations are output.

4. The method for constructing a lithology classification model according to claim 1, characterized in that: The multilayer perceptron includes a third multilayer perceptron and a fourth multilayer perceptron; the corresponding subgraph representation and subnode representation are obtained according to all the output lithology sub-features, including: Inputting all the output lithology sub-features after pooling processing into the third multi-layer perceptron, and outputting the corresponding sub-graph representation; All of the output lithology sub-features are input into the fourth multi-layer perceptron, and the corresponding sub-node representations are output.

5. The method for constructing a lithology classification model according to claim 1, characterized in that: The step of comparing the graph representation and the subnode representation to obtain a corresponding model graph representation and model subnode representation includes: Obtaining a first comparison result through a preset first discriminator according to the graph representation and the child node representation; The weight of the graph representation is updated according to the first comparison result to obtain the corresponding model graph representation, and the sub-node representation is updated according to the first comparison result to obtain the corresponding model sub-node representation.

6. The method for constructing a lithology classification model according to claim 1, characterized in that: The step of comparing the subgraph representation and the node representation to obtain a corresponding model subgraph representation and model node representation includes: Obtaining a second comparison result through a preset second discriminator according to the subgraph representation and the node representation; The weight of the subgraph representation is updated according to the second comparison result to obtain the corresponding model subgraph representation, and the node representation is updated according to the second comparison result to obtain the corresponding model node representation.

7. The method for constructing a lithology classification model according to claim 1, characterized in that: The step of constructing a lithology adjacency matrix according to the co-occurrence relationship of all lithologies comprises: constructing an initial lithology adjacency matrix according to the co-occurrence relationship of all the lithologies in the lithology data set; Obtaining the lithology adjacency matrix through probability conversion according to the initial lithology adjacency matrix; The lithology adjacency submatrix is ​​obtained by diffusion processing according to the lithology adjacency matrix.

8. The method for constructing a lithology classification model according to claim 7, characterized in that: The adjacency submatrix satisfies: S=exp(tAD -1 -t); Where S is the lithology adjacency matrix, t is the preset diffusion time, A is the lithology adjacency matrix, D -1 is the inverse matrix of the preset diagonal matrix D.

9. A lithology classification method, characterized in that: include: Acquiring an optical image to be processed; The optical image to be processed is input into a lithology classification model, and a lithology classification result corresponding to the optical image to be processed is output; wherein the lithology classification model is obtained by using the lithology classification model construction method according to any one of claims 1 to 8.

Images