Microsomatic paleontology identification method and device

By using a microbial paleobiological recognition model with local information extraction, image blocking and self-attention mechanism in microbial paleobiological recognition, the problem of time-consuming, labor-intensive and accurate microbial paleobiological recognition in traditional methods is solved, and automated recognition is achieved and efficient identification, suitable for formation division and oil and gas exploration.

CN118918581AInactive Publication Date: 2024-11-08CHENGDU UNIVERSITY OF TECHNOLOGY
View PDF 10 Cites 0 Cited by

Patent Information

Application Number
CN202410887211.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-03
Publication Date
2024-11-08
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The traditional microbial paleobiological identification method relies on manual observation, which is time-consuming and labor-intensive. Due to the wide variety of microbial paleobi species and complex morphology, the consistency and accuracy of the identification results are difficult to ensure, especially in oil and gas exploration and stratigraphic division. .

Method used

A microbial paleobiological recognition method is adopted, and the target microbial paleobiological recognition model is used. This model includes a local information extraction module, an image blocking module, a self-attention module and a classification module. Through convolutional operations, image blocking, and multi-head attention The mechanism and multi-layer perceptron perform image processing, automatically identify the microbial paleontology categories, and output prediction results.

Benefits of technology

Automatic identification of microbial paleontology has been realized, the recognition efficiency and accuracy have been improved, the calculation cost has been reduced, the identification results are more in line with the professional knowledge of geologists, and are suitable for formation division and oil and gas exploration.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118918581A_ABST
    Figure CN118918581A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of oil exploration, and discloses a microbody paleontology recognition method and device, and the method comprises the steps: obtaining a to-be-recognized image; inputting the to-be-recognized image into the target microbody paleontology recognition model, performing automatic recognition, and outputting a target prediction result; according to the method, a local information extraction module captures local information of an image by adopting convolution operation; an input image is decomposed into image blocks through an image partitioning module, each image block is converted into a vector, and a position code is added to each image block; processing each image block by using a multi-head attention mechanism and a multi-layer perceptron through a self-attention module, calculating the relationship between different vectors, capturing the spatial information of the image, improving the nonlinear fitting capability of a model through the multi-layer perceptron, and mapping a feature vector to the probability distribution of a prediction category; and learning information of the self-attention module is summarized through the classification module, and class tokens are introduced to summarize information of all image blocks, so that automatic identification of microsomatic paleontology classes is realized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of petroleum exploration, and in particular to a method and device for identifying micropaleontology. Background Art

[0002] Automatic identification of micropaleontology is of great significance in the fields of oil and gas exploration and paleoenvironmental research. Micropaleontology, such as foraminifera and ostracods, are common tiny fossil components in sedimentary rocks. Their types and distribution are crucial to understanding geological history, paleoecological environment, stratigraphic division, and the formation and distribution of oil and gas reservoirs.

[0003] There are many types of micropaleontology, each with its own specific living environment and era characteristics. Therefore, in paleostratigraphy, specific micropaleontology fossils are often used as markers of geological ages. Their appearance and disappearance can help scientists compare strata and divide ages. In addition, the distribution and abundance of micropaleontology can reflect the conditions of the ancient environment, such as water temperature, salinity, redox conditions, etc. By analyzing these tiny fossils, scientists can reconstruct the environment and climate of the ancient Earth.

[0004] Due to the extremely small size of micropaleontology, traditional identification methods mainly rely on manual observation and classification of scanning electron microscope images of micropaleontology. This method is not only time-consuming and labor-intensive, but also limited by expert experience and subjective judgment, making it difficult to ensure the consistency and accuracy of the identification results. In addition, the variety and complex morphology of micropaleontology pose great challenges to manual identification. At the same time, as the number of samples for oil processing and analysis continues to increase, this further increases the workload of researchers.

[0005] In recent years, with the rapid development of artificial intelligence, machine learning, image processing and computer vision technology, automated microfossil identification has become a feasible and effective new method. By using high-resolution scanning equipment and complex image processing algorithms, microfossils can be automatically identified and classified, which significantly improves work efficiency, reduces human resource consumption, and improves the accuracy and consistency of identification.

[0006] AlexNet is the first deep convolutional neural network to win the ImageNet large-scale visual recognition challenge, marking a new stage for convolutional neural networks in the field of computer vision. After that, ZFnet broke the black box of deep neural networks, visualized the features of convolutional neural network layers, increased the transparency of network internal operations, and made deep convolutional neural networks no longer mysterious. ResNet successfully trained hundreds of layers of networks by introducing a residual learning framework, solving the gradient vanishing problem of deep networks. The Inception network introduced a multi-scale feature extraction mechanism, capturing image features of different scales by parallelizing convolution kernels of different sizes. The recognition accuracy of SEnet (97%) exceeded the recognition accuracy of human eyes (95%) for the first time. After that, complex convolutional neural networks (such as DenseNet, MobileNet) further improved the effect of convolutional neural networks and applied them to complex image data sets (such as JFT-300M). Summary of the invention

[0007] Based on the above, the purpose of the present invention is to propose a method and device for identifying micropaleontology to achieve automatic identification of micropaleontology categories.

[0008] To achieve the above object, the present invention adopts the following technical solutions:

[0009] The first aspect of the present application provides a method for identifying micropaleontology, which comprises:

[0010] Acquiring an image to be identified, wherein the image to be identified shows micro-paleontology;

[0011] The image to be identified is input into the target micropaleontology identification model, the micropaleontology category in the image to be identified is automatically identified, and the target prediction result is output;

[0012] Among them, the target micropaleontology identification model includes a local information extraction module, an image segmentation module, a self-attention module, and a classification module, so that the local information extraction module uses convolution operation to capture the local information of the image; the image segmentation module decomposes the input image into image blocks of fixed size, and each image block is converted into a vector form, and a position code is added to each image block; the self-attention module uses a multi-head attention mechanism and a multi-layer perceptron including a feedforward neural network to process each image block converted into a vector form, calculate the relationship between different vectors to capture the spatial information of the image, and improve the nonlinear fitting ability of the model through the multi-layer perceptron and map the feature vector to the probability distribution of the predicted category; the classification module summarizes the learning information of the self-attention module, and introduces class tokens to summarize the information of all image blocks to generate target prediction results.

[0013] In some embodiments, before placing the image to be identified into the target micropaleontology identification model, the method further includes training and obtaining the target micropaleontology identification model, wherein the training and obtaining the target micropaleontology identification model includes:

[0014] Take photos to collect micropaleontology in the target stratum and obtain micropaleontology image data;

[0015] The micropaleontological image data are classified, cleaned and normalized in turn to achieve data standardization;

[0016] The standardized micropaleontological image data are divided into training data set, validation data set and test data set according to the preset ratio, and the MicroViT model is constructed as the initial micropaleontological identification model;

[0017] Based on the training data set and the validation data set, the initial micropaleontology identification model is optimized using a stochastic gradient descent method, and iterated until a stable and accurate optimized micropaleontology identification model is obtained;

[0018] Using the optimized micropaleontology recognition model to perform recognition prediction on the micropaleontology image data in the test data set to obtain recognition prediction results;

[0019] The identification prediction results are compared with the labeled micropaleontology of the test data set for numerical analysis, and the optimized micropaleontology identification model is visualized and analyzed. The GrradCAM algorithm is used to visualize the category heat map to determine the reliability of the optimized micropaleontology identification model, so as to obtain the target micropaleontology identification model.

[0020] In some embodiments, the target strata include the Lulehe Formation, the lower section of the Lower Ganchaigou Formation, the upper section of the Lower Ganchaigou Formation, the Upper Ganchaigou Formation, the Lower Youshashan Formation, the Upper Youshashan Formation, and the Shizigou Formation in the Qaidam Basin; and / or,

[0021] The target stratigraphic micropaleontology includes Microhylidae, Orthohylidae, Qaidamylidae, Banmeishidae, Qaidamylidae, Nanxingshidae, and Microhylidae; and / or,

[0022] The preset ratio is 7:2:1; and / or,

[0023] The normalization process is implemented by using a normalization formula, which is:

[0024]

[0025] In the formula, norm is the value after standardization, x iis the image pixel value, max(x) is the maximum value of the image pixel, and min(x) is the minimum value of the image pixel.

[0026] In some embodiments, the sequential classification of micropaleontological image data includes labeling micropaleontological objects and storing the same micropaleontological objects in different categories; and / or,

[0027] Data cleaning includes removing damaged micropaleontological image data and blurred micropaleontological image data.

[0028] In some embodiments, the optimized micropaleontology recognition model includes 2 local information extraction modules, 5 self-attention modules and 1 classification module; and / or,

[0029] The optimizer used in the step of optimizing the initial micropaleontology identification model using the stochastic gradient descent method is Adam, and the loss function is the cross entropy loss function.

[0030] In some embodiments, the initial micropaleontology identification model is optimized using a stochastic gradient descent method based on the training data set and the validation data set, and iterated until a stable and accurate optimized micropaleontology identification model is obtained, comprising:

[0031] The micropaleontological images of the training data set are input into the preliminary micropaleontological automatic identification model for M iterations, and the cross entropy loss function is used to optimize the loss each time so that the loss value is continuously close to the true value. The final accuracy and loss are continuously iterated until they tend to be stable, so as to obtain a stable and accurate optimized micropaleontological identification model, M≥≥25.

[0032] In some embodiments, the standardized micropaleontological image data is divided into a training data set, a validation data set, and a test data set according to a preset ratio, and a MicroViT model is constructed as an initial micropaleontological identification model, including:

[0033] The number of image blocks is preset by the image block module, and the image of the test data set is evenly divided into multiple image blocks according to the condition of equal size, each image block is converted into a one-dimensional vector, and a position code is added;

[0034] Input all image blocks into the self-attention module, repeatedly iterate and learn the association relationship between the image blocks, the self-attention module mainly includes a multi-head attention mechanism and a multi-layer perceptron, and obtain the operation result;

[0035] The calculation results are put into the classification module to obtain the prediction results, and the initial micropaleontology automatic identification model is obtained after analyzing the prediction results.

[0036] In some embodiments, the optimized micropaleontological identification model is visualized and analyzed, and a GradCAM algorithm is used to visualize a category heat map to determine the reliability of the optimized micropaleontological identification model to obtain the target micropaleontological identification model, including:

[0037] The images of the test data set are placed in the optimized micropaleontology recognition model for forward propagation, the feature map of the multi-layer perceptron is calculated, and the prediction scores of the categories of interest of the images of the test data set are obtained;

[0038] Calculate the gradient average of the feature map to obtain the weight of the feature map for category prediction;

[0039] Multiply the weights by the feature map to get the original category heat map;

[0040] Apply the ReLU activation function to the original category heat map to obtain features that have a positive contribution to the predicted category, and obtain the target category heat map;

[0041] After combining the target category heat map to determine whether the optimized micropaleontology identification model is consistent with the professional knowledge of domain experts, the target micropaleontology automatic identification model is obtained.

[0042] In some embodiments, the operation process of the target micropaleontology identification model includes:

[0043] Put the image to be recognized into the image extraction module;

[0044] The image processed by the image extraction module is divided into N equal-sized and non-overlapping image blocks, and then the image blocks are expanded into one-dimensional vectors and position codes are added;

[0045] The image block is then placed in a self-attention module, which includes a multi-head attention mechanism and a multi-layer perceptron. The multi-head attention mechanism enables each image block to obtain global information, and then the result of the multi-head attention mechanism is placed in a multi-layer perceptron. The multi-layer perceptron includes two fully connected feedforward neural networks, and a nonlinear activation function is inserted between the two fully connected feedforward neural networks. Layer normalization is required before passing through the multi-head attention mechanism and the multi-layer perceptron.

[0046] After the multi-head attention mechanism and multi-layer perceptron, residual link operation is performed;

[0047] The class token is introduced into the image block through the classification module, and is learned together with the image block after learning by the self-attention module. The classification module includes a multi-head attention module and a multi-layer perceptron module. Layer normalization is performed before each multi-head attention module and multi-layer perceptron module, and a residual connection is added after each multi-head attention module and multi-layer perceptron module. The class token is used to learn the information of all image blocks, make a comprehensive judgment, and finally obtain the classification effect.

[0048] The second aspect of the present application provides a micro-paleontological identification device, which comprises:

[0049] at least one processor; and,

[0050] a memory communicatively connected to the at least one processor; wherein,

[0051] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the micropaleontology identification method of any of the above-mentioned embodiments.

[0052] The beneficial effects of the present invention are:

[0053] The present invention obtains an image to be identified; then, the image to be identified is input into a target micropaleontology identification model, the micropaleontology category in the image to be identified is automatically identified, and a target prediction result is output. The target micropaleontology identification model includes a local information extraction module, an image segmentation module, a self-attention module, a multi-layer perceptron, and a classification module, so that the local information of the image is captured by the local information extraction module using convolution operation; the input image is decomposed into image blocks of fixed size by the image segmentation module, and each image block is converted into a vector form, and a position code is added to each image block; each image block converted into a vector form is processed by the self-attention module using a multi-head attention mechanism and a feedforward neural network, and the relationship between different vectors is calculated to capture the spatial information of the image; the nonlinear fitting ability of the model is improved by the multi-layer perceptron and the feature vector is mapped to the probability distribution of the prediction category; the learning information of the self-attention module is summarized by the classification module, and the class token is introduced to summarize the information of all image blocks to generate a target prediction result.

[0054] The present invention is used to realize automatic identification of micropaleontology. In the target micropaleontology identification model, the self-attention module can enhance the global information collection ability of the image and simulate the characteristics of human eye observation to the greatest extent. The classification module is used to integrate the previous operation results, and the integrated learned information is finally used for intelligent identification of micropaleontology categories. It not only well identifies the representative micropaleontology categories of different strata, which is far superior to the existing technical solutions, but also is more conducive to the development of related work such as stratum division and oil and gas exploration; and the model structure parameters are small, the degree of automation is high, and the calculation cost is effectively reduced. The identification results are more in line with the professional knowledge of geologists. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the contents of the embodiments of the present invention and these drawings without paying any creative work.

[0056] Figure 1 A schematic diagram of the steps of the micropaleontological identification method provided for the embodiment of the present application;

[0057] Figure 2 A schematic diagram of the steps for training and obtaining a target micropaleontological identification model provided for an embodiment of the present application;

[0058] Figure 3 The MicroViT model and its module schematic diagram used in the micropaleontology identification method provided in the embodiment of the present application; wherein, Figure 3 (a) is the overall architecture diagram of the MicroViT model. Figure 3 (b) is the composition diagram of the self-attention module;

[0059] Figure 4 The result of automatic identification of micropaleontology by the micropaleontology identification method provided in the embodiment of the present application;

[0060] Figure 5 The micropaleontology identification method provided in the embodiment of the present application provides the results of automatic visualization of micropaleontology identification. DETAILED DESCRIPTION

[0061] In order to make the technical problems solved by the present invention, the technical solutions adopted and the technical effects achieved clearer, the technical solutions of the embodiments of the present invention will be further described in detail below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0062] In the description of the present invention, unless otherwise clearly specified and limited, the terms "connected", "connected", and "fixed" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0063] In the present invention, unless otherwise clearly stipulated and limited, a first feature being "above" or "below" a second feature may include that the first and second features are in direct contact, or may include that the first and second features are not in direct contact but are in contact through another feature between them. Moreover, a first feature being "above", "above" and "above" a second feature includes that the first feature is directly above and obliquely above the second feature, or simply indicates that the first feature is higher in level than the second feature. A first feature being "below", "below" and "below" a second feature includes that the first feature is directly below and obliquely below the second feature, or simply indicates that the first feature is lower in level than the second feature.

[0064] In the description of this embodiment, the terms "upper", "lower", "left", "right", etc., and the orientation or position relationship are based on the orientation or position relationship shown in the drawings, and are only for the convenience of description and simplified operation, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first" and "second" are only used to distinguish in the description and have no special meaning.

[0065] Micropaleontological images are images of microfossils obtained through high-resolution microscopic imaging technology. These images are used to study the morphology, structure and evolution of micropaleontology in geological history. Different categories of micropaleontology have obvious differences in appearance. For example, the appearance of pollen is usually spherical or elliptical, while charophytes have complex cell structures, usually spiral or radial structures; some features of micropaleontology of the same genus are also different. For example, both types of charophytes have a spiral appearance structure, but the cell wall size and surface texture are different. How to capture the overall and local differences in micropaleontological images is the difficulty in identifying micropaleontology. In addition, it is also crucial to verify whether the model pays attention to the characteristics of different micropaleontology and whether these characteristics are consistent with the expert knowledge of paleontologists.

[0066] In recent years, the Transformer based on the self-attention mechanism has achieved great success in the field of natural language processing. This mechanism enables the model to take into account the global dependencies between words when processing text, so as to better understand the contextual meaning of the text. The Transformer model performs well in a variety of natural language processing tasks, such as text classification, machine translation, question-answering systems, text generation, etc., and demonstrates strong generalization capabilities. Inspired by the Transformer and convolutional neural networks, an embodiment of the present invention provides a method and device for identifying micropaleontology, which is used to automatically identify micropaleontology categories.

[0067] Specifically, in the related art, convolutional neural network is one of the main methods of image recognition. The common convolution kernel size is often smaller than the size of the input image. When performing convolution operations, the convolution kernel can capture the local information of the image very well, but the small-scale feature extraction will cause the convolutional neural network to be insufficient in processing long-distance dependencies or global information, and it is difficult to capture the remote dependencies in the image. Although the convolutional neural network will capture certain global information as the network depth continues to deepen, the information loss is relatively large. The Transformer image recognition model can solve the problem that the convolutional neural network is difficult to capture the global information of the image to a certain extent. The model divides the original image into image blocks of the same size, allowing different image blocks to learn from each other. This can avoid the problem of only extracting local information. Although the Transformer shows unique advantages in the extraction of global information, it is slightly inferior to the convolutional neural network in the extraction of local information. The micro-paleontological identification method provided by the embodiment of the present application, wherein the target micro-paleontological identification model combines the advantages of the convolutional neural network and the Transformer model, and realizes the intelligent identification of micro-paleontological.

[0068] The following is a detailed introduction, refer to Figure 1-5The present application provides a method for identifying micropaleontology, which comprises:

[0069] Step 100, obtaining an image to be identified, wherein the image to be identified shows micro-paleontology;

[0070] Step 200, input the image to be identified into the target micropaleontology identification model, automatically identify the type of micropaleontology in the image to be identified, and output the target prediction result;

[0071] Among them, the target micropaleontology identification model includes a local information extraction module, an image segmentation module, a self-attention module, and a classification module, so that the local information extraction module uses convolution operation to capture the local information of the image; the image segmentation module decomposes the input image into image blocks of fixed size, and each image block is converted into a vector form, and a position code is added to each image block; the self-attention module uses a multi-head attention mechanism and a multi-layer perceptron including a feedforward neural network to process each image block converted into a vector form, calculate the relationship between different vectors to capture the spatial information of the image, and improve the nonlinear fitting ability of the model through the multi-layer perceptron and map the feature vector to the probability distribution of the predicted category; the classification module summarizes the learning information of the self-attention module, and introduces class tokens to summarize the information of all image blocks to generate target prediction results.

[0072] The implementation method of the present application is used to realize automatic identification of micropaleontology. In the target micropaleontology identification model, the self-attention module can enhance the global information collection ability of the image and simulate the characteristics of human eye observation to the greatest extent. The classification module is used to integrate the previous operation results, and the integrated learned information is finally used for intelligent identification of micropaleontology categories. It not only well identifies the representative micropaleontology categories of different strata, which is far superior to the existing technical solutions, but also is more conducive to the development of related work such as stratum division and oil and gas exploration; and the model structure parameters are small, the degree of automation is high, and the calculation cost is effectively reduced. The identification results are more in line with the professional knowledge of geologists.

[0073] refer to Figure 1-5 In some embodiments, before placing the image to be identified into the target micropaleontology identification model, the method further includes step 000, training and obtaining the target micropaleontology identification model, wherein step 000, training and obtaining the target micropaleontology identification model includes:

[0074] Step 010, taking photos to collect micropaleontology in the target stratum and obtaining micropaleontology image data;

[0075] Step 020, classify the micropaleontological image data in turn, perform data cleaning and normalization processing to achieve data standardization;

[0076] Step 030, dividing the standardized micropaleontological image data into a training data set, a validation data set and a test data set according to a preset ratio, and constructing a MicroViT model as an initial micropaleontological identification model; specifically, the preset ratio may be 7:2:1; specifically, the MicrooViT model may be written in a Pytorch framework (version 1.12.0);

[0077] Step 040, based on the training data set and the validation data set, optimizing the initial micropaleontology recognition model using a stochastic gradient descent method, and iterating until a stable and accurate optimized micropaleontology recognition model is obtained;

[0078] Step 050, using the optimized micropaleontology recognition model to perform recognition prediction on the micropaleontology image data in the test data set to obtain recognition prediction results;

[0079] Step 060, perform numerical analysis on the identification prediction results compared with the labeled micropaleontology of the test data set, and perform visual analysis on the optimized micropaleontology identification model, use the GradCAM algorithm to visualize the category heat map, and determine the reliability of the optimized micropaleontology identification model to obtain the target micropaleontology identification model.

[0080] refer to Figure 1-5 In some embodiments, the target strata include the Lulehe Formation, the lower section of the Lower Ganchaigou Formation, the upper section of the Lower Ganchaigou Formation, the Upper Ganchaigou Formation, the Lower Youshashan Formation, the Upper Youshashan Formation, and the Shizigou Formation in the Qaidam Basin. Figure 1-5 In some embodiments, the target stratum micropaleontology includes Microlake Cyperus, Orthostar Cyperus, Qaidam Cyperus, Semi-Meistar Cyperus, Chaibeigou Cyperus, Nanxing Cyperus, and Microhyalopsis, and thus the embodiments of the present application can efficiently, accurately, and quickly predict the micropaleontology of different strata in the Qaidam Basin. Specifically, the images of micropaleontology in the Qaidam Basin can be collected by means of manual scanning electron microscopy photography.

[0081] refer to Figure 1-5 In some embodiments, the sequential classification of micropaleontological image data includes labeling the micropaleontological objects and storing the same micropaleontological objects in different categories.

[0082] refer to Figure 1-5 In some embodiments, data cleaning includes removing damaged micropaleontological image data and blurred micropaleontological image data.

[0083] refer to Figure 1-5 In some implementations, the normalization process is implemented using a normalization formula, which is:

[0084]

[0085] In the formula, norm is the value after standardization, x i is the image pixel value, max(x) is the maximum value of the image pixel, and min(x) is the minimum value of the image pixel.

[0086] refer to Figure 1-5 In some embodiments, the optimized micropaleontology identification model includes 2 local information extraction modules, 5 self-attention modules and 1 classification module.

[0087] refer to Figure 1-5 In some embodiments, the optimizer used in the step of optimizing the initial micropaleontology identification model using the stochastic gradient descent method is Adam, and the loss function is the cross entropy loss function.

[0088] refer to Figure 1-5 In some embodiments, based on the training data set and the validation data set, the initial micropaleontology identification model is optimized using a stochastic gradient descent method, and iterated until a stable and accurate optimized micropaleontology identification model is obtained, comprising:

[0089] The micropaleontological images of the training data set are input into the preliminary micropaleontological identification model for M iterations, and the cross entropy loss function is used to optimize the loss each time so that the loss value is continuously close to the true value. The final accuracy and loss are continuously iterated until they tend to be stable, so as to obtain a stable and accurate optimized micropaleontological identification model, M≥≥25.

[0090] In addition, for ease of understanding, Figure 3 The overall architecture of the MicroViT model is demonstrated, specifically showing the core components of the MicroViT model, namely the self-attention module and the multi-layer perceptron. Figure 4 The results of the automatic identification of micropaleontology using the micropaleontology identification method are shown. Figure 4 (a) It can be seen that after 25-30 iterations, the accuracy and loss are stable. In the training data set, the accuracy can reach 93% and the loss can be reduced to 0.15. The confusion matrix also confirms the high recognition accuracy. Specifically, Figure 4 The dotted line in (a) is the iteration number-loss value relationship curve, and the solid line is the iteration number-accuracy relationship curve.

[0091] refer to Figure 1-5 In some embodiments, the step 030, dividing the standardized micropaleontological image data into a training data set, a validation data set, and a test data set according to a preset ratio, and constructing a MicroViT model as an initial micropaleontological identification model, comprises:

[0092] Step 031, preset the number of image blocks through the image block module, evenly divide the image of the test data set into multiple image blocks according to the condition of equal size, convert each image block into a one-dimensional vector, and add a position code;

[0093] Step 032: input all image blocks into the self-attention module, repeatedly iterate and learn the association relationship between the image blocks, the self-attention module mainly includes a multi-head attention mechanism and a multi-layer perceptron, and obtain the calculation result;

[0094] Step 033, putting the calculation results into the classification module to obtain the prediction results, and after analyzing the prediction results, obtaining the initial micropaleontology automatic identification model.

[0095] refer to Figure 1-5 In some embodiments, in step 060, the optimized micropaleontological identification model is visualized and analyzed, and a GradCAM algorithm is used to visualize a category heat map to determine the reliability of the optimized micropaleontological identification model, so as to obtain the target micropaleontological identification model, including:

[0096] Step 061, placing the image of the test data set into the optimized micro-paleontology automatic identification model for forward propagation, calculating the feature map of the multi-layer perceptron and obtaining the prediction score of the category of interest of the image of the test data set;

[0097] Step 062: Calculate the gradient average of the feature map to obtain the weight of the feature map for category prediction;

[0098] Step 063, multiply the weight and the feature map to obtain the original category heat map;

[0099] Step 064, applying the ReLU activation function to the original category heat map to obtain features that have a positive contribution to the predicted category, and obtain a target category heat map;

[0100] Step 065: After judging whether the preliminary micropaleontology recognition model is consistent with the professional knowledge of domain experts in combination with the target category heat map, the target micropaleontology automatic recognition model is obtained.

[0101] Explanatory, it can be expressed through the formula, as follows:

[0102]

[0103] Among them, A k Represents the feature map of the multi-layer perceptron, y c represents the prediction score of category c, Represents the weight of the multi-layer perceptron feature map, and Z represents the feature map Ak The total number of elements, L C Represents the target feature heatmap of category c.

[0104] refer to Figure 5 , Figure 5 The micropaleontology identification method provided in the embodiment of the present application is a visualization result of the automatic identification of micropaleontology. Among them, column a represents the micropaleontology picture, column b represents the visualization result of the corresponding micropaleontology picture, and the depth of the color represents the importance of the area for the network to make the final classification decision. The darker the color (for example, the closer to red), the greater the contribution to the model's prediction. The model pays more attention to these areas when making decisions. On the contrary, the lighter the color (for example, the closer to blue), the less influence the area has on the prediction.

[0105] refer to Figure 1-5 In some embodiments, the operation process of the target micropaleontology identification model in step 200 includes:

[0106] Step 210: putting the image to be identified into the image extraction module;

[0107] Explanatory, assume that the image to be recognized is x∈R H×W×C , H, W, C are the height, width and number of channels of the image respectively, then

[0108] Where O(i, j) is the value of the output feature map at position (i, j). I(i·s+mp, j·s+np) is the value of the image at (i·s+mp, j·s+np), K(m, n) is the value of the convolution kernel at position (m, n), and kH and kw are the height and width of the convolution kernel, respectively.

[0109] Step 220, dividing the image processed by the image extraction module into N equal-sized and non-overlapping image blocks, then expanding the image blocks into one-dimensional vectors, and adding position codes;

[0110] Explanatory, X′ e =Concat([x p1 ;x p2 ;…;x pN ])*E;

[0111] X e =X′ e +x pos

[0112] In the formula, x pi is the i-th image block; the Concat function is used to concatenate all image blocks together according to the specified dimension; E represents the learning matrix, which is used for continuous iteration of the model; xpos Refers to position encoding, which can embed position information into the feature vector of the sequence and adaptively learn the embedding parameters of the position through the training process. This encoding method enables the model to better capture the local details and global structure in the image and achieve better performance.

[0113] Step 230: Put the image block into a self-attention module, which includes a multi-head attention mechanism and a multi-layer perceptron; wherein, each image block obtains global information through the multi-head attention mechanism, and then puts the result of the multi-head attention mechanism into the multi-layer perceptron, which includes two fully connected feedforward neural networks, and a nonlinear activation function is inserted between the two fully connected feedforward neural networks; wherein, layer normalization is required before passing through the multi-head attention mechanism and the multi-layer perceptron;

[0114] Step 240, performing residual link operation after passing through the multi-head attention mechanism and the multi-layer perceptron;

[0115] Explanation: The core of the multi-head attention mechanism is to repeatedly learn through the Q, K, and V matrices so that each image block obtains global information. The Q matrix is ​​used to query the relationship between the current image block and other image blocks, and at the same time indicates which information the current image block needs to pay attention to. The K matrix is ​​used to calculate the similarity between the query image block to determine which information needs to be paid attention to. The V matrix calculates the specific information of Q and K and is used to calculate the target self-attention output. The Q, K, and V matrices are used to learn the relationship between the image block and the W q , W k , W v Multiply them respectively to get, where W q , W k , W v is a learnable matrix. This also gives it the ability to learn iteratively. The target attention result operation is to multiply the Q matrix by the transposed matrix of K and divide it by a constant d represents the dimension of the vector. Finally, it is multiplied with the V matrix and softmax normalized. Then the result of the multi-head attention mechanism is put into a multi-layer perceptron. The multi-layer perceptron is usually composed of two fully connected feedforward neural networks, and a nonlinear activation function (such as ReLU, GeLU) is inserted between the two fully connected feedforward neural networks. Before passing through the multi-head attention mechanism and the multi-layer perceptron, layer normalization is required, and after passing through the multi-head attention mechanism and the multi-layer perceptron, residual link operation is performed:

[0116] Q, K, V = X e (W q , W k , W v )

[0117] X e=LN(X e )

[0118]

[0119] X e+1 =LN(MHSA(Q, K, V))

[0120] Step 250, introducing the class token into the image block through the classification module, and learning it together with the image block after learning by the self-attention module, the classification module includes a multi-head attention module and a multi-layer perceptron module, layer normalization is performed before each multi-head attention module and multi-layer perceptron module, and a residual connection is added after each multi-head attention module and multi-layer perceptron module. The class token is used to learn the information of all image blocks, make a comprehensive judgment, and finally obtain the classification effect.

[0121] In summary, the micropaleontology identification method combines the self-attention neural network algorithm to identify micropaleontology. It not only has a simple model, low cost, and high computational efficiency, but also achieves accurate, efficient, and universal automatic identification of micropaleontology. Compared with the existing technology, it combines modern artificial intelligence image technology, has obvious technical progress, outstanding substantive characteristics, and significant progress.

[0122] In addition, some embodiments of the present application provide a micropaleontology identification device, which includes at least one processor and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the micropaleontology identification method of any of the above-mentioned embodiments.

[0123] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A method for identifying micropaleontology, characterized in that: include: Acquiring an image to be identified, wherein the image to be identified shows micro-paleontology; The image to be identified is input into the target micropaleontology identification model, the micropaleontology category in the image to be identified is automatically identified, and the target prediction result is output; Among them, the target micropaleontology identification model includes a local information extraction module, an image segmentation module, a self-attention module, and a classification module, so that the local information extraction module uses convolution operation to capture the local information of the image; the image segmentation module decomposes the input image into image blocks of fixed size, and each image block is converted into a vector form, and a position code is added to each image block; the self-attention module uses a multi-head attention mechanism and a multi-layer perceptron including a feedforward neural network to process each image block converted into a vector form, calculate the relationship between different vectors to capture the spatial information of the image, and improve the nonlinear fitting ability of the model through the multi-layer perceptron and map the feature vector to the probability distribution of the predicted category; the classification module summarizes the learning information of the self-attention module, and introduces class tokens to summarize the information of all image blocks to generate target prediction results.

2. The micropaleontological identification method according to claim 1, characterized in that: Before placing the image to be identified into the target micropaleontology identification model, the method further includes training and obtaining the target micropaleontology identification model, wherein the training and obtaining the target micropaleontology identification model includes: Take photos to collect micropaleontology in the target stratum and obtain micropaleontology image data; The micropaleontological image data are classified, cleaned and normalized in turn to achieve data standardization; The standardized micropaleontological image data are divided into training data set, validation data set and test data set according to the preset ratio, and the MicroViT model is constructed as the initial micropaleontological identification model; Based on the training data set and the validation data set, the initial micropaleontology identification model is optimized using a stochastic gradient descent method, and iterated until a stable and accurate optimized micropaleontology identification model is obtained; Using the optimized micropaleontology recognition model to perform recognition prediction on the micropaleontology image data in the test data set to obtain recognition prediction results; The identification prediction results are compared with the labeled micropaleontology of the test data set for numerical analysis, and the optimized micropaleontology identification model is visualized and analyzed. The GradCAM algorithm is used to visualize the category heat map to determine the reliability of the optimized micropaleontology identification model, so as to obtain the target micropaleontology identification model.

3. The micropaleontological identification method according to claim 2, characterized in that: The target strata include the Lulehe Formation, the lower section of the Lower Ganchaigou Formation, the upper section of the Lower Ganchaigou Formation, the Upper Ganchaigou Formation, the Lower Youshashan Formation, the Upper Youshashan Formation, and the Shizigou Formation in the Qaidam Basin; and / or, The target stratigraphic micropaleontology includes Microhylidae, Orthohylidae, Qaidamylidae, Banmeishidae, Qaidamylidae, Nanxingshidae, and Microhylidae; and / or, The preset ratio is 7:2:1; and / or, The normalization process is implemented by using a normalization formula, which is: In the formula, norm is the value after standardization, x i is the image pixel value, max(x) is the maximum value of the image pixel, and min(x) is the minimum value of the image pixel.

4. The micropaleontological identification method according to claim 2, characterized in that: The sequential classification of micropaleontological image data includes labeling micropaleontological objects and storing the same micropaleontological objects in different categories; and / or, Data cleaning includes removing damaged micropaleontological image data and blurred micropaleontological image data.

5. The micropaleontological identification method according to claim 2, characterized in that: The optimized micropaleontology recognition model includes 2 local information extraction modules, 5 self-attention modules and 1 classification module; and / or, The optimizer used in the step of optimizing the initial micropaleontology identification model using the stochastic gradient descent method is Adam, and the loss function is the cross entropy loss function.

6. The micropaleontological identification method according to claim 5, characterized in that: The method of optimizing the initial micropaleontology identification model based on the training data set and the validation data set using a stochastic gradient descent method, and iterating until a stable and accurate optimized micropaleontology identification model is obtained, comprises: The micropaleontological images of the training data set are input into the preliminary micropaleontological automatic identification model for M iterations, and the cross entropy loss function is used to optimize the loss each time so that the loss value is continuously close to the true value. The final accuracy and loss are continuously iterated until they tend to be stable, so as to obtain a stable and accurate optimized micropaleontological identification model, M≥25.

7. The micropaleontological identification method according to claim 2, characterized in that: The standardized micropaleontological image data are divided into a training data set, a verification data set and a test data set according to a preset ratio, and a MicroViT model is constructed as an initial micropaleontological identification model, including: The number of image blocks is preset by the image block module, and the image of the test data set is evenly divided into multiple image blocks according to the condition of equal size, each image block is converted into a one-dimensional vector, and a position code is added; Input all image blocks into the self-attention module, repeatedly iterate and learn the association relationship between the image blocks, the self-attention module mainly includes a multi-head attention mechanism and a multi-layer perceptron, and obtain the operation result; The calculation results are put into the classification module to obtain the prediction results, and the initial micropaleontology automatic identification model is obtained after analyzing the prediction results.

8. The micropaleontological identification method according to claim 2, characterized in that: The method of visually analyzing the optimized micropaleontological identification model, visualizing the category heat map using the GradCAM algorithm, and judging the reliability of the optimized micropaleontological identification model to obtain the target micropaleontological identification model includes: The images of the test data set are placed in the optimized micropaleontology recognition model for forward propagation, the feature map of the multi-layer perceptron is calculated, and the prediction scores of the categories of interest of the images of the test data set are obtained; Calculate the gradient average of the feature map to obtain the weight of the feature map for category prediction; Multiply the weights by the feature map to get the original category heat map; Apply the ReLU activation function to the original category heat map to obtain features that have a positive contribution to the predicted category, and obtain the target category heat map; After combining the target category heat map to determine whether the optimized micropaleontology identification model is consistent with the professional knowledge of domain experts, the target micropaleontology automatic identification model is obtained.

9. The micropaleontological identification method according to any one of claims 1 to 8, characterized in that: The operation process of the target micropaleontology identification model includes: Put the image to be recognized into the image extraction module; The image processed by the image extraction module is divided into N equal-sized and non-overlapping image blocks, and then the image blocks are expanded into one-dimensional vectors and position codes are added; The image block is then placed in a self-attention module, which includes a multi-head attention mechanism and a multi-layer perceptron. The multi-head attention mechanism enables each image block to obtain global information, and then the result of the multi-head attention mechanism is placed in a multi-layer perceptron. The multi-layer perceptron includes two fully connected feedforward neural networks, and a nonlinear activation function is inserted between the two fully connected feedforward neural networks. Layer normalization is required before passing through the multi-head attention mechanism and the multi-layer perceptron. After the multi-head attention mechanism and multi-layer perceptron, residual link operation is performed; The class token is introduced into the image block through the classification module, and is learned together with the image block after learning by the self-attention module. The classification module includes a multi-head attention module and a multi-layer perceptron module. Layer normalization is performed before each multi-head attention module and multi-layer perceptron module, and a residual connection is added after each multi-head attention module and multi-layer perceptron module. The class token is used to learn the information of all image blocks, make a comprehensive judgment, and finally obtain the classification effect.

10. A micro-paleontological identification device, characterized in that: include: at least one processor; as well as, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the micropaleontology identification method described in any one of claims 1 to 9.

Citation Information

Patent Citations

  • Micro paleontology fossil image detection, classification and discovery method, system and application

    CN113128335A

  • Single-sample and small-sample microbody paleontology fossil image identification method and system

    CN114399763A

  • Railway scene unusual invasion target identification method based on small sample learning

    CN115393802A

  • Lithology identification method and system, computer equipment and storage medium

    CN116935115A

  • Remote sensing image classification method based on visual Transform

    CN117746141A