Rock slice mineral type identification method and related device

Abstract

The invention discloses a rock slice mineral type identification method and a related device, and relates to the technical field of computer vision identification, and the method comprises the steps: carrying out the image registration and image segmentation of an orthogonal polarized image and a single polarized image of a rock slice, and obtaining an orthogonal partial photon image and a single partial photon image corresponding to the spatial position of each mineral in the rock slice, taking the orthogonal partial photon image and the single partial photon image corresponding to the spatial position of the mineral as input for each mineral in the rock slice, and utilizing the trained mineral type identification model to determine the type of the mineral. And rock slice mineral type identification is realized. According to the method, the feature fusion module is fused in the feature extraction module of the deep learning model, and the feature fusion module is used for fusing the features of the orthogonal polarization photon image and the features of the single polarization photon image to obtain the trained mineral type recognition model to complete rock slice mineral type recognition, so that the recognition precision can be improved.

Description

Technical Field

[0001] The present application relates to the field of computer vision recognition technology, and in particular to a rock thin section mineral type recognition method and related devices based on a deep learning algorithm. Background Art

[0002] Mineral composition is the basic element of lithology type and the basis for describing the evolution of mineralization environment and conducting oil and gas exploration. Rapid and accurate high-quality identification of mineral types has an important impact on improving the mining efficiency and performance of mineral resources.

[0003] At present, the traditional methods of mineral type identification are mainly divided into laboratory analysis and rock thin section identification. Laboratory analysis mainly relies on geophysical or chemical means (such as X-ray diffraction technique (XRD), X-ray fluorescence analysis method (XRF), hyperspectral scanning technology, scanning electron microscope (SEM)-energy dispersive X-ray fluorescence spectrometer (EDX), comprehensive automatic mineral petrology detection method (Quantitative Evaluation of Minerals by Scanning electron microscopy, QEMSCAN, i.e. scanning electron microscope mineral quantitative evaluation), mineral dissociation analyzer (Mineral Liberation Analyser, MLA), electron probe, Raman spectroscopy analysis method) to obtain elemental information, mineral proportion and spectral information of minerals. By constructing linear or nonlinear models for the element content in different types of mineral samples, mineral type identification is carried out in the study area. Although laboratory analysis methods can quantify mineral components and determine mineral types, some experimental instruments need to grind mineral samples during implementation, which destroys the structure of mineral samples. In this way, mixed information of a single point position of the mineral sample is obtained, and the mineral properties of all points within the spatial range of the mineral sample cannot be obtained. In addition, some experimental instruments are expensive and have high requirements for the working environment and instrument accuracy. The testing cost of newly collected mineral samples is high. The rock thin section identification method mainly uses different polarization angles to observe the particle shape and optical properties of mineral particles in rock thin sections under a polarizing microscope, and combines the crystal optical characteristics and polarization characteristics of the mineral to identify the structure, texture and mineral composition of the mineral and determine the mineral category. The rock thin section identification method can observe the entire microscopic morphological distribution, protrusion state, interference color, etc. of the mineral in the rock thin section. Although the spatial distribution of minerals can be obtained, it requires a high level of knowledge and experience for the identification personnel. It takes a lot of effort to rotate the microscope to observe the mineral characteristics under different polarization angles. At the same time, under the cutting angle of the rock thin section, when some mineral characteristics are not obvious, different rock and mineral identification personnel also have different conclusions on the same mineral. When faced with large quantities of rock thin sections, it is difficult to achieve rapid and high-quality mineral type identification.

[0004] Traditional methods have played an important role in the implementation of engineering applications in geology, and have achieved a series of results, providing key references for geological mineral prediction, oil and gas exploration, and mineral deposit environmental evaluation. However, in the process of mineral type identification, researchers are required to have high experience. Improving the efficiency of large-scale mineral type identification, reducing experimental testing costs and the workload of geologists, and enhancing the objectivity of experimental results are of great research significance for geological engineering applications and intelligent prospecting. With the rapid development of computer image technology, image recognition based on machine learning and deep learning technology is becoming more and more widely used in geological images, which can solve the problems existing in the above traditional methods.

[0005] Image processing methods based on machine learning usually process raw data into a one-dimensional data set, failing to pay attention to the spatial structural characteristics of the sample, resulting in low recognition efficiency. Image processing methods based on deep learning perform convolutional coding by integrating the multi-scale features of orthogonal polarized images and single polarized images, and build a model to distinguish different mineral types, thereby achieving recognition of different minerals. However, in terms of image fusion, the features extracted from orthogonal polarized images by convolutional neural networks and the features extracted from single polarized images by convolutional neural networks are spliced, and the joint representation of mineral features by two different data sources is not considered, resulting in low recognition accuracy. Summary of the invention

[0006] The purpose of this application is to provide a rock thin section mineral type identification method and related devices, which can improve the identification accuracy.

[0007] To achieve the above objectives, this application provides the following solutions:

[0008] In a first aspect, the present application provides a method for identifying mineral types in rock thin sections, the method comprising:

[0009] Obtain orthogonal polarization images and single polarization images of rock thin sections;

[0010] Performing image registration and image segmentation on the orthogonal polarization image and the single polarization image to obtain an orthogonal polarization sub-image and a single polarization sub-image corresponding to the spatial position of each mineral in the rock slice;

[0011] For each mineral in the rock thin section, the orthogonal polarized photon image and the single polarized photon image corresponding to the spatial position of the mineral are taken as input, and the type of the mineral is determined using a trained mineral type recognition model to achieve rock thin section mineral type recognition; wherein the trained mineral type recognition model is a model obtained by integrating a feature fusion module into the feature extraction module of the deep learning model, and the feature fusion module is used to fuse the features of the orthogonal polarized photon image and the features of the single polarized photon image.

[0012] Optionally, performing image registration and image segmentation on the orthogonal polarization image and the single polarization image to obtain an orthogonal polarization sub-image and a single polarization sub-image corresponding to the spatial position of each mineral in the rock slice specifically includes:

[0013] Taking the orthogonal polarization image as a reference, the single polarization image is moved and rotated to obtain a converted single polarization image, so as to perform image registration on the orthogonal polarization image and the single polarization image, so that the spatial positions of the same mineral in the converted single polarization image and the orthogonal polarization image correspond to each other;

[0014] The orthogonal polarization image and the converted single polarization image are respectively segmented to obtain an orthogonal polarization sub-image and a single polarization sub-image corresponding to the spatial position of each mineral in the rock slice.

[0015] Optionally, the deep learning model is an EfficientNet-B0 model, which includes a first branch, a second branch, and a third branch, wherein the output end of the first branch and the output end of the second branch are both connected to the input end of the third branch; the first branch and the second branch have the same structure, the input of the first branch is an orthogonal polarized sub-image, and the input of the second branch is a single polarized sub-image, the first branch includes a first stage, a second stage, a third stage, a fourth stage, and a fifth stage connected in sequence; the third branch includes a sixth stage, a seventh stage, an eighth stage, and a ninth stage connected in sequence;

[0016] The feature fusion module includes a first feature fusion unit, a second feature fusion unit and a third feature fusion unit; at this time, in the trained mineral type recognition model, the first input end of the first feature fusion unit is connected to the third stage of the first branch, the second input end is connected to the third stage of the second branch, the first output end is connected to the fourth stage of the first branch, and the second output end is connected to the fourth stage of the second branch; the first input end of the second feature fusion unit is connected to the fourth stage of the first branch, the second input end is connected to the fourth stage of the second branch, the first output end is connected to the fifth stage of the first branch, and the second output end is connected to the fifth stage of the second branch; the first input end of the third feature fusion unit is connected to the fifth stage of the first branch, the second input end is connected to the fifth stage of the second branch, and the output end is connected to the sixth stage of the third branch.

[0017] Optionally, the first stage includes sequentially connected convolution layers, batch normalization layers, and activation function layers; the second stage, the third stage, the fourth stage, the fifth stage, the sixth stage, the seventh stage, and the eighth stage all include a mobile flip bottleneck convolution module; the ninth stage includes sequentially connected convolution layers, batch normalization layers, activation function layers, average pooling layers, and classification layers, and the classification layer is a fully connected layer.

[0018] Optionally, the first feature fusion unit and the second feature fusion unit have the same structure, the first feature fusion unit includes a weight calculation subunit and a weighted feature calculation subunit connected in sequence; the third feature fusion unit includes a weight calculation subunit, a weighted feature calculation subunit and a feature fusion calculation subunit connected in sequence;

[0019] The calculation formula of the weight calculation subunit is as follows:

[0020]

[0021] Among them, Attention is the weight; Sigmoid is the activation function layer; Conv 1×1 is a 1×1 convolutional layer; f ortho is the feature of the orthogonal polarization sub-image input to the weight calculation sub-unit; f single is the feature of the single polarization photon image input to the weight calculation subunit;

[0022] The calculation formula of the weighted feature calculation subunit is:

[0023]

[0024] in, is the weighted feature of the orthogonal polarized photon image; is the weighted feature of the single polarization photon image;

[0025] The calculation formula of the feature fusion calculation subunit is:

[0026]

[0027] Among them, f fuse Conv is the fusion feature; 3×3 is a 3×3 convolutional layer; Concat is a concatenation layer.

[0028] Optionally, before using the orthogonal polarization image and the single polarization image corresponding to the spatial position of the mineral as input and using a trained mineral type recognition model to determine the type of the mineral, the rock thin section mineral type recognition method further includes:

[0029] Acquire a data set; the data set includes sample features and sample labels of each sample mineral in a plurality of sample rock thin sections, the sample features are sample orthogonal polarization images and sample single polarization images corresponding to the spatial positions of the sample minerals, and the sample labels are the types of the sample minerals;

[0030] Performing data standardization processing on the data set to obtain a standardized data set;

[0031] Randomly dividing the standardized data set to obtain a training set, a validation set, and a test set;

[0032] Performing data enhancement processing on the training set to obtain an enhanced training set; the data enhancement processing includes: rotation, brightness transformation and adding noise;

[0033] The enhanced training set, the validation set and the test set are used as inputs to train the initial mineral type recognition model to obtain a trained mineral type recognition model.

[0034] In a second aspect, the present application provides a rock thin section mineral type identification device, the rock thin section mineral type identification device comprising:

[0035] An image acquisition module, used to acquire orthogonal polarization images and single polarization images of rock thin sections;

[0036] An image processing module, used for performing image registration and image segmentation on the orthogonal polarization image and the single polarization image, to obtain an orthogonal polarization sub-image and a single polarization sub-image corresponding to the spatial position of each mineral in the rock slice;

[0037] A mineral type prediction module is used to determine the type of each mineral in a rock thin section by using an orthogonal polarized photon image and a single polarized photon image corresponding to the spatial position of the mineral as input, and to realize rock thin section mineral type identification by using a trained mineral type identification model; wherein the trained mineral type identification model is a model obtained by integrating a feature fusion module into a feature extraction module of a deep learning model, and the feature fusion module is used to fuse the features of the orthogonal polarized photon image and the features of the single polarized photon image.

[0038] In a third aspect, the present application provides a computer device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any of the rock thin section mineral type identification methods described above.

[0039] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the rock thin section mineral type identification methods described above.

[0040] In a fifth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements any of the rock thin section mineral type identification methods described above.

[0041] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0042] The present application provides a method and related device for identifying the mineral type of a rock thin section, which performs image registration and image segmentation on the orthogonal polarized image and the single polarized image of the rock thin section, obtains the orthogonal polarized sub-image and the single polarized sub-image corresponding to the spatial position of each mineral in the rock thin section, integrates a feature fusion module into the feature extraction module of the deep learning model, and the feature fusion module is used to fuse the features of the orthogonal polarized sub-image and the features of the single polarized sub-image to obtain a trained mineral type recognition model. For each mineral in the rock thin section, the orthogonal polarized sub-image and the single polarized sub-image corresponding to the spatial position of the mineral are used as input, and the type of the mineral is determined using the trained mineral type recognition model to achieve rock thin section mineral type recognition. The present application comprehensively considers the features of the orthogonal polarized sub-image and the single polarized sub-image, and performs feature fusion in the feature extraction stage of the orthogonal polarized sub-image and the single polarized sub-image to improve the recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0044] Figure 1 This is an application environment diagram of a rock thin section mineral type identification method provided in Example 1 of the present application.

[0045] Figure 2 A schematic flow chart of a method for identifying mineral types in rock thin sections provided in Example 1 of the present application.

[0046] Figure 3 A schematic diagram of the principle of a method for identifying mineral types in rock thin sections provided in Example 1 of the present application.

[0047] Figure 4 Schematic diagram of the network structure of the trained mineral type recognition model provided in Example 1 of the present application.

[0048] Figure 5 This is a schematic diagram of the network structure of the image fusion part provided in Example 1 of the present application.

[0049] Figure 6 Schematic diagram of the orthogonal polarization image of the amphibole sample provided in Example 1 of the present application.

[0050] Figure 7 Schematic diagram of a single polarized photon image of a sample of hornblende provided in Example 1 of the present application.

[0051] Figure 8 Schematic diagram of the orthogonal polarization image of the biotite sample provided in Example 1 of the present application.

[0052] Fig. 9 Schematic diagram of a single polarized photon image of a biotite sample provided in Example 1 of the present application.

[0053] Fig.10 Schematic diagram of the orthogonal polarization image of the calcite sample provided in Example 1 of the present application.

[0054] Fig.11 Schematic diagram of a single polarized photon image of a calcite sample provided in Example 1 of the present application.

[0055] Fig.12 Schematic diagram of the orthogonal polarization image of the chlorite sample provided in Example 1 of the present application.

[0056] Fig.13 Schematic diagram of a single polarized photon image of a chlorite sample provided in Example 1 of the present application.

[0057] Fig.14 Schematic diagram of the orthogonal polarization image of the dolomite sample provided in Example 1 of the present application.

[0058] Fig.15 Schematic diagram of a single polarized photon image of a dolomite sample provided in Example 1 of the present application.

[0059] Fig.16 This is a schematic diagram of an orthogonal polarization image of an ooid sample provided in Example 1 of the present application.

[0060] Fig.17 A schematic diagram of a single polarized photon image of an ooid sample provided in Example 1 of the present application.

[0061] Fig.18 A schematic diagram of a sample orthogonal polarization image of garnet provided in Example 1 of the present application.

[0062] Fig.19 A schematic diagram of a single polarized photon image of a garnet sample provided in Example 1 of the present application.

[0063] Fig. 20 This is a schematic diagram of an orthogonal polarization image of a sample of potassium feldspar provided in Example 1 of the present application.

[0064] Fig.21 Schematic diagram of a single polarized photon image of a sample of potassium feldspar provided in Example 1 of the present application.

[0065] Fig. 22 Schematic diagram of the orthogonal polarization image of the muscovite sample provided in Example 1 of the present application.

[0066] Fig.23 Schematic diagram of a single polarized photon image of a sample of muscovite provided in Example 1 of the present application.

[0067] Fig.24 Schematic diagram of the orthogonal polarization image of the plagioclase sample provided in Example 1 of the present application.

[0068] Fig.25 Schematic diagram of a single polarized photon image of a plagioclase sample provided in Example 1 of the present application.

[0069] Fig.26 This is a schematic diagram of an orthogonal polarization image of a quartz sample provided in Example 1 of the present application.

[0070] Fig. 27 A schematic diagram of a single polarized photon image of a quartz sample provided in Example 1 of the present application.

[0071] Fig.28 Schematic diagram of the orthogonal polarization image of the spodumene sample provided in Example 1 of the present application.

[0072] Fig.29 Schematic diagram of a single polarized photon image of a spodumene sample provided in Example 1 of the present application.

[0073] Fig.30 A schematic diagram of an orthogonal polarization image of a serpentine sample provided in Example 1 of the present application.

[0074] Fig.31 A schematic diagram of a single polarized photon image of a serpentine sample provided in Example 1 of the present application.

[0075] Fig.32 A schematic diagram of the functional modules of a rock thin section mineral type identification device provided in Example 2 of the present application.

[0076] Fig.33 A schematic diagram of the structure of a computer device provided in Example 3 of the present application. DETAILED DESCRIPTION

[0077] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0078] Example 1

[0079] The rock thin section mineral type identification method provided in the embodiment of the present application can be applied to Figure 1In the application environment shown. Among them, the terminal communicates with the server through the network. The data storage system can store the data that the server needs to process. The data storage system can be set up separately, integrated on the server, or placed on the cloud or other servers. The terminal can send the orthogonal polarization image and the single polarization image of the rock slice to be processed to the server. After the server receives the orthogonal polarization image and the single polarization image of the rock slice to be processed, the server performs image registration and image segmentation on the orthogonal polarization image and the single polarization image of the rock slice to be processed, and obtains the orthogonal polarization sub-image and the single polarization sub-image corresponding to the spatial position of each mineral in the rock slice; for each mineral in the rock slice, the orthogonal polarization sub-image and the single polarization sub-image corresponding to the spatial position of the mineral are used as input, and the type of the mineral is determined using the trained mineral type recognition model to realize the rock slice mineral type recognition. The server can feedback the obtained mineral type prediction result to the terminal.

[0080] In addition, in some embodiments, the rock thin section mineral type identification method can also be implemented independently by a server or a terminal. For example, the terminal can directly process the orthogonal polarization image and the single polarization image of the rock thin section to be processed, or the server can obtain the orthogonal polarization image and the single polarization image of the rock thin section to be processed from the data storage system, and process the orthogonal polarization image and the single polarization image of the rock thin section to be processed.

[0081] The terminals may be, but are not limited to, various desktop computers, laptops, smart phones, tablet computers, IoT devices and portable wearable devices. IoT devices may be smart speakers, smart TVs, smart air conditioners, smart car-mounted devices, etc. Portable wearable devices may be smart watches, smart bracelets, head-mounted devices, etc. The server may be implemented as an independent server or a server cluster consisting of multiple servers, or a cloud server.

[0082] like Figure 2 and Figure 3 As shown, a method for identifying mineral types in rock slices is provided. The method is executed by a computer device, and can be executed by a computer device such as a terminal or a server alone, or by a terminal and a server together. In the embodiment of the present application, the method is applied to Figure 1 The server in the example is used to illustrate the following steps:

[0083] Step S1, obtaining an orthogonal polarization image and a single polarization image of a rock slice.

[0084] Step S2, performing image registration and image segmentation on the orthogonal polarization image and the single polarization image to obtain an orthogonal polarization sub-image and a single polarization sub-image corresponding to the spatial position of each mineral in the rock slice.

[0085] Step S3, for each mineral in the rock thin section, the orthogonal polarization image and the single polarization image corresponding to the spatial position of the mineral are used as input, and the type of the mineral is determined by using a trained mineral type recognition model to realize the rock thin section mineral type recognition; wherein the trained mineral type recognition model is a model obtained after integrating a feature fusion module into the feature extraction module of the deep learning model, and the feature fusion module is used to fuse the features of the orthogonal polarization image and the features of the single polarization image.

[0086] By implementing the above-mentioned steps S1 to S3, this embodiment provides a method for intelligently identifying the mineral types of rock thin sections based on deep learning and feature fusion algorithms, which relates to the technical field of computer vision recognition of rock thin section image data processing. A feature fusion module is integrated into the feature extraction module of the deep learning model to obtain a trained mineral type recognition model. The feature fusion module is used to fuse the features of the orthogonal polarized photon image and the features of the single polarized photon image, so that the features of the two can be fused during the feature extraction process of the orthogonal polarized photon image and the single polarized photon image. Subsequently, for each mineral in the rock thin section, the orthogonal polarized photon image and the single polarized photon image corresponding to the spatial position of the mineral are used as input, and the type of the mineral is determined using the trained mineral type recognition model to realize rock thin section mineral type recognition, so that the image feature fusion technology can be carried out in combination with the different polarization features of the mineral image to improve the recognition accuracy of the mineral type.

[0087] This embodiment uses a polarizing microscope to obtain an orthogonal polarizing image and a single polarizing image of a rock slice. The orthogonal polarizing image and the single polarizing image collected by the polarizing microscope in the same field of view both include a number of minerals, and the spatial positions of the minerals may be different. Based on this, this embodiment introduces image registration and image segmentation. Image registration is to correct the positions of the two images in order to solve the problem that the spatial positions of the minerals in the orthogonal polarizing image and the single polarizing image collected by the polarizing microscope in the same field of view are different. Because the effect of orthogonal polarization in the mineral differentiation process is better than that of single polarization, the orthogonal polarizing image is used as a reference. By shifting The single polarization image is positionally corrected by moving and rotating operations, and the final output is an orthogonal polarization image and a single polarization image that correspond to the spatial position of the same mineral. Image segmentation is to obtain a sub-image containing only one mineral. The collected orthogonal polarization image and the single polarization image after position correction are input, and the image is cropped into sub-images of 224*224 in a non-overlapping manner to screen out the sub-image containing minerals. Specifically, the orthogonal polarization image and the single polarization image are image registered and segmented to obtain the orthogonal polarization sub-image and the single polarization sub-image corresponding to the spatial position of each mineral in the rock thin section.

[0088] Among them, the orthogonal polarization image and the single polarization image are imaged and segmented to obtain the orthogonal polarization sub-image and the single polarization sub-image corresponding to the spatial position of each mineral in the rock slice, specifically including:

[0089] (1) Using the orthogonal polarization image as a reference, the single polarization image is moved and rotated to obtain a converted single polarization image, so as to perform image registration on the orthogonal polarization image and the single polarization image, so that the spatial positions of the same mineral in the converted single polarization image and the orthogonal polarization image correspond to each other.

[0090] (2) Perform image segmentation on the orthogonal polarized image and the converted single polarized image respectively to obtain the orthogonal polarized photon image and single polarized photon image corresponding to the spatial position of each mineral in the rock slice.

[0091] In this embodiment, the deep learning model can adopt any model, and specifically adopts EfficientNet-B0 as the basic framework, such as Figure 4 As shown, EfficientNet-B0 includes 9 stages, and Stage 1 includes a convolution layer with a convolution kernel of size 3×3 (i.e. Figure 4 Conv in ), 1 BatchNorm layer (batch normalization layer, i.e. Figure 4 BN in ) and 1 Sigmoid activation function layer (i.e. Figure 4Stage 2-Stage 8 are 7 MBConv (mobile inverted bottleneck convolution) modules with different structures and combined with attention mechanism (Squeeze Excitation, SE), Stage 9 includes 1 convolution layer with a convolution kernel of 1×1 size, 1 BatchNorm layer, 1 Sigmoid activation function layer, 1 average pooling layer (i.e. Figure 4 Avgpool in ) and 1 classification layer (i.e. Figure 4 The classification layer is a fully connected layer. Each MBConv module contains an attention mechanism layer.

[0092] In this embodiment, the deep learning model may be an EfficientNet-B0 model, which includes a first stage (i.e., Stage 1), a second stage (i.e., Stage 2), a third stage (i.e., Stage 3), a fourth stage (i.e., Stage 4), a fifth stage (i.e., Stage 5), a sixth stage (i.e., Stage 6), a seventh stage (i.e., Stage 7), an eighth stage (i.e., Stage 8), and a ninth stage (i.e., Stage 9), wherein the first stage, the second stage, the third stage, the fourth stage, and the fifth stage are all two, and the sixth stage, the seventh stage, the eighth stage, and the ninth stage are all one. Specifically, the EfficientNet-B0 model includes a first branch, a second branch, and a third branch, and the output end of the first branch and the output end of the second branch are both connected to the input end of the third branch. The first branch and the second branch have the same structure. The input of the first branch is an orthogonal polarized photon image, and the input of the second branch is a single polarized photon image. The first branch includes the first stage, the second stage, the third stage, the fourth stage, and the fifth stage connected in sequence. The second branch includes the first stage, the second stage, the third stage, the fourth stage, and the fifth stage connected in sequence. The third branch includes the sixth stage, the seventh stage, the eighth stage, and the ninth stage connected in sequence. Among them, the first stage includes a convolution layer, a batch normalization layer, and an activation function layer connected in sequence. The second stage, the third stage, the fourth stage, the fifth stage, the sixth stage, the seventh stage, and the eighth stage all include a moving flip bottleneck convolution module. The ninth stage includes a convolution layer, a batch normalization layer, an activation function layer, an average pooling layer, and a classification layer connected in sequence. The classification layer is a fully connected layer.

[0093] This embodiment further introduces a feature fusion module into the EfficientNet-B0 model. The feature fusion module includes a first feature fusion unit, a second feature fusion unit, and a third feature fusion unit. Figure 4 and Figure 5As shown, at this time, in the trained mineral type recognition model, the first input end of the first feature fusion unit is connected to the third stage of the first branch, the second input end is connected to the third stage of the second branch, the first output end is connected to the fourth stage of the first branch, and the second output end is connected to the fourth stage of the second branch. The first input end of the second feature fusion unit is connected to the fourth stage of the first branch, the second input end is connected to the fourth stage of the second branch, the first output end is connected to the fifth stage of the first branch, and the second output end is connected to the fifth stage of the second branch. The first input end of the third feature fusion unit is connected to the fifth stage of the first branch, the second input end is connected to the fifth stage of the second branch, and the output end is connected to the sixth stage of the third branch.

[0094] It should be noted that, at this time, in the trained mineral type recognition model, in the first branch, the third stage and the fourth stage are no longer connected, and the fourth stage and the fifth stage are no longer connected. In the second branch, the third stage and the fourth stage are no longer connected, and the fourth stage and the fifth stage are no longer connected. In the third branch, the sixth stage is no longer connected to the output end of the first branch and the output end of the second branch.

[0095] When fusing image features, this embodiment adopts a feature weighted fusion algorithm to adaptively fuse the features of the orthogonal polarization sub-image and the features of the single polarization sub-image. In order to integrate the features of the two polarization sub-images, a first feature fusion unit, a second feature fusion unit and a third feature fusion unit are specifically designed. The first feature fusion unit and the second feature fusion unit have the same structure. The first feature fusion unit includes a weight calculation subunit and a weighted feature calculation subunit connected in sequence. The second feature fusion unit includes a weight calculation subunit and a weighted feature calculation subunit connected in sequence. The third feature fusion unit includes a weight calculation subunit, a weighted feature calculation subunit and a feature fusion calculation subunit connected in sequence.

[0096] The weight calculation subunit is used to calculate the feature weights of the two images. The calculation formula is as follows:

[0097]

[0098] In formula (1), Attention is the weight; Sigmoid is the activation function layer; Conv 1×1 is a 1×1 convolutional layer; f ortho is the feature of the orthogonal polarization sub-image input to the weight calculation sub-unit; f single is the feature of the single polarization photon image input to the weight calculation subunit.

[0099] The weighted feature calculation subunit is used to calculate the weighted features of the orthogonal polarization sub-image and the features of the single polarization sub-image, and its calculation formula is:

[0100]

[0101]

[0102] In the above formula, is the weighted feature of the orthogonal polarized photon image; is the weighted feature of the single polarization photon image.

[0103] The feature fusion calculation subunit is used to calculate the fusion feature, and its calculation formula is:

[0104]

[0105] In formula (4), f fuse Conv is the fusion feature; 3×3 is a 3×3 convolutional layer; Concat is a concatenation layer.

[0106] The working process of the trained mineral type recognition model is as follows: first, the first stage, second stage and third stage in the first branch are used to extract the features of the orthogonal polarized sub-image, and the feature output size is batchsize×28 (length)×28 (width)×40 (number of channels). The first stage, second stage and third stage in the second branch are used to extract the features of the single polarized sub-image, and the feature output size is batchsize×28×28×40. The first feature fusion unit is used to carry out weighted feature calculation (i.e., formula (1)-formula (3)), and the weighted features of the orthogonal polarized sub-image and the weighted features of the single polarized sub-image are obtained respectively, and the feature output size is batchsize× 28×28×40; secondly, the weighted features of the orthogonal polarized sub-image are input into the fourth stage of the first branch, and the convolution operation is performed using the fourth stage of the first branch to extract the features of the orthogonal polarized sub-image again, and the feature output size is batchsize×14×14×80. The weighted features of the single polarized sub-image are input into the fourth stage of the second branch, and the convolution operation is performed using the fourth stage of the second branch to extract the features of the single polarized sub-image again, and the feature output size is batchsize×14×14×80. The weighted feature calculation is performed using the second feature fusion unit (i.e., formula (1)-formula (3)), and the weighted features of the orthogonal polarized sub-image are obtained respectively. and the weighted features of the single polarized photon image, the feature output size is batchsize×14×14×80; then, the weighted features of the extracted orthogonal polarized photon image are input into the fifth stage of the first branch, and the convolution operation is performed using the fifth stage of the first branch to extract the features of the orthogonal polarized photon image again, and the feature output size is batchsize×14×14×112. The weighted features of the extracted single polarized photon image are input into the fifth stage of the second branch, and the convolution operation is performed using the fifth stage of the second branch to extract the features of the single polarized photon image again, and the feature output size is batchsize×14×14×112. The third feature fusion is used The unit carries out weighted feature calculation (i.e., formula (1)-formula (4)) to obtain fused features, and the feature output size is batchsize×14×14×112; then, the extracted fused features are input into the sixth stage of the third branch for convolution operation, and the feature output size is batchsize×7×7×192, and then input into the seventh stage of the third branch for convolution operation, and the feature output size is batchsize×7×7×320, and then input into the eighth stage of the third branch for convolution operation, and the feature output size is batchsize×7×7×1280, completing the feature extraction process. Finally, the ninth stage of the third branch is used for mineral type identification.

[0107] In this embodiment, Stage 3 to Stage 6 of the EfficientNet-B0 model are selected to extract features of orthogonal polarized photon images and single polarized photon images respectively. This is because this stage is the middle stage of the network structure and the extracted image semantic feature information is rich, but it is also the stage where the most spatial information is lost. Therefore, it is considered to embed feature fusion into this stage and make full use of image feature fusion technology to calculate effective weights of feature maps from different data sources for fusion and suppress irrelevant information.

[0108] Before using the orthogonal polarized photon image and the single polarized photon image corresponding to the spatial position of the mineral as input and using the trained mineral type recognition model to determine the type of the mineral, the rock thin section mineral type recognition method of this embodiment further includes:

[0109] (1) Obtain a data set, where the data set includes sample features and sample labels of each sample mineral in multiple sample rock thin sections, where the sample features are sample orthogonal polarization images and sample single polarization images corresponding to the spatial positions of the sample minerals, and the sample labels are the types of the sample minerals.

[0110] The present embodiment pre-acquires a sample orthogonal polarization image and a sample single polarization image of each sample rock slice in a plurality of sample rock slices, performs image registration and image segmentation on the sample orthogonal polarization images and the sample single polarization images, obtains a sample orthogonal polarization sub-image and a sample single polarization sub-image corresponding to the spatial position of each sample mineral in the sample rock slice, combines the mineral feature differentiation knowledge of rock and mineral identification, organizes the sub-images containing a mineral into corresponding mineral categories, collects mineral images containing different optical features, morphologies, and spatial positions, increases the model's feature mining of different minerals, and improves the accuracy of mineral classification.

[0111] (2) Perform data standardization on the data set to obtain a standardized data set.

[0112] In this embodiment, sample orthogonal polarization images and sample single polarization images of 13 sample minerals are collected, which are: 1771 hornblende, 1933 biotite, 2033 calcite, 1983 chlorite, 2006 dolomite, 1751 oolite, 1986 garnet, 2000 potassium feldspar, 1974 muscovite, 2076 plagioclase, 2116 quartz, 2197 spodumene, and 2116 serpentine. Figure 6-Figure 31As shown in the figure, 5 schematic diagrams are given respectively. It should be noted that these numbers are the numbers of sample orthogonal polarization images and sample single polarization images, for example, 1771 amphibole samples refer to the collection of 1771 sample orthogonal polarization images and 1771 sample single polarization images of amphibole, and the calculation of the mean and variance of the sample orthogonal polarization images and the mean and variance of the sample single polarization images, and the calculation of the mean and variance of the pixel values ​​of each sample orthogonal polarization image of each sample mineral in the three bands of red, green and blue, respectively. The mean and variance of the sample orthogonal polarization image are obtained, and then the mean and variance of the sample orthogonal polarization image are used to perform data standardization processing on all sample orthogonal polarization images, such as maximum-minimum normalization, and the mean and variance of the pixel values ​​of each sample single polarization image of each sample mineral in the red, green and blue bands are calculated respectively to obtain the mean and variance of the sample single polarization image, and then the mean and variance of the sample single polarization image are used to perform data standardization processing on all sample single polarization images, such as maximum-minimum normalization. As an example, the means of the sample orthogonal polarized photon images in the red, green and blue bands are 0.408, 0.416 and 0.434, and the variances in the red, green and blue bands are 0.128, 0.134 and 0.125, respectively. The means of the sample single polarized photon images in the red, green and blue bands are 0.579, 0.565 and 0.558, and the variances in the red, green and blue bands are 0.112, 0.121 and 0.110, respectively. The calculation of the mean and variance is mainly to balance the mineral image features at different spatial positions and to speed up the model convergence during the training process.

[0113] (3) Randomly divide the standardized data set into training set, validation set and test set.

[0114] The standardized data set is randomly divided into training set, test set and validation set in the ratio of 6:3:1. The division results are shown in Table 1, where the sample orthogonal polarization images and sample single polarization images in the training set, test set and validation set are corresponding, that is, the sample orthogonal polarization images and sample single polarization images corresponding to the minerals in the same spatial position are used to obtain the orthogonal polarization features and single polarization features of the same mineral when extracting features.

[0115] Table 1 Division results

[0116]

[0117]

[0118] (4) Performing data enhancement processing on the training set to obtain an enhanced training set. The data enhancement processing includes: rotation, brightness transformation and adding noise.

[0119] In this embodiment, the training set can be rotated 90 degrees, rotated 180 degrees, brightness transformed, and data enhancement processing such as adding Gaussian noise can be performed.

[0120] (5) The enhanced training set, validation set and test set are used as input to train the initial mineral type recognition model to obtain a trained mineral type recognition model.

[0121] Set the model training initialization parameters, such as training batch Batch = 32, training round Epoch = 100, use Adam optimizer, use cross entropy loss function to distinguish the loss of predicted value and true value, and set evaluation criteria: Accuracy, Precision, Recall and F1_score. During training, the pre-trained weights of the EfficientNet-B0 model are pre-assigned to the corresponding structure of the EfficientNet-B0 model in the initial mineral type recognition model as the initial value, and then the training set is used for training to obtain the trained model. At this time, the trained model is trained with the test set to obtain the accuracy and loss value, and the model weight parameters with higher accuracy are saved. Then, it is judged whether the loss value converges. If it converges, the training is completed and the trained mineral type recognition model is obtained. If it does not converge, it is trained again. After obtaining the trained mineral type recognition model, its performance is verified using the validation set. The results show that it has good recognition accuracy.

[0122] This embodiment provides a big data-driven mineral type identification method based on deep learning and feature fusion algorithms, specifically including: data set preprocessing: preprocessing the original orthogonal polarization and single polarization image data, including image registration, image segmentation, data standardization, data set division, data enhancement and mineral prediction; constructing an orthogonal polarization feature extraction module and a single polarization feature extraction module respectively, selecting a feature extraction stage with rich semantic information, constructing a weight calculation module for dual-link polarization image features, adaptively selecting more meaningful parts of different polarization image features for weighted fusion, and effectively improving the recognition accuracy of mineral types by iterating image features of different spatial scales. This embodiment improves the prediction accuracy of rock thin section mineral images and the robustness of the model, improves the efficiency of large-scale mineral classification, and reduces the cost and subjective errors of manual mineral identification.

[0123] Compared with the prior art, the advantages of this embodiment are:

[0124] (1) It provides an efficient and low-cost deep learning prediction method for intelligent identification of rock thin section minerals.

[0125] (2) Through data enhancement, the feature recognition capability of deep learning for geological image data is improved, the robustness of the model is enhanced, and it can be applied to the intelligent classification of multiple types of mineral images.

[0126] (3) Through deep learning image feature extraction combined with image feature fusion technology, the orthogonal polarization and single polarization image features of different spatial scales are comprehensively extracted, which improves the mineral differentiation effect of geological rock thin section mineral images when the polarization features are not obvious, and enhances the prediction accuracy and stability of the model for mineral identification.

[0127] (4) The deep learning transfer learning technology is used to combine the pre-trained weights of EfficientNet-B0 and the feature fusion technology of different polarized images, which improves the training efficiency and stability of the model and reduces the annotation cost of geological images.

[0128] The embodiment of the present application also provides an application scenario, which applies the above-mentioned rock thin section mineral type identification method. Specifically, the rock thin section mineral type identification method provided in this embodiment can be applied in the mineral type identification scenario. The mineral type identification scenario includes a data acquisition link and a data processing link. The data acquisition link is used to obtain the orthogonal polarization image and the single polarization image of the rock thin section. The data processing link is used to process the orthogonal polarization image and the single polarization image of the rock thin section to obtain the type of each mineral in the rock thin section. The rock thin section mineral type identification method provided in this embodiment belongs to the data processing link.

[0129] Example 2

[0130] Based on the same inventive concept, the embodiment of the present application also provides a rock thin section mineral type identification device for implementing the rock thin section mineral type identification method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in the rock thin section mineral type identification device embodiment provided below can refer to the above limitations on the rock thin section mineral type identification method, and will not be repeated here.

[0131] like Fig.32 As shown, a rock thin section mineral type identification device is provided, and the rock thin section mineral type identification device comprises:

[0132] The image acquisition module M1 is used to acquire the orthogonal polarization image and the single polarization image of the rock thin section.

[0133] The image processing module M2 is used to perform image registration and image segmentation on the orthogonal polarization image and the single polarization image to obtain an orthogonal polarization sub-image and a single polarization sub-image corresponding to the spatial position of each mineral in the rock slice.

[0134] The mineral type prediction module M3 is used to determine the type of each mineral in the rock thin section by taking the orthogonal polarized photon image and the single polarized photon image corresponding to the spatial position of the mineral as input, and using the trained mineral type recognition model to realize the rock thin section mineral type recognition; wherein the trained mineral type recognition model is a model obtained after integrating the feature fusion module into the feature extraction module of the deep learning model, and the feature fusion module is used to fuse the features of the orthogonal polarized photon image and the features of the single polarized photon image.

[0135] Example 3

[0136] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Fig.33 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for identifying the mineral type of a rock thin section is implemented.

[0137] Those skilled in the art will understand that Fig.33 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0138] In an exemplary embodiment, a computer device is also provided, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the rock thin section mineral type identification method described in Example 1.

[0139] Example 4

[0140] An embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the rock thin section mineral type identification method described in Embodiment 1 is implemented.

[0141] Example 5

[0142] An embodiment of the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the rock thin section mineral type identification method described in Example 1.

[0143] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0144] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0145] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for identifying mineral types in rock thin sections, characterized in that: The rock thin section mineral type identification method comprises: Obtain orthogonal polarization images and single polarization images of rock thin sections; Performing image registration and image segmentation on the orthogonal polarization image and the single polarization image to obtain an orthogonal polarization sub-image and a single polarization sub-image corresponding to the spatial position of each mineral in the rock slice; For each mineral in the rock thin section, the orthogonal polarized photon image and the single polarized photon image corresponding to the spatial position of the mineral are taken as input, and the type of the mineral is determined using a trained mineral type recognition model to achieve rock thin section mineral type recognition; wherein the trained mineral type recognition model is a model obtained by integrating a feature fusion module into the feature extraction module of the deep learning model, and the feature fusion module is used to fuse the features of the orthogonal polarized photon image and the features of the single polarized photon image.

2. The rock thin section mineral type identification method according to claim 1, characterized in that: Performing image registration and image segmentation on the orthogonal polarization image and the single polarization image to obtain an orthogonal polarization sub-image and a single polarization sub-image corresponding to the spatial position of each mineral in the rock slice, specifically includes: Taking the orthogonal polarization image as a reference, the single polarization image is moved and rotated to obtain a converted single polarization image, so as to perform image registration on the orthogonal polarization image and the single polarization image, so that the spatial positions of the same mineral in the converted single polarization image and the orthogonal polarization image correspond to each other; The orthogonal polarization image and the converted single polarization image are respectively segmented to obtain an orthogonal polarization sub-image and a single polarization sub-image corresponding to the spatial position of each mineral in the rock slice.

3. The rock thin section mineral type identification method according to claim 1, characterized in that: The deep learning model is an EfficientNet-B0 model, which includes a first branch, a second branch, and a third branch, wherein the output end of the first branch and the output end of the second branch are both connected to the input end of the third branch; the first branch and the second branch have the same structure, the input of the first branch is an orthogonal polarized photon image, and the input of the second branch is a single polarized photon image, the first branch includes a first stage, a second stage, a third stage, a fourth stage, and a fifth stage connected in sequence; the third branch includes a sixth stage, a seventh stage, an eighth stage, and a ninth stage connected in sequence; The feature fusion module includes a first feature fusion unit, a second feature fusion unit and a third feature fusion unit; At this time, in the trained mineral type recognition model, the first input end of the first feature fusion unit is connected to the third stage of the first branch, the second input end is connected to the third stage of the second branch, the first output end is connected to the fourth stage of the first branch, and the second output end is connected to the fourth stage of the second branch; The first input end of the second feature fusion unit is connected to the fourth stage of the first branch, the second input end is connected to the fourth stage of the second branch, the first output end is connected to the fifth stage of the first branch, and the second output end is connected to the fifth stage of the second branch; The first input end of the third feature fusion unit is connected to the fifth stage of the first branch, the second input end is connected to the fifth stage of the second branch, and the output end is connected to the sixth stage of the third branch.

4. The rock thin section mineral type identification method according to claim 3 is characterized in that: The first stage includes sequentially connected convolutional layers, batch normalization layers, and activation function layers; the second, third, fourth, fifth, sixth, seventh, and eighth stages all include mobile flip bottleneck convolution modules; the ninth stage includes sequentially connected convolutional layers, batch normalization layers, activation function layers, average pooling layers, and classification layers, and the classification layer is a fully connected layer.

5. The rock thin section mineral type identification method according to claim 3 is characterized in that: The first feature fusion unit has the same structure as the second feature fusion unit, and the first feature fusion unit includes a weight calculation subunit and a weighted feature calculation subunit connected in sequence; The third feature fusion unit includes a weight calculation subunit, a weighted feature calculation subunit and a feature fusion calculation subunit connected in sequence; The calculation formula of the weight calculation subunit is as follows: Among them, Attention is the weight; Sigmoid is the activation function layer; Conv 1×1 is a 1×1 convolutional layer; f ortho is the feature of the orthogonal polarization sub-image input to the weight calculation sub-unit; f single is the feature of the single polarization photon image input to the weight calculation subunit; The calculation formula of the weighted feature calculation subunit is: in, is the weighted feature of the orthogonal polarized photon image; is the weighted feature of the single polarization photon image; The calculation formula of the feature fusion calculation subunit is: Among them, f fuse Conv is the fusion feature; 3×3 is a 3×3 convolutional layer; Concat is a concatenation layer.

6. The rock thin section mineral type identification method according to claim 1, characterized in that: Before determining the type of the mineral using the trained mineral type recognition model by taking the orthogonal polarization photon image and the single polarization photon image corresponding to the spatial position of the mineral as input, the rock thin section mineral type recognition method further includes: Acquire a data set; the data set includes sample features and sample labels of each sample mineral in a plurality of sample rock thin sections, the sample features are sample orthogonal polarization images and sample single polarization images corresponding to the spatial positions of the sample minerals, and the sample labels are the types of the sample minerals; Performing data standardization processing on the data set to obtain a standardized data set; Randomly dividing the standardized data set to obtain a training set, a validation set, and a test set; Performing data enhancement processing on the training set to obtain an enhanced training set; the data enhancement processing includes: rotation, brightness transformation and adding noise; The enhanced training set, the validation set and the test set are used as inputs to train the initial mineral type recognition model to obtain a trained mineral type recognition model.

7. A rock thin section mineral type identification device, characterized in that: The rock thin section mineral type identification device comprises: An image acquisition module, used to acquire orthogonal polarization images and single polarization images of rock thin sections; An image processing module, used for performing image registration and image segmentation on the orthogonal polarization image and the single polarization image, to obtain an orthogonal polarization sub-image and a single polarization sub-image corresponding to the spatial position of each mineral in the rock slice; A mineral type prediction module is used to determine the type of each mineral in a rock thin section by using an orthogonal polarized photon image and a single polarized photon image corresponding to the spatial position of the mineral as input, and to realize rock thin section mineral type identification by using a trained mineral type identification model; wherein the trained mineral type identification model is a model obtained by integrating a feature fusion module into a feature extraction module of a deep learning model, and the feature fusion module is used to fuse the features of the orthogonal polarized photon image and the features of the single polarized photon image.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the rock thin section mineral type identification method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the rock thin section mineral type identification method described in any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the rock thin section mineral type identification method described in any one of claims 1 to 6 is implemented.

Images