A pore identification method and system for core images

Abstract

This invention provides a method and system for identifying pores in core images, belonging to the field of petroleum exploration and development. The method involves adding DropBlock and BN layers to the convolutional blocks of the original U-Net network to obtain a reconstructed convolutional block. Furthermore, the feature pyramid module (FPN) replaces the jumps between upsampling and downsampling in the original U-Net model to obtain an improved U-Net architecture. The improved U-Net architecture is trained to obtain a pore identification model for identifying core images. The image to be tested is input into the pore identification model to obtain pore identification results for the core image. This method enhances the stability and reliability of prediction results, helping operators make faster decisions and optimize petroleum exploration and development plans.

Description

Technical Field

[0001] The present invention belongs to the field of petroleum exploration and development, and in particular relates to a pore identification method and system for core images. Background Art

[0002] Tight sandstone reservoirs possess complex reservoir characteristics, primarily characterized by deep burial depths, widespread distribution, and complex sedimentation. Reservoir properties exhibit low porosity, low permeability, and strong heterogeneity. Pores are primarily micro- and nano-pores, with narrow and dispersed throats, hindering the migration and accumulation of oil and gas. These factors complicate reservoir classification and evaluation, requiring consideration of the interplay of multiple factors, making the selection of high-quality reservoirs challenging.

[0003] In the classification and evaluation of tight sandstone reservoirs, traditional techniques provide effective classification and discrimination of carbonate reservoirs using a variety of geological and geophysical parameters. By constructing multivariate distribution models, the accuracy and reliability of reservoir evaluation have been improved. In recent years, deep learning technology has attracted widespread attention, spawning a series of salient object detection methods based on deep neural networks. These methods abandon traditional feature extraction methods and instead rely on encoders to automatically extract salient features from images, thereby accurately detecting key objects in the image. However, the accuracy and efficiency of deep learning methods for identifying core pores in complex backgrounds need to be improved. Summary of the Invention

[0004] In order to solve the problem of pore identification in core images under complex rock formations, the present invention provides a pore identification method and system for core images.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] A pore identification method for core images, comprising:

[0007] Obtain scanning electron microscope images of core pores;

[0008] Based on the original U-Net network, a DropBlock layer and a BN layer are added to the convolution block of the original U-Net network to obtain a reconstructed convolution block, and the jump between upsampling and downsampling in the original U-Net network is replaced by a feature pyramid module FPN to obtain an improved U-Net architecture. The improved U-Net architecture is trained using the core pore scanning electron microscope image to obtain a pore recognition model.

[0009] The core image to be identified is input into the pore identification model to obtain the pore identification result of the core image.

[0010] Preferably, the pore recognition model includes an encoder, which consists of repeated applications of two 3×3 convolutions, each followed by a nonlinear Mish activation function and a 2×2 maximum pooling operation.

[0011] Preferably, in the pore recognition model, the features of the core image to be identified are extracted by reconstructing the convolution block, the obtained features are input into the feature pyramid module, and the feature maps of different scales are fused; the feature vector output by the feature pyramid module is connected with the feature vector obtained after upsampling to obtain the pore recognition result of the core image.

[0012] Preferably, during the training of the improved U-Net architecture, positive pixels and negative pixels in the training set are weighted, specifically:

[0013] ;

[0014] Where N represents the number of samples, represents the positive prediction weight, is the true label, which takes the value 0 or 1. Represents the probability value predicted by the pore identification model.

[0015] Preferably, before using the core pore scanning electron microscope image to train the improved U-Net architecture, the core pore scanning electron microscope image is enhanced, specifically by performing operations including scaling, flipping, shifting and rotating, and adding impulse noise to the input slice of the image.

[0016] Preferably, the pore identification result of the core image now includes the location, size, outline and size calculation parameters of the rock pores.

[0017] Preferably, the performance of the improved U-Net architecture is evaluated using precision, mean absolute error, and Dice coefficient.

[0018] The present invention also provides a pore identification system for core images, specifically comprising:

[0019] The data acquisition module is used to obtain the scanning electron microscope image of the core pores.

[0020] The model construction module is used to add a DropBlock layer and a BN layer to the convolution block of the original U-Net network as a basis to obtain a reconstructed convolution block, and use a feature pyramid module FPN to replace the jump between upsampling and downsampling in the original U-Net network to obtain an improved U-Net architecture; the improved U-Net architecture is trained using the core pore scanning electron microscope image to obtain a pore recognition model.

[0021] The recognition module is used to input the core image to be recognized into the pore recognition model to obtain the pore recognition result of the core image.

[0022] The pore identification method of a core image provided by the present invention has the following beneficial effects:

[0023] This paper adds DropBlock and BN layers to the convolutional blocks of the original U-Net network to obtain a reconstructed convolutional block. The DropBlock layer better preserves the structural information of the input data, while the BN layer accelerates network training and convergence, preventing gradient vanishing and overfitting. The feature pyramid module (FPN) replaces the jumps between upsampling and downsampling in the original U-Net model, fusing features of different resolutions and improving feature extraction capabilities in complex backgrounds, resulting in an improved U-Net architecture. This improved U-Net architecture is then trained to obtain a pore recognition model for identifying pores in core images. This pore recognition model effectively identifies and labels core pores, providing valuable support for subsequent research on reservoir classification and evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] To more clearly illustrate the embodiments of the present invention and its design, the following briefly introduces the drawings required for this embodiment. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be derived from these drawings without inventive effort.

[0025] Figure 1 The figure is a flow chart of a method for identifying pores in a core image according to an embodiment of the present invention.

[0026] Figure 2 This is a sample diagram of a data set in an embodiment of the present invention.

[0027] Figure 3 This is a diagram of the improved U-Net network architecture in an embodiment of the present invention.

[0028] Figure 4 This is a structural diagram of the reconstructed convolution module in an embodiment of the present invention.

[0029] Figure 5 This is a structural diagram of the feature pyramid module in an embodiment of the present invention.

[0030] Figure 6 Schematic diagram of pore identification results in an embodiment of the present invention. DETAILED DESCRIPTION

[0031] In order to enable those skilled in the art to better understand the technical solution of the present invention and to be able to implement it, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solution of the present invention and are not intended to limit the scope of protection of the present invention.

[0032] Example:

[0033] The present invention provides a pore identification method for core images, such as Figure 1 As shown, the specific steps include:

[0034] Step 1: Obtain sandstone pore scanning electron microscope image data from different oil fields, combine the researchers' prior knowledge to label the images, and generate a dataset that can be used for deep learning research. Some examples of the dataset, such as Figure 2 shown.

[0035] Step 2: Perform data preprocessing. Standardize the input data to improve the efficiency and effectiveness of model training and adapt to the model input requirements. The input size of the model is 512×512 pore SEM images, so a data augmentation strategy is adopted instead of a sampling strategy to enhance the generalization potential and robustness of the model. The data augmentation methods implemented in the proposed network include scaling, flipping, shifting, and rotation. In addition, impulse noise is added to the input slices of the image to improve the generalization ability of the neural network. While processing the noise, the input image is elastically transformed, randomly sheared, scaled, and rotated to keep the input image size unchanged. Therefore, by applying these small transformations to the image during training, the diversity of the training dataset can be increased and the robustness of the proposed model can be improved.

[0036] Step 3: Based on the original U-Net network, the DropBlock layer and the BN layer are added to the convolution block of the original U-Net network to obtain a reconstructed convolution module. The feature pyramid module FPN is used to replace the jump between upsampling and downsampling in the original U-Net network to obtain an improved U-Net architecture. The improved U-Net architecture is trained using core pore scanning electron microscope images to obtain a pore recognition model.

[0037] The preprocessed dataset was divided into a training set, a validation set, and a test set. The training set data was imported to train the improved U-Net network. The accuracy of the pore recognition model was verified using the validation set and the test set.

[0038] Step 3.1: Add the DropBlock layer and the BN layer to the original convolution module to obtain the reconstructed convolution module, as shown in Figure 4As shown. DropBlock is a regularization technology, an improvement to the traditional Dropout method. By randomly blocking neurons in an area instead of a single neuron, the redundant connections in the neural network are reduced, thereby improving the generalization ability of the model. Batch Normalization (BN) is an important regularization and training acceleration technology. By standardizing the input distribution of the intermediate layer of the neural network, it significantly improves the training speed and stability of the deep network. The jump connection between upsampling and downsampling in the original UNet model is replaced by the feature pyramid module FPN. The feature pyramid module structure is as follows Figure 5 As shown, the improved U-Net architecture is obtained.

[0039] Step 3.2: The improved U-Net architecture takes a 512×512 image as input and outputs a 512×512 mask. Figure 3 As shown, the input image is processed by two 3×3 reconvolutional convolutional modules, which sequentially perform a 3×3 convolution, a DropBlock layer for regularization, a BN layer for batch normalization, and a RELU activation function. After the convolution, a 2×2 MAX pooling operation is performed to downsample the input image features. The number of feature channels doubles with each downsampling step, and the depth of the contraction layer is 5.

[0040] Step 3.3: Input the features at five depths into the feature pyramid module (FPN), FPN is Figure 3 The circular processing structure in the middle inputs the feature vector to the decoder. Each step in the expansion path involves upsampling the feature map using a 2×2 upsampling, which halves the number of feature channels at each depth. The resulting upsampled feature vector is concatenated with the corresponding feature vector in the feature pyramid. The concatenation operation is followed by two 3×3 convolutions, each followed by a Mish activation function. The FPN method fuses feature maps of different scales by constructing a feature fusion path consisting of a bottom-up network, a top-down network, lateral connections, and convolutions, thereby generating a feature pyramid with rich multi-scale information. The input is the five depth-wise feature maps corresponding to the encoder of the backbone network structure, and the output is fed into the decoder of the backbone network. The FPN adds additional weights to each input, allowing the network to learn the importance of specific input features. This method achieves a fast normalization fusion that combines dynamic learning behavior and accuracy. In addition, batch normalization and the ReLU nonlinear activation function are used to improve efficiency.

[0041] Step 3.4: In the last layer of the decoder, the resulting 512×512×64 feature map is passed through two 3×3 convolutions, followed by a Mish activation function and a final 1×1 convolution block, and finally a sigmoid activation function is applied to output the mask.

[0042] Step 3.5: The goal of network training is to increase the probability of correctly classifying each pixel in the mask. To achieve this, a weighted binary cross-entropy loss is applied to each training example. To achieve weighted binary cross-entropy, the positive and negative pixels in the training set are weighted. Positive pixels refer to the white portion of the labeled image (the main body), while negative pixels refer to the black background. The formula for weighted binary cross-entropy loss is:

[0043] ;

[0044] Where N represents the number of samples, represents the positive prediction weight, is the true label, which takes the value 0 or 1. represents the predicted value of the pore identification model.

[0045] Step 3.6: Evaluate the model performance using metrics such as precision, mean absolute error (MAE), and Dice coefficient.

[0046] Step 4: Input the core image to be identified into the pore identification model to obtain the pore identification results of the core image. The identification results include the location, size, outline, and size calculation parameters of the rock pores (including area, perimeter, roundness, and concavity).

[0047] The present invention also provides a pore identification system for core images, specifically comprising:

[0048] The data acquisition module is used to obtain the scanning electron microscope image of the core pores.

[0049] A model construction module is used to add DropBlock layers and BN layers to the convolution blocks of the original U-Net network as a basis to obtain a reconstructed convolution module, and to replace the jumps between upsampling and downsampling in the original U-Net network with the feature pyramid module FPN to obtain an improved U-Net architecture. The improved U-Net architecture is trained using scanning electron microscope images of core pores to obtain a pore recognition model.

[0050] The recognition module is used to input the core image to be recognized into the pore recognition model to obtain the pore recognition result of the core image.

[0051] Build a program software. Connect the above-mentioned core image pore recognition system to the program software, extract the pores in the core image through interactive operation, and automatically calculate the pore shape parameters such as pore perimeter, shape factor, major axis length, flatness ratio and concavity. Figure 6 shown.

[0052] The present invention has the following beneficial effects: (1) Providing a scientific basis for oil exploration and development: The high-precision pore identification and prediction method provided by the present invention provides an important geophysical parameter basis for oil exploration and development, helps to more accurately understand the reservoir characteristics of tight sandstone oil layers, helps to carry out reservoir classification evaluation, guides the exploration and development decision-making of oil and gas resources, reduces exploration risks, and improves economic benefits.

[0053] (2) Achieving high-precision identification and prediction, and improving the stability and reliability of prediction results: This method carefully screens high-definition scanning electron microscope image data of rocks with significant correlations and preprocesses the data images to effectively remove outliers and improve data quality. Furthermore, by constructing and training a deep learning model, the accuracy of pore location recognition is significantly improved. At the same time, the introduction of model optimization further enhances the stability and reliability of the prediction results.

[0054] (3) The image editing equipment equipped with the method of the present invention has the functions of identifying images, calculating parameters, exporting results, and saving prediction results, allowing direct comparison between the original image and the prediction results. This not only simplifies the pore identification and prediction process and improves work efficiency, but also makes the prediction results more intuitive and easy to understand. This helps oil explorers make decisions more quickly and optimize exploration and development plans, thereby improving overall work quality and economic benefits.

[0055] Each module in the aforementioned core image pore identification system can be implemented in whole or in part through software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a computer device memory in the form of software, so that the processor can call and execute the corresponding operations of each module.

[0056] It should be pointed out that the specific implementation methods described above can enable those skilled in the art to understand the invention more comprehensively, but do not limit the invention in any way. Therefore, although the present specification and examples have described the invention in detail, those skilled in the art should understand that the invention can still be modified or replaced by equivalents; and all technical solutions and improvements that do not deviate from the spirit and scope of the invention are included in the scope of protection of the patent for the invention. Any figure mark in the claims should not be regarded as limiting the claims involved. Any simple change or equivalent replacement of the technical solution that can be obviously obtained by any person familiar with the art within the technical scope disclosed in the present invention falls within the scope of protection of the present invention.

Claims

1. A pore identification method for core images, characterized in that: The following steps are involved: Obtain scanning electron microscope images of core pores; Based on the original U-Net network, the DropBlock layer and the BN layer are added to the convolution block of the original U-Net network to obtain a reconstructed convolution block. The feature pyramid module FPN is used to replace the jump between upsampling and downsampling in the original U-Net network to obtain an improved U-Net architecture. The improved U-Net architecture is trained using the core pore scanning electron microscope image to obtain a pore recognition model; the pore recognition model includes an encoder, which is composed of two repeated applications of 3×3 convolutions, each followed by a nonlinear Mish activation function and a 2×2 maximum pooling operation; Input the core image to be identified into the pore recognition model, extract features of the core image to be identified through the reconstructed convolution block, input the obtained features into the feature pyramid module, and fuse feature maps of different scales; The feature vector output by the feature pyramid module is connected with the feature vector obtained after upsampling to obtain the pore recognition result of the core image.

2. The pore identification method of a core image according to claim 1, characterized in that: During the training of the improved U-Net architecture, the positive and negative pixels in the training set are weighted as follows: ； Where N represents the number of samples, represents the positive prediction weight, is the true label, which takes the value 0 or 1. Represents the probability value predicted by the pore identification model.

3. The pore identification method of a core image according to claim 1, characterized in that: The method also includes enhancing the core pore scanning electron microscope image before training the improved U-Net architecture using the core pore scanning electron microscope image, specifically performing operations including scaling, flipping, shifting and rotating, and adding impulse noise to the input slice of the image.

4. The method for pore identification of a core image according to claim 1, characterized in that: The pore identification result of the core image specifically includes the location, size, outline and size calculation parameters of the rock pores.

5. The method for pore identification of a core image according to claim 1, characterized in that: It also includes the evaluation of the performance of the improved U-Net architecture using precision, mean absolute error, and Dice coefficient.

6. A pore identification system for core images, characterized in that: include: A data acquisition module is used to obtain scanning electron microscope images of core pores; The model building module is used to add DropBlock and BN layers to the convolutional blocks of the original U-Net network as the basis to obtain a reconstructed convolutional block. The feature pyramid module FPN replaces the jumps between upsampling and downsampling in the original U-Net network to obtain an improved U-Net architecture. The improved U-Net architecture is trained using the core pore scanning electron microscope image to obtain a pore recognition model; the pore recognition model includes an encoder, which is composed of two repeated applications of 3×3 convolutions, each followed by a nonlinear Mish activation function and a 2×2 maximum pooling operation; The recognition module is used to input the core image to be identified into the pore recognition model, extract features of the core image to be identified by reconstructing the convolution block, input the obtained features into the feature pyramid module, and fuse feature maps of different scales; The feature vector output by the feature pyramid module is connected with the feature vector obtained after upsampling to obtain the pore recognition result of the core image.

Images