Well-passing crack image separation method and device, electronic equipment and medium

Abstract

The invention relates to a through-well crack image separation method and device, electronic equipment and a medium, and belongs to the technical field of computer vision, and the method comprises the steps: carrying out the preprocessing of a through-well crack image data set, and obtaining a preprocessed data set; adding a channel and a space fusion attention mechanism on the basis of the Deeplabv3 + model to obtain a GADeeplabv3 + semantic segmentation model; training a GADeeplabv3 + semantic segmentation model on the basis of the data set to obtain a GADeeplabv3 + semantic segmentation model which is completely trained; obtaining a target well-passing crack image; and inputting the target well-passing crack image into the fully trained GADeeplabv3 + semantic segmentation model to obtain a crack image after separation. According to the method, the channel and space attention mechanism is added on the basis of the Deeplabv3 + model, so that the extraction precision and efficiency of the through-well crack are improved.

Description

Technical Field

[0001] The present invention relates to the field of computer vision technology, and in particular to a method, device, electronic equipment and medium for separating through-well fracture images. Background Art

[0002] With the rapid development of social economy, the consumption of oil and natural gas resources is increasing day by day. The exploration and development of fractured oil and gas reservoirs is of great significance to the stable supply of oil and gas resources, because the cracks in fractured oil and gas reservoirs can affect the distribution of oil and gas, so the effective detection of cracks has a certain guiding role in the exploration and development of fractured oil and gas reservoirs. Among them, through-well fractures are a special type of well logging fractures, which often represent the cross-section of the well wall, so the research on the detection of through-well fractures is particularly important.

[0003] At present, major oil fields mainly use manual analysis, which requires rich experience and professional knowledge, is time-consuming and labor-intensive, and has subjective errors. In addition, the complex downhole environment often leads to incomplete and inaccurate data collection, further increasing the difficulty of fracture detection.

[0004] In summary, the prior art lacks a through-well fracture image segmentation method to improve the efficiency and accuracy of through-well fracture extraction. Summary of the invention

[0005] In view of this, it is necessary to provide a method, device, electronic device and medium for separating through-well fracture images to solve the problem of low efficiency and accuracy in extracting through-well fracture images in the prior art. In order to solve the above problems, the present invention provides a method for segmenting a through-well fracture image, comprising: Preprocessing the through-well fracture image data set to obtain a preprocessed data set; Based on the Deeplabv3+ model, the channel and spatial fusion attention mechanism is added to obtain the GA_Deeplabv3+ semantic segmentation model; Training a GA_Deeplabv3+ semantic segmentation model based on the training set in the data set to obtain a fully trained GA_Deeplabv3+ semantic segmentation model; Acquire target through-well fracture images; The target through-well fracture image is input into the well-trained GA_Deeplabv3+ semantic segmentation model to obtain a segmented fracture image.

[0006] In a possible implementation, the GA_Deeplabv3+ semantic segmentation model is obtained by adding channel and spatial fusion attention mechanisms to the Deeplabv3+ model, including: The channel and spatial fusion attention mechanism is added to the spatial pyramid pooling module of the Deeplabv3+ model to obtain an improved spatial pyramid pooling module; The GA_Deeplabv3+ semantic segmentation model is obtained based on the improved spatial pyramid pooling module.

[0007] In one possible implementation, the channel and spatial fusion attention mechanism includes a channel fusion attention mechanism and a spatial fusion attention mechanism. The channel fusion attention mechanism is used to weight the input feature map in the channel dimension, and the spatial fusion attention mechanism is used to weight the input feature map in the spatial dimension.

[0008] In a possible implementation, the GA_Deeplabv3+ semantic segmentation model includes: an input end, a backbone feature extraction network, an improved spatial pyramid pooling module, a decoder, and an output end; The input end is used to set the resolution of the input picture to obtain a first image; The backbone feature extraction network is used to extract a first feature of the first image; The improved spatial pyramid pooling module is used to reduce the dimension of the first feature to obtain a second feature; The decoder is used for decoding the second feature to obtain a second image; The output end is used to output the second image.

[0009] In a possible implementation manner, a Gabor filter is added to the input end.

[0010] In a possible implementation, a regularization term based on a sinusoidal fitting strategy and a crack detection loss are added to the cross entropy loss function at the output end to obtain a total loss function.

[0011] In a possible implementation, the expression of the total loss function is:

[0012] In the formula, represents the total loss function; represents the weight of the cross entropy loss function, represents the cross entropy loss function; represents the weight of crack detection loss; represents the crack detection loss, represents the weight of the regularization term based on the sine fitting strategy; represents the regularization term based on the sine fitting strategy; where:

[0013]

[0014]

[0015] In the formula, C represents the number of categories; represents the probability of the true label, i.e. the cross-well fracture area; It indicates the probability of the area predicted as a cross-well fracture by the GA_Deeplabv3+ semantic segmentation model; , Represents the i-th parameter of the output sine fit; Represents the i-th parameter of the target sine fit.

[0016] On the other hand, the present invention also provides a through-well fracture image separation device, comprising: A data set acquisition module is used to preprocess the through-well fracture image data set to obtain a preprocessed data set; The semantic segmentation model acquisition module is used to add channel and spatial fusion attention mechanisms to the Deeplabv3+ model to obtain the GA_Deeplabv3+ semantic segmentation model; A training segmentation model module is used to train a GA_Deeplabv3+ semantic segmentation model based on a training set in the data set to obtain a fully trained GA_Deeplabv3+ semantic segmentation model; A target image acquisition module is used to acquire a target through-well fracture image; The segmented image module is used to input the target through-well fracture image into the fully trained GA_Deeplabv3+ semantic segmentation model to obtain a segmented fracture image.

[0017] On the other hand, the present invention also provides an electronic device, including a memory and a processor, wherein: The memory is used to store programs; The processor is coupled to the memory and is used to execute the program stored in the memory to implement the steps in the through-well fracture image segmentation method described in any one of the above implementations.

[0018] On the other hand, the present invention further provides a computer-readable storage medium for storing a computer-readable program or instruction, which, when executed by a processor, can implement the steps of a through-well fracture image segmentation method described in any of the above implementations.

[0019] The beneficial effects of the present invention are as follows: a through-well fracture image segmentation method provided by the present invention preprocesses a through-well fracture image dataset to obtain a preprocessed dataset; a channel and space fusion attention mechanism is added on the basis of the Deeplabv3+ model to obtain a GA_Deeplabv3+ semantic segmentation model, and the channel and space fusion attention mechanism emphasizes the channel that contributes most to the task and suppresses irrelevant or redundant channels by assigning different weights to each channel, thereby improving the performance of the model; a GA_Deeplabv3+ semantic segmentation model with complete training is obtained based on the dataset training set GA_Deeplabv3+ semantic segmentation model; a target through-well fracture image is obtained; the target through-well fracture image is input into the GA_Deeplabv3+ semantic segmentation model with complete training to obtain a separated fracture image. The present invention improves the extraction accuracy and efficiency of through-well fractures by adding channel and space attention mechanisms on the basis of the Deeplabv3+ model. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A method flow chart of an embodiment of a method for segmenting a through-well fracture image provided by the present invention; Figure 2 For the present invention Figure 1 A schematic flow chart of an embodiment of S102; Figure 3 It is a schematic diagram of the GA_Deeplabv3+ semantic segmentation model of the present invention and the average intersection-over-union ratio curve obtained by extracting cross-well fractures; Figure 4 Comparison chart of GA_Deeplabv3+ semantic segmentation model and deeplabv3+ semantic segmentation model used in the present invention; Figure 5 A schematic flow chart of an embodiment of a through-well fracture image separation device provided by the present invention; Figure 6 A schematic structural diagram of an embodiment of an electronic device provided by the present invention. DETAILED DESCRIPTION

[0021] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not used to limit the scope of the present invention.

[0022] The present invention provides a method, device, electronic equipment and medium for separating through-well fracture images, which are described below respectively.

[0023] Figure 1 A schematic flow chart of an embodiment of the method for segmenting a through-well fracture image provided by the present invention is shown in FIG. Figure 1As shown, the through-well fracture image segmentation method includes: S101, preprocessing the through-well fracture image data set to obtain a preprocessed data set; S102. Based on the Deeplabv3+ model, the channel and spatial fusion attention mechanism is added to obtain the GA_Deeplabv3+ semantic segmentation model; S103, training a GA_Deeplabv3+ semantic segmentation model based on a training set in the data set to obtain a fully trained GA_Deeplabv3+ semantic segmentation model; S104, acquiring a target through-well fracture image; S105, inputting the target through-well fracture image into the fully trained GA_Deeplabv3+ semantic segmentation model to obtain a segmented fracture image.

[0024] Compared with the prior art, the present embodiment provides a method for segmenting through-well fracture images, preprocessing a through-well fracture image dataset to obtain a preprocessed dataset; adding a channel and space fusion attention mechanism on the basis of the Deeplabv3+ model to obtain a GA_Deeplabv3+ semantic segmentation model, the channel and space fusion attention mechanism assigns different weights to each channel, emphasizes the channel that contributes most to the task, and suppresses irrelevant or redundant channels, thereby improving the performance of the model; training the GA_Deeplabv3+ semantic segmentation model based on the dataset to obtain a fully trained GA_Deeplabv3+ semantic segmentation model; obtaining a target through-well fracture image; inputting the target through-well fracture image into the fully trained GA_Deeplabv3+ semantic segmentation model to obtain a separated fracture image. The present invention improves the extraction accuracy and efficiency of through-well fractures by adding a channel and space attention mechanism on the basis of the Deeplabv3+ model.

[0025] It should be noted that the application scenario of the embodiment of the present invention is to extract through-well fracture images. It can be understood that the present invention is not limited to extracting through-well fracture images, but can also be applied to other scenarios that require accurate image extraction.

[0026] In a specific embodiment of the present invention, in step S101, preprocessing the through-well fracture image dataset to obtain a preprocessed dataset specifically includes: A data set of logging images containing cross-well fractures was collected and calibrated, and then divided into training set and test set in a ratio of 8:2. Data enhancement was performed at the input end of the training to increase the complexity of the training samples and obtain the preprocessed data set.

[0027] In some embodiments of the present invention, in step S102, as Figure 2 As shown, the GA_Deeplabv3+ semantic segmentation model is obtained by adding channel and spatial fusion attention mechanism on the basis of Deeplabv3+ model, including: S201. Add channel and spatial fusion attention mechanism to the spatial pyramid pooling module of Deeplabv3+ model to obtain an improved spatial pyramid pooling module. S202. Obtain the GA_Deeplabv3+ semantic segmentation model based on the improved spatial pyramid pooling module.

[0028] In some embodiments of the present invention, the GA_Deeplabv3+ semantic segmentation model includes: an input end, a backbone feature extraction network, an improved spatial pyramid pooling module, a decoder, and an output end; The input end is used to set the resolution of the input picture to obtain a first image; The backbone feature extraction network is used to extract a first feature of the first image; The improved spatial pyramid pooling module is used to reduce the dimension of the first feature to obtain a second feature; The decoder is used for decoding the second feature to obtain a second image; The output end is used to output the second image.

[0029] In a specific embodiment of the present invention, the GA_Deeplabv3+ semantic segmentation model includes an input end, a backbone feature extraction network, an improved spatial pyramid pooling module, a decoder, and an output end. The input end sets the input resolution to 256×256, the backbone feature extraction network uses Resnet or Xception network, and the spatial pyramid pooling module creates a convolution layer for each branch, including a 1×1 convolution and multiple 3×3 convolutions, and each branch uses a different expansion rate. In this embodiment, the spatial pyramid is improved by adding a channel and spatial fusion attention mechanism (CBAM). The decoder part receives features from the ASSP module and low-level features, and generates the final extraction result through several convolutional layers and ReLU activation. The decoder uses skip connections to retain the information of low-level features. The output end uses sc_loss as the loss function for back propagation.

[0030] In some embodiments of the present invention, the channel and spatial fusion attention mechanism includes a channel fusion attention mechanism and a spatial fusion attention mechanism. The channel fusion attention mechanism is used to weight the input feature map in the channel dimension, and the spatial fusion attention mechanism is used to weight the input feature map in the spatial dimension.

[0031] In a specific embodiment of the present invention, the channel and spatial fusion attention mechanism (CBAM) includes two parts: channel attention and spatial attention. Among them, channel attention emphasizes the features of important channels by weighting the input feature map in the channel dimension. Given the dimension of the input feature map X is C×H×W, where C is the number of channels, H and W are height and width respectively.

[0032] Use global average pooling and global maximum pooling to generate channel descriptors: (1) (2) These two descriptors are processed through a fully connected layer to obtain the channel attention weights: (3) in, is the Sigmoid activation function, and MLP stands for multi-layer perceptron (generally including two linear layers).

[0033] Finally, the weights are applied to the input feature map: (4) Spatial attention emphasizes the features of important positions by weighting the input feature map in the spatial dimension. The input feature map X is averaged and max-pooled in the channel dimension to obtain two two-dimensional feature maps: (5) (6) Concatenate these two feature maps in the channel dimension: (7) Generate spatial attention weights through a convolutional layer: (8) Finally, the weights are applied to the input feature map: (9) The final output of CBAM is to combine the channel attention mechanism with the spatial attention mechanism: (10) It should be noted that the Gabor filter is a linear filter used to detect features such as edges and textures in an image, and has directional selectivity and frequency selectivity. The core of the Gabor filter is a convolution kernel composed of a Gaussian envelope function and a sine wave. In some embodiments of the present invention, a Gabor filter is added to the input end and the output end, and the Gabor filter is applied to the convolution layer. The expression of the Gabor filter is: (11) Among them, (x, y) represents the spatial coordinates, is the standard deviation of the Gaussian envelope function, which is used to control the scale of the filter; is the wavelength of the sine wave, which determines the frequency characteristics of the filter; It is the phase shift of the sine wave that determines the directivity of the filter.

[0034] In practical applications, input data is often a multi-channel image. For example, an RGB image has three color channels, each of which contains independent information. Therefore, the Gabor kernel can be applied to multi-channel convolution operations. Assume that the input image has Cin channels and the output has Cout channels. By repeating the Gabor kernel in the input and output channel dimensions, each input channel can be connected to each output channel. Specifically, the Gabor kernel is repeated Cin×Cout times to form a weight matrix of size Cin×Cout. This extended weight matrix can be directly applied to the convolution operation to ensure that each input channel is processed by the Gabor filter and output feature maps of multiple channels. In this way, the Gabor filter can adapt to multi-channel input and output in convolutional neural networks and support complex image processing tasks.

[0035] In order to ensure that the characteristics of the Gabor filter are not changed during the training process, this embodiment adopts a fixed Gabor kernel weight. When training a neural network, the weights of the convolutional layer are usually updated to better adapt to the task. However, the core characteristics of the Gabor filter are its direction and frequency selectivity, and if the weights are allowed to update, the filter may lose these characteristics. Therefore, this embodiment turns off the gradient calculation when the network is initialized to ensure that the Gabor kernel remains unchanged throughout the training process.

[0036] In some embodiments of the present invention, a regularization term based on a sinusoidal fitting strategy and a crack detection loss are added to the cross entropy loss function at the output end to obtain a total loss function.

[0037] In some embodiments of the present invention, the expression of the total loss function is: (12) In the formula, represents the total loss function; represents the weight of the cross entropy loss function, represents the cross entropy loss function; represents the weight of crack detection loss; represents the crack detection loss, represents the weight of the regularization term based on the sine fitting strategy; represents the regularization term based on the sine fitting strategy.

[0038] Specifically, cross entropy loss is a loss function widely used in machine learning and deep learning, especially in classification problems. It is used to measure the difference between the probability distribution of the model output and the target true distribution, with the goal of making the distribution predicted by the model as close to the true distribution as possible. Its formula is as follows: (13) Wherein, C represents the number of categories, and the default value of C in this embodiment is 2; represents the true label, i.e. the probability of the well fracture area, is the probability of the GA_Deeplabv3+ semantic segmentation model predicting the area as a well fracture. The cross entropy loss can be understood as the closer the probability distribution predicted by the model is to the true distribution, the smaller the cross entropy loss. When the prediction is correct and the confidence is high, the loss is close to zero. If the probability predicted by the model deviates significantly from the true category, the loss will become larger.

[0039] Fracture area loss is essentially a special binary cross entropy loss. Binary loss is a loss function used for binary classification problems. It measures the difference between the binary classification probability predicted by the model and the actual label, and the goal is to make the prediction result output by the model as close to the actual label as possible. In this embodiment, the image is divided into two categories: the cross-well fracture area and the background. The formula is as follows: (14) Among them, the formula represents the true label, i.e. the probability of the well fracture area, is the probability that the GA_Deeplabv3+ semantic segmentation model predicts that it is a well-crossing fracture area. If the true label is a well-crossing fracture area, only That is, the higher the probability of predicting the well-through fracture area, the smaller the loss. If the true label is a non-well-through fracture area, only attention will be paid to , that is, the higher the probability of the non-through-well fracture area is predicted, the smaller the loss is.

[0040] The regularization term of the sine fitting strategy, specifically, first extracts the coordinates of non-zero values ​​(crack locations) from the input binary mask. In order to fit a sine curve to these coordinates, the curve_fit method in the Scipy library is used. This method estimates the amplitude, frequency, phase shift, and offset of the sine function by minimizing the error between the target curve and the fitted curve. The initial prediction values ​​include the initial value of the amplitude, the frequency estimate based on the coordinates, the initial value of the phase is zero, and the initial estimate of the offset.

[0041] Perform sinusoidal fitting on the model output and target respectively. Compare the fitting parameters of the two and calculate the difference between them: (15) (16) represents the i-th parameter of the output sine fit, Represents the i-th parameter of the target sine fit. The exponential function exp() is used to convert the difference into a regularization value between 0 and 1. The smaller the difference, the closer the regularization value is to 1, indicating that the model output has a higher similarity with the target's periodic characteristics. Finally, the regularization values ​​of all samples are averaged as the final sine regularization term.

[0042] In addition, in order to verify the reliability and superiority of the present invention in through-well fracture extraction. Figure 3 The average intersection-over-union curve obtained by using the GA_Deeplabv3+ semantic segmentation model and sc_loss for logging fracture extraction is given, and the mean intersection-over-union ratio (MIoU) is 92.2%. Figure 4 A comparison of the effects of using unmodified deeplabv3+, GA_Deeplabv3+ and sc_loss on the extraction of cross-well fractures is given. Figure 3 and Figure 4 It can be seen from the figure that the embodiment of the present invention improves the effect of through-well fracture extraction.

[0043] In order to better implement a method for segmenting a through-well fracture image in an embodiment of the present invention, based on a method for segmenting a through-well fracture image, correspondingly, as follows Figure 5 As shown, an embodiment of the present invention further provides a through-well fracture image separation device, a through-well fracture image separation device 500 includes: The data set acquisition module 501 is used to preprocess the through-well fracture image data set to obtain a preprocessed data set; A semantic segmentation model acquisition module 502 is used to add channel and spatial fusion attention mechanisms to the Deeplabv3+ model to obtain a GA_Deeplabv3+ semantic segmentation model; A training segmentation model module 503 is used to train a GA_Deeplabv3+ semantic segmentation model based on the data set to obtain a fully trained GA_Deeplabv3+ semantic segmentation model; A target image acquisition module 504 is used to acquire a target through-well fracture image; The segmented image module 505 is used to input the target through-well fracture image into the fully trained GA_Deeplabv3+ semantic segmentation model to obtain a segmented fracture image.

[0044] The through-well fracture image separation device 500 provided in the above embodiment can implement the technical solution described in the above embodiment of a through-well fracture image separation method. The specific implementation principles of the above modules or units can refer to the corresponding contents in the above embodiment of a through-well fracture image separation method, which will not be repeated here.

[0045] like Figure 6 As shown, the present invention also provides an electronic device 600. The electronic device 600 includes a processor 601, a memory 602 and a display 603. Figure 6 Only some components of the electronic device 600 are shown, but it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0046] In some embodiments, the processor 601 may be a central processing unit (CPU), a microprocessor or other data processing chip, used to run program codes or process data stored in the memory 602, such as a through-well fracture image segmentation method in the present invention.

[0047] In some embodiments, the processor 601 may be a single server or a server group. The server group may be centralized or distributed. In some embodiments, the processor 601 may be local or remote. In some embodiments, the processor 601 may be implemented in a cloud platform. In some embodiments, the cloud platform may include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-cloud, etc., or any combination thereof.

[0048] In some embodiments, the memory 602 may be an internal storage unit of the electronic device 600, such as a hard disk or memory of the electronic device 600. In other embodiments, the memory 602 may also be an external storage device of the electronic device 600, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc., equipped on the electronic device 600.

[0049] Furthermore, the memory 602 may include both an internal storage unit of the electronic device 600 and an external storage device. The memory 602 is used to store application software installed in the electronic device 600 and various data.

[0050] In some embodiments, the display 603 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode) touch device, etc. The display 603 is used to display information on the electronic device 600 and to display a visual user interface. The components 601-603 of the electronic device 600 communicate with each other via a system bus.

[0051] In one embodiment, when the processor 601 executes a through-well fracture image segmentation program in the memory 602, the following steps may be implemented: Preprocessing the through-well fracture image data set to obtain a preprocessed data set; Based on the Deeplabv3+ model, the channel and spatial fusion attention mechanism is added to obtain the GA_Deeplabv3+ semantic segmentation model; Training a GA_Deeplabv3+ semantic segmentation model based on the data set to obtain a fully trained GA_Deeplabv3+ semantic segmentation model; Acquire target through-well fracture images; The target through-well fracture image is input into the well-trained GA_Deeplabv3+ semantic segmentation model to obtain a segmented fracture image.

[0052] It should be understood that: when the processor 601 executes a through-well fracture image segmentation program in the memory 602 , in addition to the above functions, other functions may also be implemented, and details may refer to the description of the corresponding method embodiments above.

[0053] Furthermore, the embodiment of the present invention does not specifically limit the type of the electronic device 600 mentioned, and the electronic device 600 may be a portable electronic device such as a mobile phone, a tablet computer, a personal digital assistant (PDA), a wearable device, a laptop computer, etc. Exemplary embodiments of portable electronic devices include but are not limited to portable electronic devices equipped with IOS, Android, Microsoft or other operating systems. The above-mentioned portable electronic devices may also be other portable electronic devices, such as a laptop computer with a touch-sensitive surface (e.g., a touch panel). It should also be understood that in some other embodiments of the present invention, the electronic device 600 may not be a portable electronic device, but a desktop computer with a touch-sensitive surface (e.g., a touch panel).

[0054] Those skilled in the art will appreciate that all or part of the processes of the above-mentioned embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, wherein the computer-readable storage medium is a disk, an optical disk, a read-only storage memory, or a random access memory, etc.

[0055] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for segmenting a through-well fracture image, characterized in that: include: Preprocessing the through-well fracture image data set to obtain a preprocessed data set; Based on the Deeplabv3+ model, the channel and spatial fusion attention mechanism is added to obtain the GA_Deeplabv3+ semantic segmentation model; Training a GA_Deeplabv3+ semantic segmentation model based on the training set in the data set to obtain a fully trained GA_Deeplabv3+ semantic segmentation model; Acquire target through-well fracture images; The target through-well fracture image is input into the well-trained GA_Deeplabv3+ semantic segmentation model to obtain a segmented fracture image.

2. The method for segmenting cross-well fracture images according to claim 1, characterized in that: The GA_Deeplabv3+ semantic segmentation model is obtained by adding channel and spatial fusion attention mechanisms on the basis of the Deeplabv3+ model, including: The channel and spatial fusion attention mechanism is added to the spatial pyramid pooling module of the Deeplabv3+ model to obtain an improved spatial pyramid pooling module; The GA_Deeplabv3+ semantic segmentation model is obtained based on the improved spatial pyramid pooling module.

3. The method for segmenting cross-well fracture images according to claim 2, characterized in that: The channel and space fusion attention mechanism includes a channel fusion attention mechanism and a space fusion attention mechanism. The channel fusion attention mechanism is used to weight the input feature map in the channel dimension, and the space fusion attention mechanism is used to weight the input feature map in the spatial dimension.

4. The method for segmenting cross-well fracture images according to claim 5, characterized in that: The GA_Deeplabv3+ semantic segmentation model includes: an input end, a backbone feature extraction network, an improved spatial pyramid pooling module, a decoder, and an output end; The input end is used to set the resolution of the input picture to obtain a first image; The backbone feature extraction network is used to extract a first feature of the first image; The improved spatial pyramid pooling module is used to reduce the dimension of the first feature to obtain a second feature; The decoder is used for decoding the second feature to obtain a second image; The output end is used to output the second image.

5. The method for segmenting cross-well fracture images according to claim 4, characterized in that: Also includes: Gabor filters are added to both the input end and the output end.

6. The method for segmenting cross-well fracture images according to claim 1, characterized in that: Also includes: A regularization term based on a sinusoidal fitting strategy and a crack detection loss are added to the cross entropy loss function at the output end to obtain a total loss function.

7. The method for segmenting cross-well fracture images according to claim 6, characterized in that: The expression of the total loss function is: In the formula, represents the total loss function; represents the weight of the cross entropy loss function, represents the cross entropy loss function; represents the weight of crack detection loss; represents the crack detection loss, represents the weight of the regularization term based on the sine fitting strategy; represents the regularization term based on the sine fitting strategy; where: In the formula, C represents the number of categories; represents the true label, i.e. the probability of the well-crossing fracture area; It indicates the probability of the area predicted as a cross-well fracture by the GA_Deeplabv3+ semantic segmentation model; , Represents the i-th parameter of the output sine fit; Represents the i-th parameter of the target sine fit.

8. A device for separating images of cross-well fractures, characterized in that: include: A data set acquisition module is used to preprocess the through-well fracture image data set to obtain a preprocessed data set; The semantic segmentation model acquisition module is used to add channel and spatial fusion attention mechanisms to the Deeplabv3+ model to obtain the GA_Deeplabv3+ semantic segmentation model; A training segmentation model module is used to train a GA_Deeplabv3+ semantic segmentation model based on the data set to obtain a fully trained GA_Deeplabv3+ semantic segmentation model; A target image acquisition module is used to acquire a target through-well fracture image; The segmented image module is used to input the target through-well fracture image into the fully trained GA_Deeplabv3+ semantic segmentation model to obtain a segmented fracture image.

9. An electronic device, characterized in that: comprising a memory and a processor, wherein: The memory is used to store programs; The processor is coupled to the memory and is used to execute the program stored in the memory to implement the steps in the through-well fracture image segmentation method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: Used to store computer-readable programs or instructions, which, when executed by a processor, can implement the steps of a through-well fracture image segmentation method as described in any one of claims 1 to 7.

Images