Electric imaging image crack parameter quantitative calculation method for small sample training

Abstract

The invention relates to a small sample training-oriented electric imaging image crack parameter quantitative calculation method. Comprising the steps of obtaining an electric imaging logging image sample of a target work area, and performing preprocessing and data enhancement; constructing a crack segmentation model based on a fusion channel and a spatial cross attention mechanism; inputting the electric imaging logging image sample of the target work area after preprocessing and data enhancement into a crack segmentation model, and training the model by adopting a meta-learning training strategy; inputting a to-be-processed target electric imaging image into the trained crack segmentation model, and outputting a crack binarization segmentation image; and performing morphological post-processing on the crack binarization segmentation image, and calculating pixel information of the crack to obtain a crack parameter. And under the condition of extremely scarce labeled samples, high-precision and high-generalization crack identification and end-to-end full-automatic calculation from an original FMI image to quantitative geological parameters are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of geophysical well logging image processing, in particular to an electric imaging image fracture parameter quantitative calculation method for small sample training. BACKGROUND

[0002] Electric imaging logging (FMI) is one of the intuitive and effective means to identify fractures downhole. In recent years, the deep learning semantic segmentation technology has been tried to be used for automatic fracture identification of FMI images, which has shown great application potential, but the existing related methods have two core bottlenecks: one is the data dependency problem: the supervised learning model needs a large amount of high-quality labeled data to support training, and the expert labeling of FMI images is highly professional, high-cost and long-cycle, and the actual application often faces the "small sample" dilemma of insufficient labeled samples, which limits the training effect of the model. The second is the model generalization problem: the model trained based on the data of a certain work area or formation has a sharp decline in performance when directly applied to a new work area with different geological conditions, and lacks cross-area generalization ability. At the same time, the existing technology depends on a large amount of data or has unsatisfactory recognition accuracy in a small sample scene, and most of them only complete the "fracture identification" step and do not form a complete technical closed loop with "quantitative calculation" of geological parameters, which is difficult to meet the actual demand.

[0003] In the existing small sample fracture identification related technology, some schemes obtain a fracture query set and a support set, perform data augmentation, extract global features of the query set, and then perform bidirectional fusion through a network, and extract global variables of a region of interest in the support set, and then input a prediction module to train a model, so as to improve the recognition effect under a small amount of samples. However, such technology still has the following problems: there are semantic differences between different feature channel layers, and the model is difficult to fully capture multi-scale fracture information; too much attention is paid to the environment background, which reduces the recognition accuracy to a certain extent; although bidirectional fusion of features is realized, the focusing ability on fractures is insufficient under complex background, and the recognition effect is easily reduced when the background interference is strong.

[0004] Therefore, in the related art, there is an urgent need for a method capable of improving the accuracy and reliability of fracture identification. SUMMARY

[0005] Therefore, in the related art, there is an urgent need for a method capable of improving the accuracy and reliability of fracture identification.

[0006] In a first aspect, the present application provides an electric imaging image fracture parameter quantitative calculation method for small sample training. The method comprises: obtaining a target work area electric imaging logging image sample and performing preprocessing and data enhancement; A crack segmentation model based on a fusion channel and spatial cross attention mechanism is constructed, including a feature extractor, a channel attention submodule, a spatial cross attention submodule, a feature fusion module, and a prediction upsampling module. The target work area electric imaging logging image sample after the preprocessing and data enhancement is input into the crack segmentation model based on the fusion channel and spatial cross attention mechanism, and a meta-learning training strategy is used to train the model. The target electric imaging image to be processed is input into the trained crack segmentation model based on the fusion channel and spatial cross attention mechanism, and a crack binary segmentation image is output. The crack binary segmentation image is subjected to morphological post-processing, and the pixel information of the crack is calculated to obtain crack parameters.

[0007] Optionally, in an embodiment of the present application, the preprocessing includes: The blank strip region is filled by using the Criminisi algorithm.

[0008] Optionally, in an embodiment of the present application, the data enhancement includes: The target work area electric imaging logging image sample is segmented into a plurality of semantically independent blocks by using a superpixel segmentation algorithm. At least one of rotation, scaling, and gray scale transformation is performed on the blocks.

[0009] Optionally, in an embodiment of the present application, the feature extractor uses a ResNet-50 network, and the regular convolution in the last two stages is replaced by dilated convolution.

[0010] Optionally, in an embodiment of the present application, the channel attention submodule simultaneously performs global average pooling and maximum pooling on the input features, and generates a channel attention vector through a shared multilayer perceptron.

[0011] Optionally, in an embodiment of the present application, the spatial cross attention submodule generates query, key, and value tensors through three 1x1 convolutions on the input features, calculates row spatial attention and column spatial attention, and generates a spatial attention map.

[0012] Optionally, in an embodiment of the present application, the morphological post-processing of the crack binary segmentation image includes: The crack binary segmentation image is sequentially subjected to morphological opening and closing operations and noise removal processing. A thinning algorithm is used to extract a single-pixel width skeleton of the crack, and burr removal processing is performed on the skeleton.

[0013] In a second aspect, the present application also provides an electric imaging image fracture parameter quantitative calculation device for small sample training. The device comprises: a data preprocessing module configured to acquire a target work area electric imaging logging image sample and perform preprocessing and data enhancement; a fracture segmentation model construction module configured to construct a fracture segmentation model based on a fusion channel and spatial cross attention mechanism, including a feature extractor, a channel attention submodule, a spatial cross attention submodule, a feature fusion module, and a prediction upsampling module; a small sample training module configured to input the target work area electric imaging logging image sample after preprocessing and data enhancement into the fracture segmentation model based on the fusion channel and spatial cross attention mechanism, and train the model using a meta-learning training strategy; a fracture segmentation module configured to input a target electric imaging image to be processed into the trained fracture segmentation model based on the fusion channel and spatial cross attention mechanism, and output a fracture binary segmentation image; a parameter calculation module configured to perform morphological post-processing on the fracture binary segmentation image, and calculate pixel information of the fracture to obtain fracture parameters.

[0014] In a third aspect, the present application also provides a computer device. The computer device comprises a memory and a processor, the memory stores a computer program, and the processor executes the steps of the method described in each of the above embodiments.

[0015] In a fourth aspect, the present application also provides a computer readable storage medium. The computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the method described in each of the above embodiments.

[0016] The above-mentioned electric imaging image fracture parameter quantitative calculation method for small sample training has the following improvements: 1) Greatly reduce data dependence: through the combination of meta-learning and strong semantic data enhancement, excellent performance can be achieved by training on a small number of labeled samples, solving the industry core pain point of FMI image labeling difficulty.

[0017] 2) High precision and robustness: the model with the fusion attention mechanism can focus on key features; the professional post-processing process ensures the accuracy of parameter calculation, and the comprehensive average error can be as low as 3.16%.

[0018] 3) Strong generalization ability: the meta-learning paradigm enables the model to have the ability to "learn to learn", which can quickly adapt to new and unseen work area data, reducing the cost of repeated development of the model.

[0019] 4) Complete automation process: provides one-stop solution from raw FMI images to final quantitative geological parameters, greatly improves work efficiency and reduces subjective interpretation of human interpretation. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 A flowchart of a small sample training-oriented electro-imaging image fracture parameter quantitative calculation method in an embodiment; Figure 2 A structural diagram of a fracture segmentation model based on a fusion channel and spatial cross-attention mechanism in an embodiment; Figure 3 A module diagram of fusing a channel and a spatial cross-attention mechanism in an embodiment; Figure 4 A schematic diagram of a meta-learning training strategy in an embodiment; Figure 5 A structural block diagram of a small sample training-oriented electro-imaging image fracture parameter quantitative calculation device in an embodiment; Figure 6 An internal structure diagram of a computer device in an embodiment. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0022] In an embodiment, as shown in Figure 1 , a small sample training-oriented electro-imaging image fracture parameter quantitative calculation method is provided, comprising the following steps: S101: Obtain target work area electro-imaging logging image samples and perform preprocessing and data enhancement.

[0023] In the embodiments of the present application, first, a small number of electro-imaging (FMI) logging image samples with fracture labels of the target work area are obtained to form a support set, and the images in the support set are preprocessed and data enhanced. The blank strip region in the image caused by the instrument is filled based on an image repair algorithm, and the strong semantic data enhancement of the image is performed.

[0024] Specifically, in an embodiment of the present application, the preprocessing includes: The Criminisi algorithm is used to fill the blank strip region.

[0025] In an embodiment of the present application, for the instrument blank strip inherent in the FMI image, a sample block-based Criminisi algorithm is used for repair filling, specifically, a mask image of the blank strip is generated; a pixel block at the edge of the mask image is taken as a sample, and the most matched sample block is searched in the effective area of the image; the matched sample block is filled into the blank strip area; and the process is iterated until all the blank strips are filled completely.

[0026] In an embodiment of the present application, the data enhancement includes: The target work area electrical imaging logging image sample is segmented into a plurality of semantic-independent blocks by using a superpixel segmentation algorithm. At least one of rotation, scaling and gray scale transformation is performed on the blocks.

[0027] In an embodiment of the present application, first, the image is segmented into a plurality of semantic-independent blocks by using a superpixel segmentation algorithm, and then at least one of rotation, scaling and gray scale transformation is performed on the blocks, so as to greatly expand the training support set, to ensure that the enhanced image does not damage the semantic integrity of the fracture and the geological structure features, and to lay a foundation for subsequent small sample learning; finally, all the images are uniformly scaled to a size of 256x256 pixels to meet the size requirement of the model input. The superpixel segmentation algorithm is a simple linear iterative clustering (SLIC) algorithm.

[0028] S102: Construct a fracture segmentation model based on a fusion channel and spatial cross attention mechanism, including a feature extractor, a channel attention submodule, a spatial cross attention submodule, a feature fusion module and a prediction upsampling module.

[0029] As shown in FIG. 1, Figure 2 based on a full convolutional network (FCN) architecture, a fracture segmentation model is constructed which fuses a channel and spatial cross attention (FCSA) module, wherein the channel and spatial cross attention module is specially designed for small sample learning, as shown in FIG. 2, which is divided into a channel attention submodule and a spatial cross attention submodule, and an input feature map is first generated into an intermediate feature map Figure 3 by a channel attention submodule, and then input into a spatial cross attention submodule; the spatial cross attention submodule generates a spatial attention map by calculating the relevance of the same row and column positions in the query tensor and the key tensor.

[0030] ​The model further comprises a feature extractor, a feature fusion module, and a prediction up-sampling module. The feature fusion module uses a feature pyramid network (FPN) structure to fuse the multi-scale features extracted by the backbone network, including the high-level semantic features enhanced by the FCSA, which can greatly enhance the response to linear crack features and effectively suppress complex geological background noise. Finally, a prediction up-sampling module outputs a prediction result with the same resolution as the original image. The prediction up-sampling module outputs a prediction result with the same resolution as the original image.

[0031] Specifically, in an embodiment of the present application, the feature extractor uses a ResNet-50 network, and replaces the regular convolution in the last two stages with dilated convolution.

[0032] In an embodiment of the present application, the backbone network uses ResNet-50 as the feature extractor, and replaces the regular convolution in the last two stages with dilated convolution (Dilated Convolution). While maintaining the resolution of the feature map, the receptive field is greatly expanded, enabling it to capture large-scale crack context information in the FMI image. For a two-dimensional input signal X and a filter W, the output of the dilated convolution formula at position is:

[0033] where r is the dilation rate (Dilation Rate), and when r = 1, it is a standard convolution. Assuming the size of the filter is , is the index coordinate inside the filter W, usually ranging from 0 to k-1, is the weight value of the filter at position (m, n), which is a parameter that the model needs to learn through training.

[0034] In an embodiment of the present application, the channel attention sub-module performs global average pooling and max pooling on the input features simultaneously, and generates a channel attention vector through a shared multi-layer perceptron.

[0035] In an embodiment of the present application, as shown in Figure 3 , the channel attention sub-module (CAM) performs global average pooling and max pooling on the input features , and generates a channel attention vector through a shared multi-layer perceptron (MLP).

[0036]

[0037] where is the Sigmoid activation function.​

[0038] In an embodiment of the present application, the spatial cross-attention sub-module generates query, key, value tensors for the input features through three 1x1 convolutions, and calculates row and column spatial attentions to generate a spatial attention map.

[0039] In an embodiment of the present application, as shown in Figure 3 The spatial cross-attention sub-module (SCAM) takes the channel attention feature as input, generates query (Q), key (K), value (V) tensors through three convolutions. The core is to calculate row and column spatial attentions.

[0040]

[0041]

[0042]

[0043] The calculation amount of this sub-module is much smaller than that of full spatial attention, and its row and column attention patterns are highly consistent with the linearly distributed crack morphology in the FMI image, which can greatly enhance the ability to capture linear features.

[0044] S103: input the target work area electric imaging logging image sample after preprocessing and data enhancement into the crack segmentation model based on the fusion of channel and spatial cross-attention mechanism, and train the model using a meta-learning training strategy.

[0045] In an embodiment of the present application, as shown in Figure 4 The meta-learning training strategy is a Prototypical Network strategy, and the process includes: in each training iteration, randomly extract N classes from the training set, extract K samples for each class to form a support set, and calculate the feature prototype of each class through the feature extractor of the neural network; then extract a query set from the remaining samples, and optimize the entire network model parameters by minimizing the Euclidean distance between the feature of the query sample and the prototype of the belonging class, and the loss function is the negative log-likelihood loss:

[0046] wherein, is the loss value of the entire meta-learning task, is all trainable parameters of the feature extraction network f, x is a sample data of the query set, is the feature vector obtained after the sample x passes through the feature extraction network with parameters , and c is the true class label of the sample x, It is the mean of the feature vectors of all samples belonging to class c in the support set.

[0047] It should be noted that before model training, there is a pre-training step, which uses the publicly available large crack dataset (Crack500) to pre-train the crack segmentation model based on the fusion channel and spatial cross-attention mechanism, and then uses the support set to fine-tune the pre-trained model. After model training, there is a semi-supervised self-training step, which uses the model's high confidence prediction results for a large number of unlabeled FMI images as pseudo-labels and adds them to the support set, and repeatedly performs strong semantic data augmentation and meta-learning training.

[0048] S104: Input the target electro-imaging image to be processed into the trained crack segmentation model based on the fusion channel and spatial cross-attention mechanism, and output the crack binarized segmentation image.

[0049] S105: Perform morphological post-processing on the binarized segmented image of the crack and calculate the pixel information of the crack to obtain the crack parameters.

[0050] In this embodiment, the obtained binary segmented fracture image is subjected to morphological post-processing to obtain a clean fracture skeleton image. Based on the clean fracture skeleton image, the fracture pixel length is calculated by traversing and accumulating the Euclidean distance between skeleton pixels. Based on the binary segmented image, the total number of pixels in the fracture region is counted to calculate the fracture pixel area. With the skeleton pixel as the midpoint, a perpendicular line is drawn to the fracture edge in the binary image, and the average value of the perpendicular distance is calculated as the fracture pixel width. Finally, combined with the inherent scale factor of the FMI image (determined by the well diameter and imaging range), the above pixel information is converted into actual physical parameters with clear geological significance (such as meters and millimeters).

[0051] In one embodiment of this application, the morphological post-processing of the binarized segmented crack image includes: The binarized segmented image of the crack is subjected to morphological opening and closing operations and noise removal processing in sequence; A thinning algorithm is used to extract the single-pixel width skeleton of the crack, and the skeleton is then subjected to burr removal processing.

[0052] In one embodiment of this application, morphological opening and closing operations are sequentially performed on the binarized segmented image of the crack to connect the fracture cracks and remove noise. Then, the Hilditch thinning algorithm is used to extract the single-pixel width skeleton of the crack, and the skeleton is subjected to burr removal processing. The endpoints in the skeleton are identified, and short connected branches connected to the endpoints and with a length lower than a preset threshold are deleted to obtain a clean crack skeleton image.

[0053] In one embodiment of this application, the method of the present invention was rigorously and quantitatively compared with widely used benchmark models (U-Net) and current mainstream models (DeepLabV3+) on the same FMI crack test set. The results show that the method of the present invention achieves significant advantages in several core evaluation metrics. Specifically, in the key metric of segmentation accuracy, the intersection-over-union ratio (IoU), the method of the present invention achieved 75.78%, significantly higher than U-Net's 68.21% and DeepLabV3+'s 71.05%, demonstrating its more accurate identification and localization of crack pixels. In terms of accuracy, reflecting the overall pixel classification accuracy, the method of the present invention also outperformed the comparison models (U-Net: 92.15%, DeepLabV3+: 93.88%) with an excellent performance of 94.73%, showcasing its superior model reliability.

[0054] Furthermore, the method of this invention achieved average pixel accuracy (mPA) of 89.42% and F1-Score of 85.16% respectively, which are both significantly better than the comparison model, in evaluating class balance. This not only demonstrates its more balanced ability to distinguish between "cracks" and "background" targets, but also reflects its comprehensive advantage in reducing false positives and false negatives. In particular, the method of this invention achieved the highest score of 85.90% on the Dice coefficient, which is directly related to the accuracy of subsequent parameter quantification. This indicates that its predicted segmentation results have a high degree of overlap and consistency with the actual crack morphology, laying a solid foundation for subsequent extraction of crack skeletons and calculation of geometric parameters (such as length, width, and area).

[0055] In the aforementioned method for quantitative calculation of fracture parameters in electro-imaging images oriented to small-sample training, firstly, electro-imaging logging image samples of the target work area are acquired and preprocessed and data-enhanced; then, a fracture segmentation model based on a fusion channel and spatial cross-attention mechanism is constructed, including a feature extractor, a channel attention submodule, a spatial cross-attention submodule, a feature fusion module, and a prediction upsampling module; next, the preprocessed and data-enhanced electro-imaging logging image samples of the target work area are input into the fracture segmentation model based on the fusion channel and spatial cross-attention mechanism, and the model is trained using a meta-learning training strategy; then, the target electro-imaging image to be processed is input into the trained fracture segmentation model based on the fusion channel and spatial cross-attention mechanism, and a binary segmented fracture image is output; finally, morphological post-processing is performed on the binary segmented fracture image, and the pixel information of the fracture is calculated to obtain the fracture parameters. An improved method for quantitative calculation of fracture parameters in small-sample electro-imaging images based on meta-learning and multi-scale attention is provided, including a data preprocessing and strong semantic enhancement part, a small-sample feature extraction and meta-training part, and a fracture parameter quantitative calculation part. Through the original FCSA attention module and meta-learning training strategy, it can achieve high-precision and high-generalization crack identification under the condition of extremely scarce labeled samples; by designing a complete "morphological processing -> skeleton extraction -> parameter calculation" post-processing pipeline, it has for the first time realized end-to-end fully automatic calculation from raw FMI images to quantitative geological parameters.

[0056] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0057] Based on the same inventive concept, this application also provides a device for quantitatively calculating crack parameters in electro-imaging images based on few-sample training, used to implement the aforementioned method for quantitatively calculating crack parameters in electro-imaging images based on few-sample training. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the device for quantitatively calculating crack parameters in electro-imaging images based on few-sample training provided below can be found in the limitations of the method for quantitatively calculating crack parameters in electro-imaging images based on few-sample training described above, and will not be repeated here.

[0058] In one embodiment, such as Figure 5 As shown, a device 500 for quantitative calculation of crack parameters in electro-imaging images for few-sample training is provided, including: a data preprocessing module 501, a crack segmentation model construction module 503, a few-sample training module 505, a crack segmentation module 507, and a parameter calculation module 509, wherein: The data preprocessing module 501 is used to acquire electrical imaging logging image samples of the target work area and perform preprocessing and data augmentation.

[0059] The crack segmentation model construction module 503 is used to construct a crack segmentation model based on the fusion channel and spatial cross-attention mechanism, including a feature extractor, a channel attention submodule, a spatial cross-attention submodule, a feature fusion module, and a prediction upsampling module.

[0060] The small sample training module 505 is used to input the preprocessed and data-enhanced target area electrical imaging logging image samples into a fracture segmentation model based on a fusion channel and spatial cross-attention mechanism, and to train the model using a meta-learning training strategy.

[0061] The crack segmentation module 507 is used to input the target electro-imaging image to be processed into a trained crack segmentation model based on the fusion channel and spatial cross-attention mechanism, and output a crack binarized segmentation image.

[0062] The parameter calculation module 509 is used to perform morphological post-processing on the binary segmented image of the crack and calculate the pixel information of the crack to obtain the crack parameters.

[0063] In one embodiment of this application, the data preprocessing module is further configured to: The Criminisi algorithm is used to fill the blank strip regions.

[0064] In one embodiment of this application, the data preprocessing module is further configured to: The target work area electrical imaging logging image samples are segmented into multiple semantically independent blocks using a superpixel segmentation algorithm; Perform at least one of the following operations on the block: rotation, scaling, and grayscale transformation.

[0065] In one embodiment of this application, the feature extractor employs a ResNet-50 network, and replaces the conventional convolutions in the last two stages with dilated convolutions.

[0066] In one embodiment of this application, the channel attention submodule performs global average pooling and max pooling on the input features simultaneously, and generates a channel attention vector through a shared multilayer perceptron.

[0067] In one embodiment of this application, the spatial cross-attention submodule generates query, key, and value tensors from the input features through three 1×1 convolutions, and calculates row spatial attention and column spatial attention to generate a spatial attention graph.

[0068] In one embodiment of this application, the parameter calculation module is further configured to: The binarized segmented image of the crack is subjected to morphological opening and closing operations and noise removal processing in sequence; A thinning algorithm is used to extract the single-pixel width skeleton of the crack, and the skeleton is then subjected to burr removal processing.

[0069] The modules in the aforementioned device for quantitative calculation of crack parameters in electro-imaging images for small-sample training can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0070] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 6As shown, the computer device includes a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When executed by the processor, the computer program implements a method for quantitative calculation of crack parameters in electro-imaging images for small-sample training. The display screen can be a liquid crystal display (LCD) or an e-ink display. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0071] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0072] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0073] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0074] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0075] 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, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0076] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0077] The technical features of the above embodiments can be combined in any way. For the sake of brevity, 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.

[0078] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for quantitatively calculating crack parameters in electrical imaging images for small-sample training, characterized in that, The method includes: Acquire electrical imaging logging image samples of the target work area and perform preprocessing and data augmentation; A crack segmentation model based on the fusion of channel and spatial cross-attention mechanisms is constructed, including a feature extractor, a channel attention submodule, a spatial cross-attention submodule, a feature fusion module, and a prediction upsampling module; The preprocessed and data-enhanced target area electrical imaging logging image samples are input into a fracture segmentation model based on a fusion channel and spatial cross-attention mechanism, and the model is trained using a meta-learning training strategy. The target electro-imaging image to be processed is input into a pre-trained crack segmentation model based on a fusion channel and spatial cross-attention mechanism, and the output is a crack binarized segmentation image. Morphological post-processing is performed on the binarized segmented image of the crack, and the pixel information of the crack is calculated to obtain the crack parameters.

2. The method for quantitative calculation of crack parameters in electrical imaging images for small sample training according to claim 1, characterized in that, The preprocessing includes: The Criminisi algorithm is used to fill the blank strip regions.

3. The method for quantitative calculation of crack parameters in electrical imaging images for small sample training according to claim 1, characterized in that, The data enhancements include: The target work area electrical imaging logging image samples are segmented into multiple semantically independent blocks using a superpixel segmentation algorithm; Perform at least one of the following operations on the block: rotation, scaling, and grayscale transformation.

4. The method for quantitative calculation of crack parameters in electrical imaging images for small sample training according to claim 1, characterized in that, The feature extractor uses a ResNet-50 network, replacing the regular convolutions in the last two stages with dilated convolutions.

5. The method for quantitative calculation of crack parameters in electro-imaging images for small sample training according to claim 1, characterized in that, The channel attention submodule performs global average pooling and max pooling on the input features simultaneously, and generates channel attention vectors through a shared multilayer perceptron.

6. The method for quantitative calculation of crack parameters in electrical imaging images for small sample training according to claim 1, characterized in that, The spatial cross-attention submodule generates query, key, and value tensors from the input features through three 1×1 convolutions, and calculates row spatial attention and column spatial attention to generate a spatial attention graph.

7. The method for quantitative calculation of crack parameters in electro-imaging images for small sample training according to claim 1, characterized in that, The morphological post-processing of the binarized segmented crack image includes: The binarized segmented image of the crack is subjected to morphological opening and closing operations and noise removal processing in sequence; A thinning algorithm is used to extract the single-pixel width skeleton of the crack, and the skeleton is then subjected to burr removal processing.

8. A device for quantitatively calculating crack parameters in electro-imaging images for small-sample training, characterized in that, The device includes: The data preprocessing module is used to acquire electrical imaging logging image samples of the target work area and perform preprocessing and data augmentation. The crack segmentation model construction module is used to build a crack segmentation model based on the fusion channel and spatial cross-attention mechanism, including a feature extractor, a channel attention submodule, a spatial cross-attention submodule, a feature fusion module, and a prediction upsampling module; The small sample training module is used to input the preprocessed and data-enhanced target area electrical imaging logging image samples into the fracture segmentation model based on the fusion channel and spatial cross-attention mechanism, and to train the model using a meta-learning training strategy. The crack segmentation module is used to input the target electro-imaging image to be processed into a trained crack segmentation model based on the fusion channel and spatial cross-attention mechanism, and output a crack binarized segmentation image. The parameter calculation module is used to perform morphological post-processing on the binarized segmented image of the crack and calculate the pixel information of the crack to obtain the crack parameters.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images