A method and device for segmenting cracks in FMI images

Abstract

The present invention relates to the technical field of oil and gas exploration, and discloses an FMI image crack segmentation method and device, which comprises collecting and arranging a to-be-processed FMI image, wherein the FMI image has cracks not covered by blank areas, cracks covered by blank areas, information areas not covering cracks, and blank areas covering cracks; processing the to-be-processed FMI image using a target FMI image crack segmentation model to obtain an FMI image that has completed crack simulation and reconstruction; wherein the target FMI image crack segmentation model comprises an information area segmentation module and a crack segment connection module; the information area segmentation module performs crack segmentation on a portion of the to-be-processed FMI image containing crack information, while ignoring blank areas in the to-be-processed FMI image, to obtain a preliminary crack segmentation result; and processing the preliminary crack segmentation result based on the crack segment connection module to identify and simulate the connection of multiple dispersed crack segments belonging to the same crack in the preliminary crack segmentation result, which can achieve more accurate and more effective crack identification.

Description

Technical Field

[0001] The present invention relates to the technical field of oil and gas exploration, and in particular to a method and device for segmenting cracks in an FMI image. Background Art

[0002] The presence and distribution of underground fractures have a profound impact on the performance of oil and gas reservoirs and are directly related to the production capacity of oil and gas. Therefore, detailed inspection and analysis of these underground fractures is crucial for oil and gas exploration and development activities.

[0003] Generally speaking, traditional logging technology, image logging technology and core sample analysis are widely used to identify natural fractures in wells. Although these methods have been proven to be effective in practice, the high cost and limited nature of core data limit their widespread application. In this context, image logging technology, especially imaging-based logging methods (for example, FMI logging, where FMI stands for Formation Micro-scanner Image in English and Formation Microresistivity Scanning Imaging in Chinese), has shown its important value in fracture characterization and identification. In FMI data, fractures usually appear as dark continuous lines, providing an intuitive basis for fracture identification. However, traditional fracture identification methods often rely on manual analysis by experts, which is not only time-consuming and inefficient, but may also affect the accuracy of identification due to human factors.

[0004] With the development of technology and the improvement of automation level, developing a method that can automatically extract fracture information has become a research hotspot. This method can effectively improve the accuracy and efficiency of fracture identification, reduce labor costs, and has important practical significance for decision-making in oil and gas exploration and development. Researchers at home and abroad have proposed a variety of fracture identification methods based on digital image processing and machine learning technology, which have shown great potential in improving the accuracy and efficiency of fracture identification.

[0005] Although these early attempts laid the foundation for the development of fracture identification technology, they also clearly pointed out the need for further research and development of more advanced and robust fracture detection methods to better meet the practical challenges of well logging image analysis. At the same time, with the rapid development of deep learning technology, researchers have begun to explore the use of these advanced technologies to overcome the limitations of traditional methods, thereby achieving more accurate and effective fracture identification. Summary of the invention

[0006] Based on the above, the object of the present invention is to provide a method and device for FMI image crack segmentation, which can achieve more accurate and effective crack identification.

[0007] To achieve the above object, the present invention adopts the following technical solutions:

[0008] The first aspect of the present application provides a FMI image crack segmentation method, comprising:

[0009] Collecting and arranging the FMI images to be processed, wherein the FMI images show cracks not covered by blank areas, cracks covered by blank areas, information areas not covering the cracks, and blank areas covering the cracks;

[0010] Processing the to-be-processed FMI image using the target FMI image crack segmentation model to obtain an FMI image that completes the crack simulation reconstruction;

[0011] Among them, the target FMI image crack segmentation model includes an information area segmentation module and a crack segment connection module; the information area segmentation module is used to perform crack segmentation on the part containing crack information in the FMI image to be processed, while ignoring the blank area in the FMI image to be processed to obtain a preliminary crack segmentation result; the crack segment connection module is used to process the preliminary crack segmentation result to identify and simulate the connection of multiple scattered crack segments belonging to the same crack in the preliminary crack segmentation result.

[0012] In some embodiments, before using the target FMI image crack segmentation model to process the to-be-processed FMI image, the method further includes training and obtaining the target FMI image crack segmentation model, wherein training and obtaining the target FMI image crack segmentation model includes:

[0013] Collect and organize the FMI images to be tested;

[0014] Using the FMI image to be tested to create a training data set, a verification data set and a test data set, and annotating the FMI image to be tested;

[0015] The constructed training data set, the validation data set and the test data set are standardized, and a TSNet deep learning model is constructed as an initial FMI image crack segmentation model;

[0016] Based on the training data set and the validation data set, the initial FMI image crack segmentation model is optimized using a stochastic gradient descent method, and iterated until a stable and accurate optimized FMI image crack segmentation model is obtained;

[0017] Using the optimized FMI image crack segmentation model to test and evaluate the FMI images in the test data set to obtain prediction results;

[0018] The prediction results are compared and analyzed with the labeled crack segmentation data of the test data set to determine the target FMI image crack segmentation model.

[0019] In some embodiments, the optimizer used to optimize the initial FMI image crack segmentation model is AdamW; or,

[0020] Comparing and analyzing the prediction results with the labeled crack segmentation data of the test data set to determine that the target FMI image crack segmentation model includes:

[0021] Displaying the crack segmentation results of the optimized FMI image crack segmentation model on the FMI image;

[0022] Visually displaying the intermediate outputs of the information region segmentation module and the crack segment connection module;

[0023] The labeled crack segmentation data of the test data set is compared and analyzed to verify the accuracy and performance of the FMI image crack segmentation and determine the target FMI image crack segmentation model.

[0024] In some implementations, labeling the FMI image to be tested includes:

[0025] The cracks not covered by the blank area, the cracks covered by the blank area, and the blank area covering the cracks are marked.

[0026] In some implementations, using the FMI image to be tested to create a training data set, a verification data set, and a test data set, and annotating the FMI image to be tested includes:

[0027] The area containing cracks in the FMI image to be tested is cut into matrix blocks, and data cleaning is performed to obtain cleaned sample data;

[0028] The cleaned sample data are labeled, and the cleaned sample data are divided into the training data set, the verification data set and the test data set according to a preset ratio.

[0029] In some embodiments, the standardizing the constructed training data set, validation data set, and test data set includes:

[0030] The constructed training data set, the validation data set and the test data set are standardized using a standardized processing formula, wherein the standardized processing formula is:

[0031]

[0032] In the formula, x norm is the standardized value; x is the input image pixel in RGB format; mean and std use the mean and variance calculated on the million-level ImageNet dataset.

[0033] In some embodiments, the target FMI image crack segmentation model further includes a block encoding module, a bimodal architecture, a deformable convolution module, and a segmentation module.

[0034] In some embodiments, the FMI image to be processed includes a dynamic FMI image to be processed and a static FMI image to be processed, the information region segmentation module includes a first Stem module, a first feature encoding module, a second feature encoding module, and a first feature decoding module, the first Stem module includes a plurality of convolutional layers, a normalization layer, and an activation layer, the first feature encoding module and the second feature encoding module are respectively composed of four layers of basic encoding blocks and downsampling layers alternately, each basic encoding block integrates a plurality of perceptron modules, a normalization layer, and a deformable convolutional layer, the sampling layer is composed of a normalization layer and a convolutional layer with a stride of 2, the first feature decoding module is composed of four layers of basic encoding blocks and upsampling layers alternately, each basic encoding block integrates a plurality of perceptron modules, a normalization layer, and a deformable convolutional layer, the sampling layer is composed of a normalization layer and a convolutional layer with a stride of 2; wherein the information region segmentation module is used to perform crack segmentation on the portion of the FMI image to be processed containing crack information, while ignoring the blank area in the FMI image to be processed, so as to obtain a preliminary crack segmentation result, including:

[0035] Input the dynamic FMI image to be processed and the static FMI image to be processed into the first Stem module respectively, and extract the first feature information and the second feature information which are crucial for subsequent model training;

[0036] Introducing the first feature information into a first feature encoding module, introducing the second feature information into a second feature encoding module, performing feature extraction and dimension mapping processes, and obtaining first high-dimensional feature information corresponding to the first feature information and second high-dimensional feature information corresponding to the second feature information;

[0037] The first high-dimensional feature information and the second high-dimensional feature information are aggregated and sent to the first feature decoding module for decoding processing to obtain a preliminary crack segmentation result of the FMI image information area.

[0038] In some embodiments, the fracture segment connection module includes a second Stem module, a third feature encoding module, and a second feature decoding module; wherein the fracture segment connection module is used to process the preliminary fracture segmentation result to identify and simulate the connection of multiple scattered fracture segments belonging to the same fracture in the preliminary fracture segmentation result, including:

[0039] The preliminary fracture segmentation results are processed using the Argmax operation to obtain input data;

[0040] The input data is sent to the second Stem module for preliminary feature mapping to obtain mapping feature data;

[0041] Passing the mapped feature data to a third feature encoding module for calculation and refinement to obtain refined feature data;

[0042] The refined feature data is sent to the second feature decoding module to achieve overall improvement and reconstruction of the crack segmentation results.

[0043] The second aspect of the present application also provides an FMI image crack segmentation device, which includes:

[0044] at least one processor; and,

[0045] a memory communicatively connected to the at least one processor; wherein,

[0046] The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the FMI image crack segmentation method of any of the above embodiments.

[0047] The beneficial effects of the present invention include:

[0048] 1) Aiming at the characteristic that the inevitable blank bands in the FMI image to be processed hinder the continuity of cracks, a crack segmentation model for the target FMI image is constructed based on a simple and effective TSNet deep learning model. It uses two independent stages to decouple the crack segmentation of the FMI image to be processed. First, in the first stage, the information region segmentation module is used to perform accurate crack segmentation on the part of the FMI image to be processed that contains crack information; then, in the second stage, the crack segments disconnected at the blank band are connected in a simulated generation manner based on the crack segment connection module, that is, the cracks covered by the blank area are connected. This method can effectively reconstruct the integrity of the cracks and improve the continuity of the segmentation, alleviating the problem of cracks being covered by the blank area.

[0049] 2) The FMI image crack segmentation model also integrates a deformable convolution operator, which can adaptively adjust the sampling position of the convolution kernel parameters for image information according to the different shapes of the cracks, thereby improving the recognition and segmentation capabilities of complex crack shapes and improving the accuracy and efficiency of crack segmentation. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the contents of the embodiments of the present invention and these drawings without paying any creative work.

[0051] Figure 1 A schematic diagram of the steps of the FMI image crack segmentation method in an embodiment of the present invention;

[0052] Figure 2 A schematic diagram of the steps of training and obtaining a target FMI image crack segmentation model in an embodiment of the present invention;

[0053] Figure 3 The overall architecture of the target FMI image crack segmentation model proposed in the embodiment of the present invention;

[0054] Figure 4 It is a partial FMI image crack segmentation effect of the target FMI image crack segmentation model on the test data set in an embodiment of the present invention;

[0055] Figure 5 These are the respective output results of a partial FMI image passing through two stages of the target FMI image crack segmentation model in an embodiment of the present invention. DETAILED DESCRIPTION

[0056] In order to make the technical problems solved by the present invention, the technical solutions adopted and the technical effects achieved clearer, the technical solutions of the embodiments of the present invention will be further described in detail below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0057] In the description of the present invention, unless otherwise clearly specified and limited, the terms "connected", "connected", and "fixed" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0058] In the present invention, unless otherwise clearly specified and limited, a first feature being "above" or "below" a second feature may include that the first and second features are in direct contact, or may include that the first and second features are not in direct contact but are in contact through another feature between them. Moreover, a first feature being "above", "above" and "above" a second feature includes that the first feature is directly above and obliquely above the second feature, or simply indicates that the first feature is higher in level than the second feature. A first feature being "below", "below" and "below" a second feature includes that the first feature is directly below and obliquely below the second feature, or simply indicates that the first feature is lower in level than the second feature.

[0059] In the description of this embodiment, the terms "upper", "lower", "left", "right", etc., and the orientation or position relationship are based on the orientation or position relationship shown in the drawings, and are only for the convenience of description and simplified operation, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first" and "second" are only used to distinguish in the description and have no special meaning.

[0060] refer to Figure 1 , an embodiment of the present application provides a FMI image crack segmentation method, comprising:

[0061] Step 100, collect and organize the FMI images to be processed, wherein the FMI images show cracks not covered by blank areas, cracks covered by blank areas, information areas not covering cracks, and blank areas covering cracks; specifically, refer to Figure 3 The FMI images to be processed include dynamic FMI images to be processed and static FMI images to be processed, which can be obtained by the MAXIS-500 logging series of Schlumberger. In addition, the dynamic FMI images are also referred to as FMI_DYN images throughout the text, and the static FMI images are also referred to as FMI_STAT images.

[0062] Step 200, using the target FMI image crack segmentation model to process the FMI image to be processed to obtain an FMI image that completes crack simulation and reconstruction; wherein the target FMI image crack segmentation model includes an information region segmentation module and a crack segment connection module; the information region segmentation module is used to perform crack segmentation on the part of the FMI image to be processed that contains crack information, while ignoring the blank area in the FMI image to be processed to obtain a preliminary crack segmentation result; the crack segment connection module is used to process the preliminary crack segmentation result to identify and simulate the connection of multiple scattered crack segments belonging to the same crack in the preliminary crack segmentation result.

[0063] The implementation method of the present application aims at the characteristic that the blank bands inevitably generated in the FMI image to be processed hinder the continuity of the cracks, and constructs a target FMI image crack segmentation model based on a simple and effective TSNet deep learning model, which uses two independent stages to decouple the crack segmentation of the FMI image to be processed. Specifically, first, in the first stage, the information area segmentation module is used to perform accurate crack segmentation on the part containing crack information in the FMI image to be processed; then, in the second stage, the crack segments disconnected at the blank band are connected in a simulated generation manner based on the crack segment connection module, that is, the cracks covered by the blank area are connected in a simulated generation manner. This method can effectively reconstruct the integrity of the cracks and improve the continuity of the segmentation, alleviating the problem of cracks being covered by blank areas.

[0064] In addition, in step 200, before the target FMI image crack segmentation model is used to process the FMI image to be processed, the embodiment of the present application further includes step 000, training and obtaining the target FMI image crack segmentation model; wherein, reference Figure 2 Step 000, training and obtaining the target FMI image crack segmentation model includes:

[0065] Step 010: Collect and organize the FMI images to be tested.

[0066] Step 020: Use the FMI image to be tested to create a training data set, a verification data set and a test data set, and annotate the FMI image to be tested.

[0067] Step 030, standardize the constructed training data set, validation data set and test data set, and build a TSNet deep learning model as the initial FMI image crack segmentation model; explanatory, the model is written in the Pytorch framework (version 1.10.0).

[0068] Step 040: Based on the training data set and the validation data set, the initial FMI image crack segmentation model is optimized using a stochastic gradient descent method, and the optimization is performed iteratively until a stable and accurate optimized FMI image crack segmentation model is obtained.

[0069] Specifically, in some embodiments, the optimizer used to optimize the initial FMI image crack segmentation model is AdamW. AdamW is an optimization algorithm, the full name of which is "AdamWeight Decay Fix", which is a variant based on the Adam optimizer. Adam is an optimization algorithm used to train deep learning models, which combines the advantages of momentum and RMSprop. AdamW modifies the weight decay part in the Adam algorithm to make the weight decay more stable during the optimization process. More specifically, the FMI image to be tested is placed in the initial FMI image crack segmentation model for 100 iterations, and the loss is calculated each time using the predicted value and the true value of the initial FMI image crack segmentation model. The model is optimized using the loss so that the model prediction value is constantly close to the true value. After the loss becomes stable, the optimized FMI image crack segmentation model can be obtained. Experimental verification shows that the optimized FMI image crack segmentation model can achieve an IoU segmentation index of 90.8% in the test data set after 100 iterations.

[0070] Step 050: Use the optimized FMI image crack segmentation model to test and evaluate the FMI images in the test data set to obtain prediction results.

[0071] Step 060: Compare and analyze the prediction results with the labeled crack segmentation data of the test data set to determine the target FMI image crack segmentation model.

[0072] Specifically, step 060, comparing and analyzing the prediction results with the labeled crack segmentation data of the test data set to determine the target FMI image crack segmentation model includes:

[0073] Step 061, displaying the crack segmentation result of the optimized FMI image crack segmentation model on the FMI image;

[0074] Step 062: Visually display the intermediate outputs of the information region segmentation module and the crack segment connection module respectively;

[0075] Step 063: Compare and analyze the labeled crack segmentation data of the test data set to verify the accuracy and performance of the FMI image crack segmentation and determine the target FMI image crack segmentation model.

[0076] refer to Figure 4-5 , Figure 4 The crack segmentation effect of some target FMI image crack segmentation models on the test data set is demonstrated. After 100 iterations, the target FMI image crack segmentation model can achieve an IoU segmentation index of 90.8% in the test data set. Figure 5The output results of some target FMI images after two stages are shown. It can be seen that the output of the target FMI image crack segmentation model not only conforms to the design idea, but also demonstrates the efficiency and accuracy of the target FMI image crack segmentation model in processing cracks in existing information areas. Through the above steps, we can not only analyze the accuracy of FMI image crack segmentation, but also analyze the capabilities of the two stages of the target FMI image crack segmentation model. More importantly, the target FMI image crack segmentation model can also intelligently simulate the generation and connection of crack segments based on the relationship between the crack segments and the morphological trends of the cracks themselves, which further proves the rationality of our design of the two stages of the target FMI image crack segmentation model and the correctness of the decoupling idea of ​​FMI image crack segmentation. Reference Figure 3-5 , Figure 3 DYN means dynamic, STAT means static; Encoder means feature encoder, which corresponds to the first feature encoding module, the second feature encoding module, and the third feature encoding module from top to bottom and from left to right; Decoder means feature decoder, which corresponds to the first feature decoding module and the second feature decoding module from left to right; Label means the label used as a reference for comparison, that is, the corresponding label crack segmentation data, that is, the real crack segmentation data. Figure 4-5 Pred represents the prediction result of the second stage; Figure 5 Seg_temp represents the segmentation prediction result of the first stage, which corresponds to the preliminary crack segmentation result.

[0077] In some implementations, the step 020 of labeling the FMI image to be tested includes:

[0078] The cracks not covered by the blank area, the cracks covered by the blank area, and the blank area covering the cracks are marked.

[0079] In some implementations, the step 020 of using the FMI image to be tested to create a training data set, a verification data set, and a test data set, and annotating the FMI image to be tested includes:

[0080] Step 021, cropping the area containing cracks in the FMI image to be tested into matrix blocks, and performing data cleaning to obtain cleaned sample data; specifically, the matrix block can be a 256x256 matrix block.

[0081] Step 022: annotate the cleaned sample data, and divide the cleaned sample data into the training data set, the validation data set, and the test data set according to a preset ratio. Specifically, the preset ratio may be 7:2:1.

[0082] In some implementations, step 030, the standardization of the constructed training data set, validation data set, and test data set includes:

[0083] The constructed training data set, validation data set and test data set are standardized using a standardized processing formula, wherein the standardized processing formula is:

[0084]

[0085] In the formula, x norm is the standardized value; x is the input image pixel in RGB format; mean and std use the mean and variance calculated on the million-level ImageNet dataset.

[0086] In some embodiments, reference Figure 3 The target FMI image crack segmentation model also includes a block encoding module, a bimodal architecture, a deformable convolution module, and a segmentation module.

[0087] In some embodiments, reference Figure 3 , the FMI image to be processed includes a dynamic FMI image to be processed and a static FMI image to be processed, the information region segmentation module includes a first Stem module, a first feature encoding module, a second feature encoding module, and a first feature decoding module, the first Stem module includes a plurality of convolutional layers, a normalization layer, and an activation layer, the first feature encoding module and the second feature encoding module are respectively composed of four layers of basic encoding blocks and downsampling layers alternately, each basic encoding block integrates a plurality of perceptron modules, a normalization layer, and a deformable convolutional layer, the sampling layer is composed of a normalization layer and a convolutional layer with a stride of 2, the first feature decoding module is composed of four layers of basic encoding blocks and upsampling layers alternately, each basic encoding block integrates a plurality of perceptron modules, a normalization layer, and a deformable convolutional layer, the sampling layer is composed of a normalization layer and a convolutional layer with a stride of 2; wherein the information region segmentation module is used to perform crack segmentation on the part containing crack information in the FMI image to be processed, while ignoring the blank area in the FMI image to be processed, so as to obtain a preliminary crack segmentation result including:

[0088] Step 210: Input the dynamic FMI image to be processed and the static FMI image to be processed into the first Stem module respectively, and extract the first feature information and the second feature information which are crucial for the subsequent model training; explanatory, the algorithm formula is expressed as follows:

[0089] X DYN ,X STAT =Stem(X DYN ),Stem(X STAT )

[0090] Where, X DYN and X STAT They represent the data information of FMI_DYN image and FMI_STAT image respectively, that is, X DYN Represents the first feature information, X STAT represents the second characteristic information;

[0091] Step 220: introduce the first feature information into the first feature encoding module, introduce the second feature information into the second feature encoding module, perform feature extraction and dimension mapping processes, obtain first high-dimensional feature information corresponding to the first feature information, and second high-dimensional feature information corresponding to the second feature information; for explanation, X DYN As an example, the algorithm formula is expressed as follows:

[0092]

[0093] In the formula, the Encoder module is the first feature encoding module, which is composed of four layers (i=4) of basic encoding blocks (Block) and downsampling layers (Down) alternating. Each Block module integrates multiple perceptron modules (MLP), normalization layers (Norm), and deformable convolution layers (DCN). At the same time, the Down layer is mainly composed of a normalization layer and a convolution layer with a stride of 2. Inside the Encoder module, the calculation process of the i-th layer is essentially a comprehensive application of the above two operations:

[0094]

[0095] In the above formula That is the first high-dimensional feature information. In addition, it can be used Represents the second highest dimensional feature information.

[0096] Step 230: aggregate the first high-dimensional feature information and the second high-dimensional feature information and send them to the first feature decoding module for decoding processing to obtain a preliminary crack segmentation result of the FMI image information area. Explanatory, the algorithm formula is expressed as follows:

[0097]

[0098] Seg temp =Decoder(X i )

[0099] Where, X i Segment is the aggregation result. tempTo represent the preliminary crack segmentation results, the Decoder module, i.e. the first feature decoding module, also uses the design of the basic coding block. However, a key difference is that it integrates an upsampling layer, which is mainly composed of a normalization layer and a transposed convolution layer with a stride of 2. The specific operation of the i-th layer is similar to that of the encoder:

[0100] X=Block i (Up i (X+X i ))

[0101] Seg temp =SegLayer(X)

[0102] In the above implementation, the information region segmentation module is characterized in that this strategy ensures that the analysis focus is concentrated on all potential crack segments in the dynamic FMI image and the static FMI image to be processed, rather than the interaction or connectivity between the crack segments.

[0103] Further, in some embodiments, reference Figure 3 The crack segment connection module includes a second Stem module, a third feature encoding module, and a second feature decoding module; wherein the crack segment connection module is used to process the preliminary crack segmentation result to identify and simulate the connection of multiple scattered crack segments belonging to the same crack in the preliminary crack segmentation result, including:

[0104] Step 240, using Argmax operation to process the preliminary crack segmentation result to obtain input data;

[0105] Step 250: Send the input data to the second Stem module for preliminary feature mapping to obtain mapping feature data;

[0106] Step 260: passing the mapped feature data to a third feature encoding module for calculation and refinement to obtain refined feature data;

[0107] Step 270: Send the refined feature data to the second feature decoding module to achieve overall improvement and reconstruction of the crack segmentation result.

[0108] Explanatory, the algorithm formula is expressed as follows:

[0109] Seg final =Decoder(Encoder(Stem(X)))

[0110] In the formula, Seg final is the final FMI image crack segmentation result, which corresponds to the overall improvement and reconstruction of the crack segmentation result, and X is the representation of the input data.

[0111] In addition, some embodiments of the present application also provide an FMI image crack segmentation device, which can achieve more accurate and effective crack identification. The FMI image crack segmentation device includes at least one processor and a memory connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the FMI image crack segmentation method mentioned in any of the above embodiments.

[0112] As described above, in the corresponding implementation, the FMI image to be processed includes a dynamic FMI image to be processed and a static FMI image to be processed. In view of the unique characteristics of the two modal images of FMI, a parallel processing strategy is adopted to make full use of the complementarity of the two data modes to improve the accuracy and efficiency of the crack segmentation task, thereby avoiding the waste of data resources. In addition, the advantage of the crack segment connection module is that it aims to achieve accurate simulation segmentation of cracks in blank areas. Specifically, although the cracks in the FMI image to be processed may be divided into multiple crack segments due to the discontinuity of the image (for example, blank bands, i.e., blank areas), these segments show similar trends, directions or forms, thereby providing clues to connect these segments. By identifying and connecting those scattered crack segments that may belong to the same crack based on the crack segment connection module, the accurate simulation segmentation of cracks covered by blank areas is achieved by capturing the intrinsic associations between these crack segments.

[0113] In summary, the target FMI image crack segmentation model proposed in the present invention is used to perform FMI image crack segmentation, which not only has high computational efficiency and can achieve good crack segmentation accuracy on the FMI image, but also further reduces the improvement of segmentation quality through two decoupled designs.

[0114] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A method for segmenting cracks in an FMI image, characterized in that: include: Collecting and arranging the FMI images to be processed, wherein the FMI images show cracks not covered by blank areas, cracks covered by blank areas, information areas not covering the cracks, and blank areas covering the cracks; Processing the to-be-processed FMI image using the target FMI image crack segmentation model to obtain an FMI image that completes the crack simulation reconstruction; The target FMI image crack segmentation model includes an information region segmentation module and a crack segment connection module; the information region segmentation module is used to perform crack segmentation on the part containing crack information in the to-be-processed FMI image, while ignoring the blank area in the to-be-processed FMI image, so as to obtain a preliminary crack segmentation result; the crack segment connection module is used to process the preliminary crack segmentation result, so as to identify and simulate the connection of multiple scattered crack segments belonging to the same crack in the preliminary crack segmentation result; The crack segment connection module includes a second Stem module, a third feature encoding module, and a second feature decoding module; wherein the crack segment connection module is used to process the preliminary crack segmentation result to identify and simulate the connection of multiple scattered crack segments belonging to the same crack in the preliminary crack segmentation result, including: The preliminary fracture segmentation results are processed using the Argmax operation to obtain input data; The input data is sent to the second Stem module for preliminary feature mapping to obtain mapping feature data; Passing the mapped feature data to a third feature encoding module for calculation and refinement to obtain refined feature data; The refined feature data is sent to the second feature decoding module to achieve overall improvement and reconstruction of the crack segmentation results.

2. The FMI image crack segmentation method according to claim 1, characterized in that: Before using the target FMI image crack segmentation model to process the to-be-processed FMI image, the method further includes training and obtaining the target FMI image crack segmentation model. The training and obtaining the target FMI image crack segmentation model includes: Collect and organize the FMI images to be tested; Using the FMI image to be tested to create a training data set, a verification data set and a test data set, and annotating the FMI image to be tested; The constructed training data set, the validation data set and the test data set are standardized, and a TSNet deep learning model is constructed as an initial FMI image crack segmentation model, wherein the model structure of the TSNet deep learning model is the same as the model structure of the target FMI image crack segmentation model; Based on the training data set and the validation data set, the initial FMI image crack segmentation model is optimized using a stochastic gradient descent method, and iterated until a stable and accurate optimized FMI image crack segmentation model is obtained; Using the optimized FMI image crack segmentation model to test and evaluate the FMI images in the test data set to obtain prediction results; The prediction results are compared and analyzed with the labeled crack segmentation data of the test data set to determine the target FMI image crack segmentation model.

3. The FMI image crack segmentation method according to claim 2, characterized in that: The optimizer used to optimize the initial FMI image crack segmentation model is AdamW; or, Comparing and analyzing the prediction results with the labeled crack segmentation data of the test data set to determine that the target FMI image crack segmentation model includes: Displaying the crack segmentation results of the optimized FMI image crack segmentation model on the FMI image; Visually displaying the intermediate outputs of the information region segmentation module and the crack segment connection module; The labeled crack segmentation data of the test data set is compared and analyzed to verify the accuracy and performance of the FMI image crack segmentation and determine the target FMI image crack segmentation model.

4. The FMI image crack segmentation method according to claim 2, characterized in that: The labeling of the FMI image to be tested includes: The cracks not covered by the blank area, the cracks covered by the blank area, and the blank area covering the cracks are marked.

5. The FMI image crack segmentation method according to claim 2, characterized in that: The step of using the FMI image to be tested to create a training data set, a verification data set and a test data set, and annotating the FMI image to be tested includes: The area containing cracks in the FMI image to be tested is cut into matrix blocks, and data cleaning is performed to obtain cleaned sample data; The cleaned sample data are labeled, and the cleaned sample data are divided into the training data set, the verification data set and the test data set according to a preset ratio.

6. The FMI image crack segmentation method according to claim 2, characterized in that: The standardization of the constructed training data set, the validation data set and the test data set includes: The constructed training data set, the validation data set and the test data set are standardized using a standardized processing formula, wherein the standardized processing formula is: In the formula, χ norm is the value after normalization; χ is the input image pixel in RGB format; mean and std are the mean and variance calculated on the million-level ImageNet dataset.

7. The FMI image crack segmentation method according to any one of claims 1 to 6, characterized in that: The FMI image to be processed includes a dynamic FMI image to be processed and a static FMI image to be processed. The information region segmentation module includes a first Stem module, a first feature encoding module, a second feature encoding module, and a first feature decoding module. The first Stem module includes a plurality of convolutional layers, a normalization layer, and an activation layer. The first feature encoding module and the second feature encoding module are respectively composed of four layers of basic encoding blocks and downsampling layers alternately. Each basic encoding block integrates a plurality of perceptron modules, a normalization layer, and a deformable convolutional layer. The sampling layer is composed of a normalization layer and a convolutional layer with a stride of 2. The first feature decoding module is composed of four layers of basic encoding blocks and upsampling layers alternately. Each basic encoding block integrates a plurality of perceptron modules, a normalization layer, and a deformable convolutional layer. The sampling layer is composed of a normalization layer and a convolutional layer with a stride of 2. The information region segmentation module is used to perform crack segmentation on the portion of the FMI image to be processed that contains crack information, while ignoring the blank area in the FMI image to be processed, so as to obtain a preliminary crack segmentation result, including: Input the dynamic FMI image to be processed and the static FMI image to be processed into the first Stem module respectively, and extract the first feature information and the second feature information which are crucial for subsequent model training; Introducing the first feature information into a first feature encoding module, introducing the second feature information into a second feature encoding module, performing feature extraction and dimension mapping processes, and obtaining first high-dimensional feature information corresponding to the first feature information and second high-dimensional feature information corresponding to the second feature information; The first high-dimensional feature information and the second high-dimensional feature information are aggregated and sent to the first feature decoding module for decoding processing to obtain a preliminary crack segmentation result of the FMI image information area.

8. An FMI image crack segmentation device, characterized in that: include: at least one processor; as well as, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the FMI image crack segmentation method according to any one of claims 1 to 7.

Images