Logging imaging fracture-cavity segmentation and parameter extraction method based on deep learning

Abstract

The invention discloses a logging imaging fracture-cavity segmentation and parameter extraction method based on deep learning. The method comprises the following steps: step 1, making a data set; step 2, model training and selection: inputting the data set made in the step 1 into a deep learning model for training, and generating two key outputs, namely last. Pt and best.pt, by the model in the training process; best.pt is selected as a model in the actual segmentation process; 3, further designing a client interface of the logging imaging crack segmentation and parameter extraction system on the basis of completing data set production and model training and selection; step 4, crack image processing and curve fitting algorithm implementation, and crack inclination angle parameter extraction; and 5, opening a client interface, selecting a data source, starting detection, outputting a result, and displaying the result. According to the invention, crack inclination angle parameter information can be accurately extracted; and an integrated and visual operation platform is provided for a user.

Description

Technical Field

[0001] The present invention relates to the field of well logging imaging, and in particular to a well logging imaging fracture and hole segmentation and parameter extraction method based on deep learning. Background Art

[0002] Well logging imaging technology is a technology that uses various physical methods, including acoustic detection and resistivity detection, to scan and image the surrounding strata and present the detection results to users. It plays an important role in the exploration and development of mineral resources such as oil and natural gas. It can intuitively reflect the lithology and structure of the strata, help geologists understand the characteristics of underground reservoirs, and thus provide important basis for reservoir evaluation, well site deployment and mining plan formulation.

[0003] The development of well logging imaging technology from early simple logging curves to today's high-resolution imaging is of milestone significance.

[0004] Fractures and caves in the formation are the main space for oil and gas storage and migration. Accurate segmentation of fractures and caves and extraction of parameters are crucial for evaluating the permeability, oil and gas content, and production capacity of the reservoir. Traditional fracture segmentation and parameter extraction methods generally include computer vision methods such as threshold segmentation, edge detection, and morphological processing. However, this method is easily disturbed by image noise and texture. In addition, the method of estimating fracture parameters based on geostatistical principles not only requires a large amount of sample data, but may also be limited by insufficient data or complex spatial distribution. Therefore, it is of great significance to develop an automated and intelligent system for logging imaging fracture segmentation and parameter extraction. Summary of the invention

[0005] The purpose of the present invention is to provide a method for well logging imaging fracture-hole segmentation and parameter extraction based on deep learning, which can accurately understand the direction of underground fractures by automatically and accurately segmenting well logging imaging fractures and extracting fracture parameters, which is of great significance to oil and gas reservoir development, disaster prediction, etc.

[0006] Explanation of terms involved in the present invention: Ciflog: A professional logging data processing and interpretation software.

[0007] FMI: Full-bore formation microresistivity imaging tool.

[0008] ERMI: Enhanced Reservoir Monitoring Imager.

[0009] STAR-II: A logging tool that collects relevant physical parameters of the formation surrounding the wellbore.

[0010] JSON: A lightweight data exchange format that uses a text format that is completely independent of programming languages ​​to store and represent data.

[0011] TXT: Plain text file format, which is the most basic file format. It contains only text characters and does not contain any special format information.

[0012] Last.pt: The performance status of the deep learning model at the last iteration of training.

[0013] Best.pt: The version of the model that performs best during training.

[0014] Xmin, Xmax, Ymin and Ymax: the X value, Y value and minimum X value, Y value of the X-axis and Y-axis coordinates.

[0015] Labelme: An open source software for labeling deep learning training.

[0016] A method for well logging imaging fracture-cavity segmentation and parameter extraction based on deep learning, the method comprising the following steps: Step 1: Dataset preparation: First, the logging resistivity data is processed using Ciflog software to obtain a full borehole image, which is then cut to accurately fit the input requirements of the deep learning model. Finally, the imaged fractures are manually annotated according to different fracture types. Step 2: Model training and selection: Input the data set produced in step 1 into the deep learning model for training. During the training process, the model will generate two key outputs, namely last.pt and best.pt. Among them, last.pt represents the performance status of the deep learning model in the last iteration of training; and best.pt reflects the state of the model when it achieves the best segmentation effect in the entire iteration training cycle; best.pt is selected as the model in the actual segmentation process; Step 3: On the basis of completing the data set preparation, model training and selection, the client interface of the logging imaging fracture segmentation and parameter extraction system is further designed; the client interface design content includes: basic information display area, image display window, parameter setting module, detection result display module and user operation module; the client interface serves as a bridge for the user to interact with the back-end program; Step 4: Process the crack image and implement the curve fitting algorithm. Extract the coordinates of the lowest and highest points of the fitting curve, connect the two points to construct a triangle, and obtain the inclination information of the crack by calculating the angle between the two points. Step 5. Open the client interface, select the data source to start detection, output the results, and display the results.

[0017] As a further technical solution of the present invention, the data set preparation in step 1 includes: First, the logging resistivity data is processed using Ciflog software. In this process, the appropriate data source is selected, and the corresponding well and instrument are determined. The available instruments include FMI, ERMI, and STAR-II. After the selection is completed, the software will generate the original image. In order to enrich the data diversity, the original image is enhanced to generate static and dynamic images. Subsequently, the generated image is preprocessed to eliminate the common white strips in the logging image, and finally the full borehole image is obtained. In order to facilitate the input of the full borehole image into the deep learning model for deep feature extraction and learning, two scales were used to produce the image, and the full borehole image was cut according to the 2m cutting standard. The cut image can more accurately adapt to the input requirements of the deep learning model; Finally, the imaged cracks are manually annotated according to different crack types. After the annotation is completed, the corresponding JSON file is generated and then converted into a TXT file.

[0018] As a further technical solution of the present invention, in step three, the basic information display area is used to clearly display the system name and the design team name, so that the user can quickly understand the basic information and development entity of the system. The image display window includes window 1 and window 2. Window 1 is used to intuitively present the crack segmentation results, and synchronously display the detection box and the corresponding label, so that the user can clearly see the position and category of each segmented target; Window 2, on the one hand, displays the Mask result, accurately presents the detailed outline of the crack; on the other hand, it can switch to display the original image, which is convenient for users to compare and view the image differences before and after processing.

[0019] As a further technical solution of the present invention, in step three, the parameter setting module provides a setting function for the detection parameter threshold, and the user can flexibly adjust the confidence threshold and the intersection-over-union ratio threshold according to actual needs to optimize the accuracy and reliability of fracture segmentation; the detection result display module comprehensively and in detail displays the results of fracture detection, covering macro-statistical information such as the total number of detected targets, the proportion of total segmented area, and the detection time, as well as a target selection function, allowing the user to further view the detailed information of a specific target, including the detection target type, the proportion of a single target segmented area, confidence, target box position, and inclination information; the user operation module provides a variety of operation options including opening pictures, opening folders, opening videos, and opening cameras, supporting users to import logging imaging data to be detected from different data sources; and it also has a save function, which is convenient for users to save detection results and related data; an exit button is set to facilitate users to safely exit the system after use.

[0020] As a further technical solution of the present invention, in step 4, the crack image processing and curve fitting algorithm implementation includes: in the target selection area at the bottom of the client interface, the user can accurately select a specific segmentation target from multiple detection targets for separate analysis. The system presets 8 types of cracks, namely high-conductivity cracks, semi-open cracks, feather-shaped induced cracks, faults, dissolution cracks, local cracks, block-shaped induced cracks and high-resistance cracks; the segmentation area ratio is used to display the ratio of the selected single target to the total area of ​​the image, and the confidence level reflects the reliability of crack identification of a single target. The target frame position information is displayed in real time behind Xmin, Xmax, Ymin and Ymax. These data provide strong support for user analysis; The calculation of the crack inclination requires a series of complex image processing operations on the segmented Mask image: Preprocessing: delete the detection border, adjust the Mask color, remove the redundant information in the crack detection parameters, and facilitate the image binarization operation; Morphological processing: Open and close the binary image, and use morphological methods such as corrosion and expansion to further refine the crack morphology; Skeleton extraction: Use the skeleton extraction algorithm to extract the specific curve shape of the crack from the processed image, and preliminarily obtain the approximate curve shape of the crack; Denoising: Use corrosion operation to remove burr noise in the curve, making the curve shape smoother and clearer, and more accurately reflecting the true shape of the crack; Interpolation operation: After extracting the coordinates of the pixel points that reflect the crack characteristics, interpolation operation is performed on them to reasonably estimate additional points between known pixel points, optimize the continuity and smoothness of the curve, and make the transition of the crack curve more natural and smooth.

[0021] Curve fitting: The smoothed crack curve is fitted by constructing a complex polynomial. The powerful function expression ability of the complex polynomial is used to accurately describe the shape of the crack curve, and a smooth and flat crack curve shape is obtained, providing a reliable basis for subsequent in-depth analysis.

[0022] Beneficial effects achieved by the present invention: 1. In order to solve the problems of difficulty in identifying electrical imaging cracks and inability to extract crack parameters, a parameter calculation process is proposed. This method can accurately extract crack inclination parameter information.

[0023] 2. Through the front-end and back-end interactive design of basic information display area, image display window, parameter setting module, test result display module and user operation module, an integrated and visual operation platform is provided for users.

[0024] 3. Through the detection results of pictures, folders and videos, the logging imaging fractures can be efficiently identified and segmented, providing important technical support for geological exploration, oil reservoir development and other fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a flow chart of a method for well logging imaging fracture and cavity segmentation and parameter extraction based on deep learning.

[0026] Figure 2 The figure is a flow chart of the well logging electrical imaging data preprocessing process in an embodiment of the present invention.

[0027] Figure 3 This is a client interface diagram of a deep learning-based well logging imaging fracture segmentation system according to an embodiment of the present invention.

[0028] Figure 4 This is a flow chart of the process of segmenting fractures using well logging electrical imaging according to an embodiment of the present invention.

[0029] Figure 5 This is a theoretical diagram for calculating the well logging fracture dip parameters according to an embodiment of the present invention.

[0030] Figure 6 1 is a diagram showing the segmentation results of the well logging imaging fracture-cavity segmentation system according to an embodiment of the present invention (taking a high-conductivity fracture and a semi-open fracture as an example). DETAILED DESCRIPTION

[0031] The technical solution of the present invention is described in detail below in conjunction with the specific drawings.

[0032] See also Figure 1 This embodiment provides a method for well logging imaging fracture-hole segmentation and parameter extraction based on deep learning, and the method comprises the following steps: Step 1: Dataset preparation: First, the logging resistivity data is processed using Ciflog software to obtain a full borehole image, which is then cut to accurately fit the input requirements of the deep learning model. Finally, the imaged fractures are manually annotated according to different fracture types. See also Figure 2 ,The dataset preparation in step 1 includes: First, the logging resistivity data is processed using Ciflog software. In this process, the appropriate data source is selected, and the corresponding well and instrument are determined. The available instruments include FMI, ERMI, and STAR-II. After the selection is completed, the software will generate the original image. In order to enrich the data diversity, the original image is enhanced to generate static and dynamic images. Subsequently, the generated image is preprocessed to eliminate the common white strips in the logging image, and finally the full borehole image is obtained. In order to facilitate the input of full borehole images into the deep learning model for deep feature extraction and learning, two ratios (1:10 and 1:20) were used to produce the images, and the full borehole images were cut according to the 2m cutting standard. The cut images can more accurately adapt to the input requirements of the deep learning model; Finally, the imaged fractures are manually annotated according to different fracture types. After the annotation is completed, the corresponding JSON file is generated, and then the JSON file is converted into a TXT file for direct use by the deep learning model. Through the above series of operations, high-quality and standardized training data are provided for the deep learning model, which helps to improve the accuracy and efficiency of logging imaging fracture and cavity segmentation and parameter extraction.

[0033] Step 2: Model training and selection: Input the data set produced in step 1 into the deep learning model for training. During the training process, the model will generate two key outputs, namely last.pt and best.pt. Among them, last.pt represents the performance status of the deep learning model in the last iteration of training; and best.pt reflects the state of the model when it achieves the best segmentation effect in the entire iteration training cycle; best.pt is selected as the model in the actual segmentation process; such selection can ensure that the performance of the model is maximized in the well logging imaging fracture and cavity segmentation and parameter extraction tasks, so as to obtain more accurate and reliable segmentation results; Step 3: On the basis of completing the data set preparation, model training and selection, the client interface of the logging imaging fracture segmentation and parameter extraction system is further designed; the client interface design content includes: basic information display area, image display window, parameter setting module, detection result display module and user operation module; the client interface serves as a bridge for the user to interact with the back-end program; See also Figure 3 ,In step three, the basic information display area is used to clearly display ,the system name and the design team name, so that users can quickly understand ,the basic information and development entity of the system.

[0034] The image display window includes window 1 and window 2. Window 1 is used to intuitively present the crack segmentation results, synchronously display the detection frame and the corresponding label, so that the user can clearly see the position and category of each segmented target; Window 2, on the one hand, displays the Mask result and accurately presents the detailed outline of the crack; on the other hand, it can switch to display the original image, which is convenient for users to compare and view the image differences before and after processing.

[0035] The parameter setting module provides a detection parameter threshold setting function, and the user can flexibly adjust the confidence threshold and the intersection-over-union ratio threshold according to actual needs to optimize the accuracy and reliability of crack segmentation.

[0036] The detection result display module comprehensively and in detail displays the results of crack detection, including macro-statistical information such as the total number of detected targets, the proportion of total segmented area, and the detection time, as well as a target selection function, allowing users to further view detailed information of specific targets, including the detection target type, the proportion of the segmented area of ​​a single target, confidence, target box position, and inclination information.

[0037] The user operation module provides a variety of operation options including opening pictures, opening folders, opening videos and opening cameras, supporting users to import logging imaging data to be detected from different data sources; it also has a save function, which is convenient for users to save detection results and related data; and an exit button is set to facilitate users to safely exit the system after use.

[0038] Through the above-mentioned client interface design, this system provides users with an integrated and visual operation platform, which effectively improves the work efficiency and accuracy of logging imaging fracture segmentation and parameter extraction.

[0039] Step 4: Crack image processing and curve fitting algorithm implementation; In step 4, the crack image processing and curve fitting algorithm implementation include: in the target selection area at the bottom of the client interface, users can accurately select specific segmented targets from multiple detection targets for separate analysis. The system presets 8 types of cracks, namely high-conductivity cracks, semi-open cracks, feather-shaped induced cracks, faults, dissolution cracks, local cracks, block-shaped induced cracks and high-resistance cracks; the segmentation area ratio is used to display the ratio of the selected single target to the total area of ​​the image, and the confidence level reflects the reliability of crack identification of a single target. The target frame position information is displayed in real time behind Xmin, Xmax, Ymin and Ymax. These data provide strong support for user analysis; See also Figure 4 , the calculation of the crack inclination requires a series of complex image processing operations on the segmented Mask image: Preprocessing: delete the detection border, adjust the Mask color, remove the redundant information in the crack detection parameters, and facilitate the image binarization operation; Morphological processing: Open and close the binary image, and use morphological methods such as corrosion and expansion to further refine the crack morphology; Skeleton extraction: Use the skeleton extraction algorithm to extract the specific curve shape of the crack from the processed image, and preliminarily obtain the approximate curve shape of the crack; Denoising: Use corrosion operation to remove burr noise in the curve, making the curve shape smoother and clearer, and more accurately reflecting the true shape of the crack; Interpolation operation: After extracting the coordinates of the pixel points that reflect the crack characteristics, interpolation operation is performed on them to reasonably estimate additional points between known pixel points, optimize the continuity and smoothness of the curve, and make the transition of the crack curve more natural and smooth.

[0040] Curve fitting: The smoothed crack curve is fitted by constructing a complex polynomial. The powerful function expression ability of the complex polynomial is used to accurately describe the shape of the crack curve, and a smooth and flat crack curve shape is obtained, providing a reliable basis for subsequent in-depth analysis.

[0041] Crack dip parameter extraction: Extract the coordinates of the lowest and highest points of the fitting curve, connect the two points to construct a triangle, and obtain the crack dip information by calculating the angle between the two points. For specific theoretical references, Figure 5 (Calculation theory of fracture dip parameters in well logging).

[0042] Step 5. Open the client interface, select the data source to start the test, output the results, and display the results. The specific test results are as follows: Figure 6 .

[0043] It should be noted that, in this article, the term "comprises" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of more restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0044] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for well logging imaging fracture-cavity segmentation and parameter extraction based on deep learning, characterized in that: The method comprises the following steps: Step 1: Dataset preparation: First, the logging resistivity data is processed using Ciflog software to obtain a full borehole image, which is then cut to fit the input requirements of the deep learning model. Finally, the imaged fractures are manually annotated according to different fracture types. Step 2: Model training and selection: Input the data set produced in step 1 into the deep learning model for training. During the training process, the model will generate two key outputs, namely last.pt and best.pt. Among them, last.pt represents the performance status of the deep learning model in the last iteration of training; and best.pt reflects the state of the model when it achieves the best segmentation effect in the entire iteration training cycle; best.pt is selected as the model in the actual segmentation process; Step 3: On the basis of completing the data set preparation, model training and selection, the client interface of the logging imaging fracture segmentation and parameter extraction system is further designed; the client interface design content includes: basic information display area, image display window, parameter setting module, detection result display module and user operation module; the client interface serves as a bridge for the user to interact with the back-end program; Step 4: Process the crack image and implement the curve fitting algorithm. Extract the coordinates of the lowest and highest points of the fitting curve, connect the two points to construct a triangle, and obtain the inclination information of the crack by calculating the angle between the two points. Step 5. Open the client interface, select the data source to start detection, output the results, and display the results.

2. The method for well logging imaging fracture-cavity segmentation and parameter extraction based on deep learning according to claim 1, characterized in that: The dataset preparation in step 1 includes: First, the logging resistivity data is processed using Ciflog software. In this process, the data source is selected, and the corresponding well and instrument are determined. The available instruments include FMI, ERMI, and STAR-II. After the selection is completed, the software will generate the original image. In order to enrich the data diversity, the original image is enhanced to generate static images and dynamic images. Subsequently, the generated image is preprocessed to eliminate the common white strips in the logging image, and finally the full borehole image is obtained. In order to facilitate the input of the full borehole image into the deep learning model for deep feature extraction and learning, two scales were used to produce the image, and the full borehole image was cut according to the 2m cutting standard. The cut image more accurately adapts to the input requirements of the deep learning model; Finally, the imaged cracks are manually annotated according to different crack types. After the annotation is completed, the corresponding JSON file is generated and then converted into a TXT file.

3. The method for well logging imaging fracture-cavity segmentation and parameter extraction based on deep learning according to claim 2, characterized in that: In step three, the basic information display area is used to clearly display the system name and the design team name, so that users can quickly understand the basic information of the system and the development entity. The image display window includes window 1 and window 2. Window 1 is used to intuitively present the crack segmentation results, and synchronously display the detection box and the corresponding label, so that the user can clearly see the position and category of each segmented target; Window 2, on the one hand, displays the Mask result, accurately presents the detailed outline of the crack; on the other hand, it can switch to display the original image, which is convenient for users to compare and view the image differences before and after processing.

4. The method for well logging imaging fracture-cavity segmentation and parameter extraction based on deep learning according to claim 1, characterized in that: In step three, the parameter setting module provides a setting function for the detection parameter threshold. The user can flexibly adjust the confidence threshold and the intersection-over-union ratio threshold according to actual needs to optimize the accuracy and reliability of crack segmentation; the detection result display module displays various results of crack detection, covering macroscopic statistical information, and target selection function, allowing users to further view detailed information of specific targets, including detection target type, single target segmentation area ratio, confidence, target frame position and inclination information; The operation options provided by the user operation module include opening pictures, opening folders, opening videos and opening cameras, which supports users to import logging imaging data to be detected from different data sources; it also has a save function, which is convenient for users to save detection results and related data; and an exit button is set to facilitate users to safely exit the system after use.

5. The method for well logging imaging fracture-cavity segmentation and parameter extraction based on deep learning according to claim 1, characterized in that: In step 4, the crack image processing and curve fitting algorithm implementation include: in the target selection area at the bottom of the client interface, the user can accurately select a specific segmentation target from multiple detection targets for separate analysis. The system presets 8 types of cracks, namely high-conductivity cracks, semi-open cracks, feather-shaped induced cracks, faults, dissolution cracks, local cracks, block-shaped induced cracks and high-resistance cracks; the segmentation area ratio is used to display the ratio of the selected single target to the total area of ​​the image, and the confidence level reflects the reliability of crack identification of a single target. The target frame position information is displayed in real time behind Xmin, Xmax, Ymin and Ymax; The calculation of the crack inclination requires a series of image processing operations on the segmented Mask image: Preprocessing: delete the detection border, adjust the Mask color, remove the redundant information in the crack detection parameters, and facilitate the image binarization operation; Morphological processing: Open and close the binary image, and use morphological methods such as corrosion and expansion to further refine the crack morphology; Skeleton extraction: Use the skeleton extraction algorithm to extract the specific curve shape of the crack from the processed image, and preliminarily obtain the approximate curve shape of the crack; Denoising: Use corrosion operation to remove burr noise in the curve, making the curve shape smoother and clearer, and more accurately reflecting the true shape of the crack; Interpolation: After extracting the coordinates of the pixels that reflect the crack characteristics, interpolation is performed on them to reasonably estimate additional points between the known pixels to optimize the continuity and smoothness of the curve; Curve fitting: The smoothed crack curve is fitted by constructing a complex polynomial. The powerful function expression ability of the complex polynomial is used to accurately describe the shape of the crack curve to obtain a smooth and flat crack curve shape.

Images