Logging imaging sine seam identification method, device and equipment based on PolylaneNet and noise perception and medium

Abstract

The invention discloses a logging imaging sine seam identification method, device and equipment based on PolylaneNet and noise perception, and a medium, and the method comprises the steps: obtaining logging resistivity structured data, and carrying out the preprocessing, and obtaining a to-be-input data set; and the initial PolylaneNet model is improved, the improved PolylaneNet model is trained based on the data set to be input, and the trained PolylaneNet model is used for logging imaging sine seam identification. The invention belongs to the field of logging imaging sine seam identification. According to the method, the accuracy of logging imaging sine seam identification can be improved.

Description

A method, apparatus, equipment, and medium for identifying sinusoidal fractures in well logging imaging based on PolylaneNet and noise perception. Technical Field

[0001] This invention relates to the field of sinusoidal seam recognition in well logging imaging, and more particularly to a method, apparatus, equipment, and medium for sinusoidal seam recognition in well logging imaging based on PolylaneNet and noise perception. Background Technology

[0002] With the continuous deepening of oil and gas exploration and development, well logging imaging technology plays an important role in formation structure analysis, fracture identification, and reservoir evaluation. Well logging resistivity imaging data can reflect the formation structure characteristics of the wellbore. Among them, geological structures such as fractures, bedding, and faults are usually represented as sinusoidal curves in the wellbore unfolded images. Therefore, accurate identification and parameter fitting of sinusoidal fractures are of great significance for the interpretation of geological structures.

[0003] Traditional sinusoidal seam identification methods often rely on manual interpretation or image processing methods such as Hough transform and curve fitting. However, in actual well logging data, due to factors such as noise interference, image quality differences, and the complex morphology of sinusoidal seams, these methods often struggle to guarantee identification accuracy and stability. Furthermore, manual interpretation is inefficient and highly subjective. Therefore, to address these issues, this invention provides a well logging imaging sinusoidal seam identification method based on PolylaneNet and noise perception. Summary of the Invention

[0004] This invention provides a method, apparatus, device, and medium for identifying sinusoidal seams in well logging imaging based on PolylaneNet and noise perception, which solves the technical problem of low accuracy in identifying sinusoidal seams in well logging imaging in the prior art and achieves the technical effect of improving the accuracy of identifying sinusoidal seams in well logging imaging.

[0005] In a first aspect, the present invention provides a method for identifying sinusoidal fractures in well logging imaging based on PolylaneNet and noise perception, comprising:

[0006] S11, acquire structured resistivity data from well logging and preprocess it to obtain the input dataset. Preprocessing includes image data conversion, data annotation, and dataset partitioning. S12, improve the initial PolylaneNet model and train it on the input dataset. Then, use the trained PolylaneNet model for well logging imaging sinusoidal fracture recognition, including S121-S124: S121, replace the third-order polynomial fitting function in the initial PolylaneNet model with a sinusoidal fitting function; S122, acquire... Take the image quality factor and construct the adaptive Huber loss function based on the dynamic threshold calculated by the adaptive error threshold function; S123, reconstruct the adaptive Huber loss function to obtain the improved PolylaneNet model, where the reconstruction terms include confidence, boundary constraints, sine curve similarity, amplitude, phase, baseline, point set fitting, and amplitude sign constraint; S124, train the improved PolylaneNet model based on the input dataset, and when the preset training requirements are met, use the improved PolylaneNet model for well logging imaging sine seam recognition.

[0007] Furthermore, the third-order polynomial fitting function in the initial PolylaneNet model is replaced with a sine fitting function, including: replacing the third-order polynomial fitting function in the initial PolylaneNet model with a sine fitting function, including:

[0008] in, Let be the vertical coordinate of the sinusoidal seam in the image. Let be the horizontal coordinate of the sinusoidal seam in the image. For amplitude, Angular frequency, For the initial phase, For vertical displacement; based on the period of the sinusoidal vibration, the sinusoidal fitting function is updated, including: .

[0009] Furthermore, the image quality factor is obtained, and an adaptive Huber loss function is constructed based on the dynamic threshold calculated by the adaptive error threshold function. This includes: calculating the image quality factor based on the results of Sobel gradient calculation and mean calculation, including:

[0010]

[0011] in, coordinates Well logging imaging at the location grayscale value, For horizontal edge operator kernels, For the operator kernel of the vertical edge, coordinates The gradient value at the pixel level at that location. coordinates The gradient value in the vertical direction at that point. coordinates Total gradient magnitude at pixel location. The width of the image in pixels. For the image's pixel height, This is the convolution operator;

[0012]

[0013] in, For the first A clear quantification of each image sample. For the first The average gradient magnitude of each image sample The minimum gradient magnitude in the image sample. The maximum gradient magnitude in the image sample is used; an adaptive error threshold function is constructed, and the dynamic threshold is calculated, including:

[0014] in, For dynamic thresholds, Preset minimum threshold A preset maximum threshold is used; based on the dynamic threshold, an adaptive Huber loss function is constructed, including:

[0015] in, For the adaptive Huber loss function, These are the actual labeled values. For predicted labeled values.

[0016] Furthermore, regarding confidence level, boundary constraints, and sine curve similarity, including: confidence loss function.

[0017] in, This is the confidence loss value. To determine the effective number of sinusoidal seam samples, As a class balance factor, For modulation factor, For image samples The true label, This indicates the existence of a sinusoidal gap. This indicates there is no sine wave. For image samples The prediction confidence level For predicting probabilities Power; boundary constraints, including left and right boundary constraints and top and bottom boundary constraints; wherein, the loss function for the left and right boundary ordinates includes:

[0018] in, The value represents the consistency loss of the left and right boundary ordinates. For the adaptive Huber loss function, For image samples left boundary Predicted values ​​of coordinates; For image samples right boundary Predicted values ​​of coordinates The number of image samples; the loss function for the upper and lower boundary ordinates, including:

[0019] in, The upper and lower boundaries are associated with the amplitude loss values. For image samples lower boundary Predicted values ​​of coordinates For image samples upper boundary Predicted values ​​of coordinates For image samples The predicted amplitude value, The correlation coefficient between boundary distance and amplitude; the full-cycle sine curve similarity loss function includes:

[0020] in, This represents the similarity loss value of the sine curve. Image samples predicted using sine parameters The horizontal index is of coordinate values, and , For image samples The horizontal index is Authentic labeling coordinates, For image samples The initial phase, For image samples The vertical displacement, For image samples The amplitude.

[0021] Furthermore, amplitude, phase, baseline, point set fitting, and amplitude sign constraints, including: amplitude loss function, including:

[0022] in, For amplitude prediction loss value, For image samples The true amplitude value; the phase loss function, including:

[0023] in, This is the phase prediction loss value. For image samples The predicted phase value, For image samples The true phase value; the baseline loss function, including:

[0024] in, The baseline predicted loss value, For image samples The predicted baseline value, For image samples The true baseline value; the loss function for the curve crossing the set of labeled points, including:

[0025] in, The loss value is the fitting value for the point set. The number of marking points for each sinusoidal seam. For image samples The one marker coordinate, To predict the curve in place coordinate, To be accurately labeled place Coordinates; amplitude sign consistency loss function, including:

[0026] in, This represents the amplitude sign consistency loss value. The loss that occurs when the amplitude is negative.

[0027] Furthermore, the adaptive Huber loss function is reconstructed to obtain the improved PolylaneNet model, including:

[0028] in, , , , , , , , , , as well as All are preset weights. For the left boundary loss, For the right boundary loss, For the upper boundary loss, The sum of the lower boundary loss, the right boundary loss, and the left boundary loss is the consistency loss value of the ordinates of the left and right boundaries. The sum of the lower boundary loss and the upper boundary loss is the correlation loss value between the upper and lower boundaries and the amplitude. This represents the total loss.

[0029] Furthermore, the structured resistivity data from well logging is acquired and preprocessed to obtain the input dataset, including: acquiring the structured resistivity data from well logging and converting it into image data; labeling the location of the sinusoidal fractures and the sinusoidal fitting parameters in the image data, including amplitude, phase, and vertical displacement; dividing the labeled image data into training, validation, and test sets, and adapting it to the PolylaneNet model input format.

[0030] Secondly, the present invention provides a well logging imaging sinusoidal fracture recognition device based on PolylaneNet and noise perception, comprising: an information acquisition module, used to execute step S11, including: acquiring well logging resistivity structured data and preprocessing it to obtain an input dataset, wherein the preprocessing includes image data conversion, data annotation, and dataset partitioning; and a model improvement module, used to execute step S12, including: improving an initial PolylaneNet model, training the improved PolylaneNet model based on the input dataset, and using the trained PolylaneNet model for well logging imaging sinusoidal fracture recognition, including S121-S124: S121, the initial... In the PolylaneNet model, the third-order polynomial fitting function is replaced with a sine fitting function; in S122, the image quality factor is obtained, and the dynamic threshold calculated based on the adaptive error threshold function is used to construct the adaptive Huber loss function; in S123, the adaptive Huber loss function is reconstructed to obtain the improved PolylaneNet model, where the reconstruction terms include confidence, boundary constraints, sine curve similarity, amplitude, phase, baseline, point set fitting, and amplitude sign constraint; in S124, the improved PolylaneNet model is trained based on the input dataset, and when the preset training requirements are met, the improved PolylaneNet model is used for well logging imaging sine seam recognition.

[0031] Thirdly, the present invention provides an electronic device comprising: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to execute to implement a well logging imaging sinusoidal seam identification method based on PolylaneNet and noise perception as provided in the first aspect.

[0032] Fourthly, the present invention provides a non-transitory computer-readable storage medium that, when the instructions in the storage medium are executed by the processor of an electronic device, enables the electronic device to perform a well logging imaging sinusoidal seam identification method based on PolylaneNet and noise perception as provided in the first aspect.

[0033] One or more technical solutions provided in this invention have at least the following technical effects or advantages: This invention proposes a sinusoidal fracture identification method for well logging imaging based on PolylaneNet and noise perception mechanism. This invention replaces the third-order polynomial fitting function in the traditional PolylaneNet model with a sinusoidal fitting function that better matches the morphological characteristics of geological fractures in well logging imaging, so that the model can more accurately describe the periodically distributed sinusoidal fracture structure in well wall imaging, thereby significantly improving the identification accuracy.

[0034] This invention introduces an image quality factor and combines it with an adaptive error threshold function to construct an adaptive Huber loss function, enabling the model to dynamically adjust its error tolerance based on image quality during training, thereby improving its robustness to noise interference and low-quality imaging data.

[0035] This invention reconstructs the loss function and introduces multiple constraint mechanisms, including confidence constraints, boundary constraints, sine curve similarity constraints, and constraints related to amplitude, phase, baseline, point set fitting, and amplitude sign. This allows the model to simultaneously consider geometric features and physical meaning during the learning process, improving the accuracy and stability of sinusoidal seam parameter fitting. This invention not only effectively enhances the automatic identification capability of sinusoidal seams in well logging imaging but also maintains high identification accuracy and stability in complex noise environments, demonstrating significant engineering application value. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 is a flowchart illustrating a well logging imaging sinusoidal fracture identification method based on PolylaneNet and noise perception provided by the present invention; Figure 2 is a flowchart illustrating an improved PolylaneNet model provided by the present invention; Figure 3 is a flowchart illustrating an improved adaptive function provided by the present invention; Figure 4 is a flowchart illustrating the total loss function provided by the present invention; Figure 5 is a flowchart illustrating the identification effect provided by the present invention. Detailed Implementation

[0038] This invention provides a method for identifying sinusoidal seams in well logging imaging based on PolylaneNet and noise perception, which solves the technical problem of low accuracy in identifying sinusoidal seams in existing technologies.

[0039] The technical solution of this invention is to solve the above-mentioned technical problems. The overall idea is as follows: A method for identifying sinusoidal seams in well logging imaging based on PolylaneNet and noise perception, comprising: S11, acquiring structured resistivity data from well logging and preprocessing it to obtain the input dataset, wherein the preprocessing includes image data conversion, data annotation, and dataset partitioning; S12, improving the initial PolylaneNet model, training the improved PolylaneNet model based on the input dataset, and using the trained PolylaneNet model for identifying sinusoidal seams in well logging imaging, including S121-S124: S121, the initial PolylaneNet model is improved, and the improved PolylaneNet model is trained based on the input dataset, and the trained PolylaneNet model is used for identifying sinusoidal seams in well logging imaging. In S122, the third-order polynomial fitting function in the et model is replaced with a sine fitting function; the image quality factor is obtained, and the dynamic threshold calculated based on the adaptive error threshold function is used to construct the adaptive Huber loss function; in S123, the adaptive Huber loss function is reconstructed to obtain the improved PolylaneNet model, where the reconstruction terms include confidence, boundary constraints, sine curve similarity, amplitude, phase, baseline, point set fitting, and amplitude sign constraint; in S124, the improved PolylaneNet model is trained based on the input dataset, and when the preset training requirements are met, the improved PolylaneNet model is used for well logging imaging sine seam recognition.

[0040] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0041] First, it should be clarified that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0042] This invention provides a method for identifying sinusoidal seams in well logging imaging based on PolylaneNet and noise perception, as shown in Figure 1. The method includes: S11, acquiring structured data of well logging resistivity and performing preprocessing to obtain the input dataset, wherein the preprocessing includes image data conversion, data annotation and dataset partitioning.

[0043] The process involves acquiring and preprocessing structured logging resistivity data to obtain the input dataset. This includes: acquiring structured logging resistivity data and converting it into image data; labeling the location of sinusoidal fractures and sinusoidal fitting parameters in the image data, including amplitude, phase, and vertical displacement; dividing the labeled image data into training, validation, and test sets, and adapting it to the PolylaneNet model input format.

[0044] Step S11 is primarily used to construct the input dataset required for model training.

[0045] First, structured logging resistivity data is acquired, typically stored in tabular or matrix form as wellbore resistivity measurements. Then, a well logging image conversion module unfolds the structured tabular data according to depth and azimuth, converting it into two-dimensional logging image data, making it compatible with deep learning models based on visual feature extraction.

[0046] After obtaining the image data, the sinusoidal seam target in the image is labeled. The labeling process combines manual labeling with semi-automated auxiliary tools to mark the spatial position of the sinusoidal seam in the image. At the same time, the corresponding sinusoidal fitting parameters, including amplitude, initial phase and vertical displacement, are extracted and labeled to characterize the geometric features of the sinusoidal seam.

[0047] After annotation, the annotated well logging image data is randomly divided into training, validation, and test sets according to a certain ratio, and the data format is standardized to meet the input format requirements of the PolylaneNet model.

[0048] S12 involves improving the initial PolylaneNet model and training the improved PolylaneNet model based on the input dataset. The trained PolylaneNet model is then used for well logging imaging sinusoidal fracture identification, including S121-S124. Additionally, refer to Figure 2: S121, where the third-order polynomial fitting function in the initial PolylaneNet model is replaced with a sinusoidal fitting function.

[0049] The third-order polynomial fitting function in the initial PolylaneNet model is replaced with a sine fitting function, including: where the third-order polynomial fitting function in the initial PolylaneNet model is... ,in: Here are the vertical coordinates of the lane lines in the image. Here are the horizontal coordinates of the lane lines in the image. , , , The coefficients of the third-order polynomial, i.e. the coefficients that the model needs to predict (the meaning of XY here is different from that of XY in other parts of the text, while XY in other parts of the text has the same meaning).

[0050] Polynomial fitting functions cannot capture the sinusoidal waveform characteristics of sinusoidal seams (including sinusoidal seams, high-resistivity seams, bedding seams, and feather-induced seams). Therefore, the third-order polynomial fitting function in the initial PolylaneNet model is replaced with a sinusoidal fitting function, including:

[0051] in, Let be the vertical coordinate of the sinusoidal seam in the image. Let be the horizontal coordinate of the sinusoidal seam in the image. For amplitude, Angular frequency, For the initial phase, The vertical displacement is; the period of the sinusoidal oscillation is... Since the period of each sinusoidal slit in well logging imaging is fixed at 1, that is... Therefore, based on the period of the oscillation of the sine curve, the sine fitting function is updated, including: .

[0052] The modified model will no longer predict polynomial coefficients, but will instead directly regress the amplitude, phase, and vertical displacement of a sine function over one period.

[0053] S122: Obtain the image quality factor and construct the adaptive Huber loss function based on the dynamic threshold calculated by the adaptive error threshold function.

[0054] For scenarios where there is strong noise interference in well logging imaging data, the MSE (mean square error) loss function used in the original PolylaneNet model is very sensitive to outliers, which can easily cause the model to be misled by noise points during training, thus affecting the fitting accuracy of sinusoidal seam parameters.

[0055] This invention will use the Huber loss function to solve this problem. Although the traditional Huber loss is robust to outliers, the error threshold... This is a fixed value. This results in: for samples with clear images and high signal-to-noise ratios, an excessively large value... It retains too much of the quadratic loss interval, resulting in insufficient fitting accuracy for minute details; while for noisy samples, the fitting accuracy is too small. It will misjudge normal noise as a large error, causing it to enter the linear loss range, thus failing to effectively suppress noise interference.

[0056] Therefore, as shown in Figure 3, this invention proposes a noise-aware adaptive robust loss function that dynamically adjusts the threshold of Huber loss by calculating the image sharpness or noise level. Therefore, to suppress noise interference in model training and improve the model's recognition stability in low signal-to-noise ratio environments, the following steps are taken: The image quality factor is obtained, and an adaptive Huber loss function is constructed based on the dynamic threshold calculated using the adaptive error threshold function. This includes: calculating the image quality factor based on the results of Sobel gradient calculation and mean calculation, including:

[0057]

[0058] in, coordinates Well logging imaging at the location grayscale value, For horizontal edge operator kernels, For the operator kernel of the vertical edge, coordinates The gradient value at the pixel level at that location. coordinates The gradient value in the vertical direction at that point. coordinates Total gradient magnitude at pixel location. The width of the image in pixels. For the image's pixel height, This is the convolution operator;

[0059]

[0060] in, For the first A clear quantification of each image sample. For the first The average gradient magnitude of each image sample The minimum gradient magnitude in the image sample. The maximum gradient magnitude in the image sample; the image quality factor aims to quantify image sharpness and noise level, and this invention uses... Represents image samples Clear quantifications are obtained, and gradient magnitude based on the Sobel operator is used as the basic metric.

[0061] Construct an adaptive error threshold function and calculate the dynamic threshold, including:

[0062] in, For dynamic thresholds, Preset minimum threshold A preset maximum threshold is used; based on the dynamic threshold, an adaptive Huber loss function is constructed, including:

[0063] in, For the adaptive Huber loss function, These are the actual labeled values. For predicted labeled values.

[0064] S123, reconstruct the adaptive Huber loss function to obtain the improved PolylaneNet model. The reconstruction terms include confidence, boundary constraints, sine curve similarity, amplitude, phase, baseline, point set fitting, and amplitude sign constraint. See Figure 4 for further details.

[0065] Regarding confidence level, boundary constraints, and sine curve similarity, including: confidence loss function

[0066] in, This is the confidence loss value. To determine the effective number of sinusoidal seam samples, As a class balance factor, For modulation factor, For image samples The true label, This indicates the existence of a sinusoidal gap. This indicates there is no sine wave. For image samples The prediction confidence level For predicting probabilities To address the severe class imbalance between sinusoidal seams and the background in well logging images, this invention changes the confidence loss function type of the original model from the traditional binary cross-entropy loss (BCE Loss) to Focal Loss. Focal Loss reduces the weight of easily classified samples through a modulation factor, making the model pay more attention to the difficult-to-classify sinusoidal seam samples.

[0067] At the same time, introduce The parameter balance of positive and negative sample weights effectively addresses the class imbalance problem caused by the small proportion of sinusoidal fractures in well logging images, thus balancing the weight distribution of positive and negative samples; secondly, through... The parameters control the weights of difficult samples, enabling the model to automatically adjust its focus on difficult-to-identify sinusoidal seam samples, thus improving learning efficiency.

[0068] Boundary constraints include left and right boundary constraints and top and bottom boundary constraints; among which, the loss function for the left and right boundary ordinates includes:

[0069] in, The value represents the consistency loss of the left and right boundary ordinates. For the adaptive Huber loss function, For image samples left boundary Predicted values ​​of coordinates; For image samples right boundary Predicted values ​​of coordinates The number of image samples; left and right boundaries. The core function of coordinate loss is to constrain the left and right boundaries of the sinusoidal seam. Consistent coordinates align with the physical characteristics of a horizontally continuous sinusoidal seam. By setting equal loss weights, boundary misalignment is avoided, aiming to improve the consistency and rationality of boundary positioning. For the model, this reduces boundary positioning errors, avoids boundary distortion, simplifies model learning degrees of freedom, and makes the predicted shape of the sinusoidal seam more realistic.

[0070] The loss function for the upper and lower boundary ordinates includes:

[0071] in, The upper and lower boundaries are associated with the amplitude loss values. For image samples lower boundary Predicted values ​​of coordinates For image samples upper boundary Predicted values ​​of coordinates For image samples The predicted amplitude value, The correlation coefficient between boundary distance and amplitude; upper and lower boundaries The ordinate loss function correlates the distance between the upper and lower boundaries with the amplitude, incorporating the physical law of "boundary distance = amplitude × correlation coefficient", and dynamically adjusts the loss weight to achieve adaptive learning. The purpose is to strengthen the matching degree between the boundary and the amplitude, ensure the physical rationality of the prediction, indirectly improve the amplitude prediction accuracy, optimize the consistency of the upper and lower boundary positioning, and reduce the ambiguity of the model prediction.

[0072] The full-cycle sine curve similarity loss function includes:

[0073] in, This represents the similarity loss value of the sine curve. Image samples predicted using sine parameters The horizontal index is of coordinate values, and , For image samples The horizontal index is Authentic labeling coordinates, For image samples The initial phase, For image samples The vertical displacement, For image samples The amplitude.

[0074] Amplitude loss focuses on the resistivity difference of amplitude to characterize its significance. It directly optimizes amplitude prediction through adaptive Huber loss and dynamically adjusts the weights based on amplitude magnitude and boundary clarity. The aim is to improve amplitude prediction accuracy, enhance the extraction of core physical features of sinusoidal seams, help the model better capture key features of sinusoidal seams, indirectly optimize boundary localization, and improve recognition reliability.

[0075] Amplitude, phase, baseline, point set fitting, and amplitude sign constraints, including: amplitude loss function, including:

[0076] in, For amplitude prediction loss value, For image samples The true amplitude value; the amplitude loss focuses on the resistivity difference of the amplitude to characterize its significance. The amplitude prediction is directly optimized by adaptive Huber loss. The weights are dynamically adjusted by combining the amplitude size and boundary clarity. The aim is to improve the accuracy of amplitude prediction and enhance the extraction of the core physical features of the sinusoidal slit. This can help the model better capture the key features of the sinusoidal slit, indirectly optimize the boundary localization effect, and improve the reliability of recognition.

[0077] Phase loss functions include:

[0078] in, This is the phase prediction loss value. For image samples The predicted phase value, For image samples The true phase value; the phase loss is the core function of the phase in reflecting the horizontal offset of the sinusoidal slit. The phase prediction is directly optimized through adaptive Huber loss, while guiding the phase continuity of adjacent sinusoidal slits. The purpose is to improve the phase prediction accuracy and ensure the consistency of sinusoidal slit positioning. It can directly improve the horizontal positioning effect of the sinusoidal slit, help the boundary to be accurately aligned, and enhance the coherence of the recognition results.

[0079] The baseline loss function includes:

[0080] in, The baseline predicted loss value, For image samples The predicted baseline value, For image samples The true baseline value; the baseline loss focuses on the role of the baseline in representing the average position of the sinusoidal joint. The baseline prediction is directly optimized through adaptive Huber loss, which guides the baseline to change smoothly with depth. The purpose is to improve the accuracy of baseline prediction, ensure the physical rationality of the overall shape fitting of the sinusoidal joint, improve the overall shape fitting effect of the sinusoidal joint, conform to geological laws, and improve the stability of identification.

[0081] The loss function for the curve passing through the set of labeled points includes:

[0082] in, The loss value is the fitting value for the point set. The number of marking points for each sinusoidal seam. For image samples The one marker coordinate, To predict the curve in place coordinate, To be accurately labeled place The loss function for curves passing through the set of labeled points aims to improve curve fitting accuracy and boundary positioning accuracy by forcing the predicted curve to fit the labeled points, dynamically adjusting the weights according to the density of labeled points, and strengthening the importance of boundary points. It makes full use of labeled data. For the model, it can accurately capture the local morphological details of the sinusoidal seam, improve the accuracy of boundary positioning and data utilization efficiency.

[0083] The amplitude sign consistency loss function includes:

[0084] in, This represents the amplitude sign consistency loss value. The loss that occurs when the amplitude is negative.

[0085] The amplitude sign consistency loss function introduces physical constraints and uses the ReLU function to force the amplitude to be positive. The purpose is to avoid prediction reversal caused by amplitude sign errors, ensure the physical rationality of the prediction results, and is easy to calculate. It can improve the reliability of sinusoidal slit prediction, reduce subsequent post-processing steps, and avoid the generation of non-physical prediction results.

[0086] The adaptive Huber loss function is reconstructed to obtain the improved PolylaneNet model, including:

[0087] in, , , , , , , , , , as well as All are preset weights. For the left boundary loss, For the right boundary loss, For the upper boundary loss, The sum of the lower boundary loss, the right boundary loss, and the left boundary loss is the consistency loss value of the ordinates of the left and right boundaries. The sum of the lower boundary loss and the upper boundary loss is the correlation loss value between the upper and lower boundaries and the amplitude. This represents the total loss.

[0088] The total loss function integrates physical laws, geometric constraints, and data features, and adopts dynamic weight adjustment and hierarchical optimization strategies to achieve multi-task collaborative optimization of confidence, boundary location, and parameter prediction. The aim is to build a comprehensive optimization framework that takes into account both physical rationality and model performance, thereby comprehensively improving the robustness of the model and the accuracy of sinusoidal fracture location and identification, accelerating model convergence, and providing strong support for well logging sinusoidal fracture identification.

[0089] S124. The improved PolylaneNet model is trained based on the input dataset. When the preset training requirements are met, the improved PolylaneNet model is used for well logging imaging sinusoidal seam recognition.

[0090] An improved PolylaneNet was trained using a labeled dataset. With appropriate learning rate and number of iterations, the designed noise-robust loss function was used as the optimization objective. The model parameters were updated through backpropagation, and the hyperparameters were optimized on the validation set to avoid overfitting.

[0091] Model inference: Input preprocessed image data, extract features through backbone network, output parameters through sine fitting module, apply constraints to complete sine crack identification and localization, and output specific information of effective cracks, as shown in Figure 5.

[0092] In summary, this invention proposes a sinusoidal fracture identification method for well logging imaging based on PolylaneNet and a noise perception mechanism. This invention replaces the third-order polynomial fitting function in the traditional PolylaneNet model with a sinusoidal fitting function that better matches the morphological characteristics of geological fractures in well logging imaging, enabling the model to more accurately describe the periodically distributed sinusoidal fracture structure in well wall imaging, thereby significantly improving the identification accuracy.

[0093] This invention introduces an image quality factor and combines it with an adaptive error threshold function to construct an adaptive Huber loss function, enabling the model to dynamically adjust its error tolerance based on image quality during training, thereby improving its robustness to noise interference and low-quality imaging data.

[0094] This invention reconstructs the loss function and introduces multiple constraint mechanisms, including confidence constraints, boundary constraints, sine curve similarity constraints, and constraints related to amplitude, phase, baseline, point set fitting, and amplitude sign. This allows the model to simultaneously consider geometric features and physical meaning during the learning process, improving the accuracy and stability of sinusoidal seam parameter fitting. This invention not only effectively enhances the automatic identification capability of sinusoidal seams in well logging imaging but also maintains high identification accuracy and stability in complex noise environments, demonstrating significant engineering application value.

[0095] Based on the same inventive concept, this invention provides a well logging imaging sinusoidal fracture recognition device based on PolylaneNet and noise perception, comprising: an information acquisition module, used to execute step S11, including: acquiring well logging resistivity structured data and performing preprocessing to obtain an input dataset, wherein the preprocessing includes image data conversion, data annotation, and dataset partitioning; and a model improvement module, used to execute step S12, including: improving the initial PolylaneNet model, training the improved PolylaneNet model based on the input dataset, and using the trained PolylaneNet model for well logging imaging sinusoidal fracture recognition, including S121-S124: S121, ... In the initial PolylaneNet model, the third-order polynomial fitting function is replaced with a sine fitting function; in S122, the image quality factor is obtained, and an adaptive Huber loss function is constructed based on the dynamic threshold calculated by the adaptive error threshold function; in S123, the adaptive Huber loss function is reconstructed to obtain the improved PolylaneNet model, where the reconstruction terms include confidence, boundary constraints, sine curve similarity, amplitude, phase, baseline, point set fitting, and amplitude sign constraint; in S124, the improved PolylaneNet model is trained based on the input dataset, and when the preset training requirements are met, the improved PolylaneNet model is used for well logging imaging sine seam recognition.

[0096] Based on the same inventive concept, the present invention also provides an electronic device, comprising: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to execute to implement a well logging imaging sinusoidal slit identification method based on PolylaneNet and noise perception as described above.

[0097] Based on the same inventive concept, the present invention also provides a non-transitory computer-readable storage medium, which, when the instructions in the storage medium are executed by the processor of an electronic device, enables the electronic device to execute the aforementioned method for identifying sinusoidal seams in well logging imaging based on PolylaneNet and noise perception.

[0098] Since the electronic device described in this embodiment is an electronic device used to implement the information processing method in the embodiments of the present invention, those skilled in the art can understand the specific implementation methods and various variations of the electronic device in this embodiment based on the information processing method described in the embodiments of the present invention. Therefore, how the electronic device implements the method in the embodiments of the present invention will not be described in detail here. Any electronic device used by those skilled in the art to implement the information processing method in the embodiments of the present invention falls within the scope of protection of the present invention.

[0099] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0100] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in one or more blocks of the flowchart illustrations and / or one or more blocks of the block diagrams.

[0101] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means that implement the functions specified in one or more flowcharts and / or one or more block diagrams.

[0102] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, such that the instructions, which execute on the computer or other programmable apparatus, provide steps for implementing the functions specified in one or more flowcharts and / or one or more block diagrams.

[0103] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0104] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for identifying sinusoidal seams in well logging imaging based on PolylaneNet and noise perception, characterized in that, include: S11, acquire structured resistivity data from well logging and preprocess it to obtain the input dataset. Preprocessing includes image data conversion, data annotation, and dataset partitioning. S12, improve the initial PolylaneNet model and train it on the input dataset. Then, use the trained PolylaneNet model for well logging imaging sinusoidal fracture recognition, including S121-S124: S121, replace the third-order polynomial fitting function in the initial PolylaneNet model with a sinusoidal fitting function; S122, acquire the image... Like the quality factor, and based on the dynamic threshold calculated by the adaptive error threshold function, an adaptive Huber loss function is constructed; S123, the adaptive Huber loss function is reconstructed to obtain the improved PolylaneNet model, wherein the reconstruction terms include confidence, boundary constraints, sine curve similarity, amplitude, phase, baseline, point set fitting, and amplitude sign constraint; S124, the improved PolylaneNet model is trained based on the dataset to be input, and when the preset training requirements are met, the improved PolylaneNet model is used for well logging imaging sine fracture recognition.

2. The well logging imaging sinusoidal fracture identification method based on PolylaneNet and noise perception as described in claim 1, characterized in that, The third-order polynomial fitting function in the initial PolylaneNet model is replaced with a sine fitting function, including: in, Let be the vertical coordinate of the sinusoidal seam in the image. Let be the horizontal coordinate of the sinusoidal seam in the image. For amplitude, Angular frequency, For the initial phase, For vertical displacement; based on the period of the sinusoidal vibration, the sinusoidal fitting function is updated, including: 。 3. The well logging imaging sinusoidal fracture identification method based on PolylaneNet and noise perception as described in claim 1, characterized in that, Obtain the image quality factor and construct an adaptive Huber loss function based on the dynamic threshold calculated by the adaptive error threshold function. This includes calculating the image quality factor based on the results of Sobel gradient calculation and mean calculation. in, coordinates Well logging imaging at the location grayscale value, For horizontal edge operator kernels, For the operator kernel of the vertical edge, coordinates The gradient value at the pixel level at that location. coordinates The gradient value in the vertical direction at that point. coordinates Total gradient magnitude at pixel location. The width of the image in pixels. For the image's pixel height, This is the convolution operator; in, For the first A clear quantification of each image sample. For the first The average gradient magnitude of each image sample The minimum gradient magnitude in the image sample. The maximum gradient magnitude in the image sample is used; an adaptive error threshold function is constructed, and the dynamic threshold is calculated, including: in, For dynamic thresholds, Preset minimum threshold A preset maximum threshold is used; based on the dynamic threshold, an adaptive Huber loss function is constructed, including: in, For the adaptive Huber loss function, These are the actual labeled values. For predicted labeled values.

4. The well logging imaging sinusoidal fracture identification method based on PolylaneNet and noise perception as described in claim 1, characterized in that, Regarding confidence level, boundary constraints, and sine curve similarity, including: confidence loss function in, This is the confidence loss value. To determine the effective number of sinusoidal seam samples, As a class balance factor, For modulation factor, For image samples The true label, This indicates the existence of a sinusoidal gap. This indicates there is no sine wave. For image samples The prediction confidence level For predicting probabilities Power; boundary constraints, including left and right boundary constraints and top and bottom boundary constraints; wherein, the loss function for the left and right boundary ordinates includes: in, The value represents the consistency loss of the left and right boundary ordinates. For the adaptive Huber loss function, For image samples left boundary Predicted values ​​of coordinates; For image samples right boundary Predicted values ​​of coordinates The number of image samples; the loss function for the upper and lower boundary ordinates, including: in, The upper and lower boundaries are associated with the amplitude loss values. For image samples lower boundary Predicted values ​​of coordinates For image samples upper boundary Predicted values ​​of coordinates For image samples The predicted amplitude value, The correlation coefficient between boundary distance and amplitude; the full-cycle sine curve similarity loss function includes: in, This represents the similarity loss value of the sine curve. Image samples predicted using sine parameters The horizontal index is of coordinate values, and ， For image samples The horizontal index is Authentic labeling coordinates, For image samples The initial phase, For image samples The vertical displacement, For image samples The amplitude.

5. The well logging imaging sinusoidal fracture identification method based on PolylaneNet and noise perception as described in claim 4, characterized in that, Amplitude, phase, baseline, point set fitting, and amplitude sign constraints, including: amplitude loss function, including: in, For amplitude prediction loss value, For image samples The true amplitude value; the phase loss function, including: in, This represents the phase prediction loss value. For image samples The predicted phase value, For image samples The true phase value; the baseline loss function, including: in, The baseline predicted loss value, For image samples The predicted baseline value, For image samples The true baseline value; the loss function for the curve crossing the set of labeled points, including: in, The loss value is the fitting value for the point set. The number of marking points for each sinusoidal seam. For image samples The one marker coordinate, To predict the curve in place coordinate, To be accurately labeled place Coordinates; amplitude sign consistency loss function, including: in, This represents the amplitude sign consistency loss value. The loss that occurs when the amplitude is negative.

6. The well logging imaging sinusoidal fracture identification method based on PolylaneNet and noise perception as described in claim 5, characterized in that, The adaptive Huber loss function is reconstructed to obtain the improved PolylaneNet model, including: in, 、 、 、 、 、 、 、 、 、 as well as All are preset weights. For the left boundary loss, For the right boundary loss, For the upper boundary loss, The sum of the lower boundary loss, the right boundary loss, and the left boundary loss is the consistency loss value of the ordinates of the left and right boundaries. The sum of the lower boundary loss and the upper boundary loss is the correlation loss value between the upper and lower boundaries and the amplitude. This represents the total loss.

7. The well logging imaging sinusoidal fracture identification method based on PolylaneNet and noise perception as described in claim 1, characterized in that, The process involves acquiring and preprocessing structured logging resistivity data to obtain the input dataset. This includes: acquiring structured logging resistivity data and converting it into image data; labeling the location of sinusoidal fractures and sinusoidal fitting parameters in the image data, including amplitude, phase, and vertical displacement; dividing the labeled image data into training, validation, and test sets, and adapting it to the PolylaneNet model input format.

8. A well logging imaging sinusoidal fracture identification device based on PolylaneNet and noise perception, characterized in that, include: The information acquisition module is used to execute step S11, including: acquiring structured logging resistivity data and preprocessing it to obtain the input dataset, wherein the preprocessing includes image data conversion, data annotation, and dataset partitioning; the model improvement module is used to execute step S12, including: improving the initial PolylaneNet model, training the improved PolylaneNet model based on the input dataset, and using the trained PolylaneNet model for logging imaging sinusoidal fracture recognition, including S121-S124: S121, replacing the third-order polynomial fitting function in the initial PolylaneNet model. S122: Obtain the image quality factor and construct the adaptive Huber loss function based on the dynamic threshold calculated by the adaptive error threshold function; S123: Reconstruct the adaptive Huber loss function to obtain the improved PolylaneNet model, wherein the reconstruction terms include confidence, boundary constraints, sine curve similarity, amplitude, phase, baseline, point set fitting, and amplitude sign constraint; S124: Train the improved PolylaneNet model based on the input dataset, and when the preset training requirements are met, use the improved PolylaneNet model for well logging imaging sine seam recognition.

9. An electronic device, characterized in that, include: processor; A memory for storing processor-executable instructions; wherein the processor is configured to execute to implement a well logging imaging sinusoidal fracture identification method based on PolylaneNet and noise perception as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium, characterized in that, When the instructions in the storage medium are executed by the processor of the electronic device, the electronic device is able to perform a well logging imaging sinusoidal seam identification method based on PolylaneNet and noise perception as described in any one of claims 1 to 7.

Images