Intelligent segmentation model and system for natural fractures of tight sandstone core

Abstract

The invention relates to the technical field of rock core fracture segmentation, in particular to an intelligent segmentation model and system for natural fractures of a tight sandstone rock core, and the model comprises a feature extraction unit, a feature fusion unit and a segmentation unit. The feature extraction unit comprises a convolution module, a C2f module and an SPPF module; the feature fusion unit comprises a convolution module, a splicing module, a CARAFE module and an improved Neck network structure BiFPN; the dividing unit comprises a dividing head module. According to the method, the structure of an up-sampling module and a Neck network of a neck network (Neck) of an original YOLOv8m network is improved, a natural fracture segmentation network FracNet suitable for the tight sandstone core is formed, and the recognition precision and efficiency of the natural fracture of the tight sandstone core can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of core fracture segmentation, and particularly relates to an intelligent segmentation model and system for natural fractures in tight sandstone cores. Background Art

[0002] Natural fractures in cores are one of the important characteristics of tight sandstone reservoirs, which have a significant impact on the migration ability of fluids in low-porosity and low-permeability tight reservoirs, and also determine to a certain extent the effect of water flooding or gas flooding development. By segmenting the natural fractures in the core, the reservoir space and fluid channels can be more accurately evaluated, providing key data support for the evaluation of oil and gas reserves, and thus improving the accuracy of reservoir evaluation; secondly, the segmentation and identification of natural fractures in tight sandstone cores play an important role in optimizing oil and gas field development strategies; in addition, it can provide accurate data support for oil and gas field development, so as to reasonably design development means such as horizontal wells and fracture fracturing, and optimize the resource recovery rate. Therefore, the segmentation and identification of natural fractures in tight sandstone reservoirs play an important role in revealing the development characteristics and distribution patterns of natural fractures in tight sandstone reservoirs.

[0003] At present, the identification of natural fractures in tight sandstone drilling cores mainly relies on professionals to visually observe and identify the cores on site, which requires high professional experience for researchers and has a high time cost. At present, relevant intelligent identification algorithms and technologies mainly focus on road bridges and material cracks (Cracks), and combine technologies such as threshold algorithms, edge algorithms, region algorithms, matching algorithms, and fuzzy algorithms to record the position and shape of cracks. There are few records on the intelligent identification of natural fractures in tight sandstone reservoir drilling cores. The current method of manual identification is highly subjective, prone to omissions and errors, resulting in problems such as inaccurate measurement, low efficiency, and incomplete records; moreover, traditional algorithms have low accuracy in detecting crack characteristics and complex steps, which limits their applicability and makes it difficult to meet the actual research analysis, exploration and development, and engineering construction needs of the oil and gas industry for the identification of core natural fractures.

[0004] Traditional fracture segmentation methods include thresholding, edge detection, morphological processing, etc. These methods process core fracture images by manually setting rules or specific algorithms. However, these methods are insufficient in dealing with complex fracture structures and are easily affected by factors such as illumination changes and noise interference, resulting in inaccurate segmentation results. Wu Feng et al. proposed a multi-scale fracture filtering kernel superposition noise reduction method, which uses multiple filtering kernels to superpose and filter and denoise the core CT scan images, and then superposes the filtering results. On this basis, a multi-scale fracture filtering kernel superposition noise reduction method is proposed, which uses multiple filtering kernels to superpose and filter and denoise the core CT scan images, and then superposes the filtering results. Finally, the core fractures are segmented and identified. Although the authors used many filtering methods and combined them with methods such as threshold segmentation to achieve good results. This method is not universal and has low efficiency in segmentation and identification. Among them, the fracture extraction effect is as Figure 1 shown.

[0005] Lu Hongyu et al. converted the RGB of the fracture image into an HSV color space domain graph, used the color difference of the fracture filling materials to identify the fractures in the graph, and then used binaryzation processing and morphological processing. This method strongly relies on the color difference of the fracture filling materials, such as white calcite, white quartz, etc. The fractures extracted by this method have strong limitations. Using HSV to process fracture photos cannot distinguish the labels and fractures clearly. It is only the binaryzation method based on local thresholds under image region block that separates the fractures and the background, as Figure 2 shown. Each core photo needs to be processed, and the operation steps are complex, the workload is huge, and it is difficult to carry out universality. Shen Ke et al. used the improved PSPNet to identify the fractures in the electrical imaging logging. Although the network parameters were optimized, the segmentation performance and the number of parameters could not be balanced, resulting in low accuracy and efficiency of fracture identification.

[0006] In summary, the current existing technologies have low accuracy and efficiency in identifying natural fractures in tight sandstone cores. Summary of the Invention

[0007] To solve the above problems, the present invention provides an intelligent segmentation model and system for natural fractures in tight sandstone cores, which is used to improve the accuracy and efficiency of identifying natural fractures in tight sandstone cores.

[0008] To achieve the above object, the technical solution of the present invention is as follows: On the one hand, an intelligent segmentation model for natural fractures in tight sandstone cores is proposed, which is constructed based on the YOLOv8m network and includes a feature extraction unit, a feature fusion unit, and a segmentation unit; The feature extraction unit includes: a convolution module, a C2f module, and an SPPF module; The feature fusion unit includes: a convolution module, a splicing module, a CARAFE module, and an improved Neck network structure BiFPN; The segmentation unit includes: a segmentation head module Furthermore, the CARAFE module replaces the upsampling module of the neck network in the YOLOv8m network, which is used to improve the segmentation speed of the neck network and the segmentation accuracy of edge serrated cracks.

[0009] Furthermore, the improved Neck network structure BiFPN is used to enhance the feature extraction ability of the YOLOv8m network, achieve efficient aggregation of multi-scale features, and strengthen the segmentation ability of the YOLOv8m network for natural fractures in complex dense sandstone cores.

[0010] Furthermore, the CARAFE module includes an upsampling kernel prediction module and a feature recombination module; The upsampling kernel prediction module is used to reduce the number of channels of the input H×W×C feature map to through 1×1 convolution operation; and through convolution operation, the number of channels is further transformed into ; then the channels are spatially expanded, and the generated upsampling kernel is subjected to softmax normalization to ensure that the sum of its weights is 1; The feature recombination module is used to reverse-map the position of each output feature map to the input feature map; perform a dot product operation on the original feature map in the area centered on this point and the corresponding predicted upsampling kernel; different channels share the same upsampling kernel at the same position to generate the final new feature map.

[0011] Furthermore, the improved Neck network structure BiFPN is used to connect nodes with the same feature map size; it is also used to establish a new path to connect the original input node and the output node when the original input node and the output node are in the same layer; a weighted feature fusion mechanism is also introduced.

[0012] Furthermore, the weights in the weighted feature fusion mechanism are learned using the fast normalization method, and its formula is as follows: In the formula, is the number of fused feature maps at the node; is the input feature map at the node; , are the weights attached to the input feature map, and the initial values of the weights are randomly selected between 0 and 1; is a constant used to make the denominator non-zero.

[0013] Furthermore, the dataset of natural fracture images of tight sandstone cores is sourced from the area where the intelligent segmentation model of natural fractures in tight sandstone cores to be applied is constructed.

[0014] On the other hand, an intelligent segmentation method for natural fractures in tight sandstone cores is proposed, which also includes: the above-mentioned intelligent segmentation model of natural fractures in tight sandstone cores and a data input module; The data input module is used to obtain the natural fracture image of the tight sandstone core to be segmented and input it into the intelligent segmentation model of natural fractures in tight sandstone cores; The intelligent segmentation model of natural fractures in tight sandstone cores is used to receive the natural fracture image of the tight sandstone core to be segmented and output the segmentation result of the core natural fractures.

[0015] Beneficial effects: The present invention provides an efficient method for segmenting natural fractures in tight sandstone cores based on the FracNet network. First, the upsampling module of the original neck network (Neck) of YOLOv8m is replaced with the CARAFE module with lightweight calculation, enabling the neck network to fully utilize the semantics of the feature map during feature fusion. Second, after the neck network adopts the CARAFE module with lightweight calculation, the segmentation speed can be improved and the segmentation accuracy of jagged fractures at the edge can be enhanced.

[0016] The benchmark YOLOv8m network uses the FPN mode for feature fusion in the feature fusion stage. The natural fractures in tight sandstone cores have characteristics such as large differences in size. Using the FPN mode for feature fusion structure will lead to low fusion efficiency of core fractures, especially the segmentation effect of multi-scale core fractures is not ideal. Therefore, a new Neck network structure BiFPN is proposed in the neck of YOLOv8m to enhance the feature extraction ability of the Neck network, achieve efficient aggregation of multi-scale features, and strengthen the segmentation ability of the YOLOv8 network for complex natural fractures in tight sandstone cores.

[0017] In summary, the present invention effectively improves the recognition accuracy and efficiency of natural fractures in tight sandstone cores.

[0018] The additional aspects and advantages of the present invention will be partially given in the following description, partially will become apparent from the following description, or be understood through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is the effect diagram of crack extraction studied by Wu Feng et al. in the background technology of the present invention; Figure 2 It is the effect diagram of crack extraction studied by Lu Hongyu et al. in the background technology of the present invention; Figure 3 It is the flow chart of the construction and use steps of the natural fracture model of tight sandstone cores in the embodiment of the present invention; Figure 4 Flow chart of manual annotation and format conversion of core fractures in an embodiment of the present invention; Figure 5 Schematic diagram of the structure of the CARAFE upsampling operator in an embodiment of the present invention; Figure 6 Schematic diagram of the feature fusion network with four different feature fusion ideas in an embodiment of the present invention; Figure 7 Schematic diagram of the structure of the natural fracture model of tight sandstone core in an embodiment of the present invention; Figure 8 Schematic diagram of the segmentation effect of the natural fracture model of tight sandstone core in an embodiment of the present invention; Figure 9 Schematic diagram of the structure of the intelligent segmentation system for natural fractures of tight sandstone core in an embodiment of the present invention. Detailed implementation manners

[0020] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0021] The following will be further described in detail through specific implementation manners: Embodiment: An intelligent segmentation model for natural fractures of tight sandstone core. This segmentation model is constructed based on the YOLOv8m network and mainly consists of a feature extraction unit, a feature fusion unit, and a segmentation unit. Referring to Figure 7 As shown, the feature extraction unit includes: a convolution module, a C2f module, and an SPPF module; the feature fusion unit includes: a convolution module, a splicing module, a CARAFE module, and an improved Neck network structure BiFPN; the segmentation unit includes: a segmentation head module.

[0022] In the segmentation model of this embodiment, the CARAFE module is used to replace the upsampling module of the neck network in the original YOLOv8m network to improve the segmentation speed of the neck network and the segmentation accuracy of edge serrated fractures. After adopting a lightweight model in the neck network, the segmentation speed can be improved. In this embodiment, the improved Neck network structure BiFPN is used to enhance the feature extraction ability of the YOLOv8m network, achieve efficient aggregation of multi-scale features, and strengthen the segmentation ability of the YOLOv8m network for complex natural fractures of tight sandstone core.

[0023] In this embodiment, the CARAFE module is integrated to address the limitations of segmenting jagged natural core fractures. The improvement method is to replace the upsampling module in the neck network structure with the CARAFE upsampling operator (CARAFE module). The CARAFE upsampling operator is a novel upsampling method. It fully utilizes the semantics of the feature map and can effectively perform lightweight calculations. Therefore, the CARAFE module structure is used to optimize the upsampling algorithm of YOLOv8m, increasing the receptive field of the network while maintaining lightweight upsampling and obtaining a better high-resolution feature map. The advantages of the CARAFE upsampling operator are less redundancy, strong feature fusion ability, and fast operation speed. It abandons the single-kernel sampling method for samples using the nearest neighbor interpolation method and proposes a new method based on multi-core data, that is, a new method based on multi-core data is adopted to generate an adaptive content-aware sampling method.

[0024] In Figure 5 the structure of the CARAFE upsampling operator is given. The CARAFE module structure is used to optimize the upsampling algorithm of YOLOv8m, increasing the receptive field of the network while maintaining lightweight upsampling. The CARAFE module includes an upsampling kernel prediction module and a feature recombination module; the upsampling kernel prediction module is used to reduce the number of channels of the input H×W×C feature map to through a 1×1 convolution operation; and through convolution operations, the number of channels is further transformed into ; then the channels are spatially expanded, and the generated upsampling kernel is subjected to softmax normalization to ensure that the sum of its weights is 1; the feature recombination module is used to reverse-map the position of each output feature map to the input feature map; and perform a dot product operation on the original feature map in the area centered on this point and the corresponding predicted upsampling kernel; different channels share the same upsampling kernel at the same position to generate the final new feature map.

[0025] Compared with the nearest neighbor interpolation upsampling technique, the CARAFE method significantly enhances the semantic richness of the recombined feature map. This method is achieved by emphasizing the key points within the local area. In the case of dense natural small fractures in tight sandstone cores, the ability of CARAFE to enhance spatial details helps the model distinguish small fractures that are very close to each other, thereby potentially reducing the number of merged segments. It also helps to improve the localization accuracy of the model in segmenting natural fractures in tight sandstone cores. In addition, CARAFE has a wider observation range, proficient content processing, and a lightweight design, ultimately achieving fast calculations.

[0026] In the initial network structure, high-level pyramid features are directly extracted from the backbone network for prediction. This structure lacks feature fusion, resulting in relatively low segmentation accuracy. With the in-depth study of network structures, the FPN network based on the idea of feature fusion was proposed. The structure is as shown in Figure 6 (a). A new top-down path is created for feature fusion. The connected feature map obtains more abundant semantic information, and prediction can improve the accuracy to a certain extent. However, the FPN network is essentially a top-down structure. Due to the limitation of unidirectional information transmission, the accuracy is still difficult to meet the requirements. In recent years, the PANet network, which is most frequently used, is used as the Neck in YOLOv8m. The structure is as follows Figure 6 (b). It builds a bottom-up path on the basis of FPN to make up for the defect of the FPN network with only a unidirectional information flow structure. The high-level feature map has stronger semantic information, which is beneficial to object classification and segmentation. The low-level feature map has stronger position information, which is beneficial to object localization. Such a structure can greatly improve the accuracy of the target segmentation task.

[0027] At the same time, the recently proposed NAS-FPN structure has the following specific structure Figure 6 (c). The popular NAS (Neural Architecture Search) technology is used to search for the best network structure. Although the effect of this structure is the best, the network obtained based on the search is irregular, difficult to interpret and modify, and the use of NAS technology consumes a lot of time and labor intensity, so it has not been widely used.

[0028] Based on this, the present embodiment proposes a new Neck network structure, BiFPN, as follows Figure 6 (d). Compared with the PANet structure, the design changes of BiFPN are as follows: (1) Connect nodes with the same feature map size. For example, in the backbone network, nodes in the p3 layer and p5 layer with the same size as the feature map of the neck network are spliced. The remaining nodes in the backbone network do not have other-direction inputs for joint feature fusion, so these nodes are not connected in the network structure with multi-scale fusion, which simplifies the bidirectional network structure;

[0029] (2) When the original input node and the output node are in the same layer, a new path is established to connect the original input node and the output node. Such a structure can fully fuse more feature information at the cost of a small increase.

[0030] (3) Introduce the Weighted Feature Fusion mechanism. Traditional feature fusion often simply uses Concat or Shortcut to connect feature maps, without differentiating the simultaneously added feature maps. However, the input feature maps have different resolutions, and their contributions to fusing the input feature maps are also different. Therefore, simply adding or superimposing them is not the best operation. So, a simple and efficient weighted feature fusion mechanism is proposed, adding additional weights to each input, mainly to learn the importance of different input features and fuse different input features with discrimination.

[0031] In the weighted feature fusion mechanism, the weights are learned using the fast normalization method, and its formula is as follows: In the formula, is the number of fused feature maps at the node; is the input feature map at the node; and are the weights attached to the input feature map, and the initial values of the weights are randomly selected between 0 and 1; is a constant used to make the denominator non-zero.

[0032] Set attached Figure 3 As shown, the steps for constructing and using the above improved intelligent segmentation model for natural fractures in tight sandstone cores are as follows: S100. Make a dataset of natural fracture images of tight sandstone cores; Specifically, assuming that the core fracture photos in the Bozidabei block of the Tarim Basin are used as the data basis, use LabelImg to calibrate the dataset for segmenting natural fractures in tight sandstone cores. LabelImg is a graphic image annotation tool. The annotations are saved as.json files in the used format, and then the.json is converted into a.txt file in coco format using the format modification algorithm. The specific steps for manual annotation and format conversion of core fractures are as Figure 4 shown. After making the dataset of natural fractures in tight sandstone cores using LabelImg, input the dataset of natural fractures in tight sandstone cores into the network, and let the network learn the natural fractures in the tight sandstone cores in the dataset and segment the natural fractures in the tight sandstone cores. Through the pictures input at the input end, the input image size of the efficient FracNet natural fracture segmentation network for tight sandstone cores is 640*640, usually including the picture preprocessing stage, that is, scaling the input image to the input size of the network and performing normalization, etc. In the network training stage, the efficient FracNet natural fracture segmentation network for tight sandstone cores uses Mosaic data augmentation to improve the training speed and network accuracy of the model.

[0033] S200. Establish a natural fracture segmentation network for tight sandstone cores based on FracNet, that is, the improved intelligent segmentation model of natural fractures in tight sandstone cores mentioned above; S300. Use the dataset of natural fracture images of tight sandstone cores produced to train the natural fracture segmentation network for tight sandstone cores based on FracNet to obtain an intelligent segmentation model of natural fractures in tight sandstone cores; Specifically, the hardware platform configuration parameters for training the segmentation model in this embodiment are as follows: a graphics card of NVIDIA GeForce RTX A5000 GPU; software configuration: a 64-bit Windows 11 operating system, a framework of Detectron2 based on PyTorch, CUDA 11.3, OpenCV2 library, and PyCharm integrated development environment.

[0034] S400. Input the natural fracture image of the tight sandstone core to be segmented into the intelligent segmentation model of natural fractures in tight sandstone cores to segment the natural fractures in the core.

[0035] Specifically, use the dataset of natural fracture images of tight sandstone cores to train an efficient FracNet natural fracture segmentation network for cores to obtain a natural fracture segmentation model. Segmentation is the ultimate goal of the output. For different segmentation algorithms, the number of branches at the output end is not the same, usually including a classification branch and a regression branch. YOLOv8m uses CIOU_Loss to replace the Smooth L1 Loss function and uses DIOU_nms to replace the traditional NMS operation, thereby further improving the segmentation accuracy of the algorithm. The output end represents the output picture. The types and confidence levels of natural fractures in the core are obtained through the efficient FracNet natural fracture segmentation network model for cores, and the fracture area is mapped through the mask.

[0036] Correspondingly, this embodiment provides an intelligent segmentation system for natural fractures in tight sandstone cores, and its structure is as Figure 9 shown, including a data input module and the intelligent segmentation model of natural fractures in tight sandstone cores of this well mentioned above. Among them, the data input module is used to obtain the natural fracture image of the tight sandstone core to be segmented and input it into the intelligent segmentation model of natural fractures in tight sandstone cores; the intelligent segmentation model of natural fractures in tight sandstone cores is used to receive the natural fracture image of the tight sandstone core to be segmented and output the segmentation result of the natural fractures in the core.

[0037] Compare and demonstrate the method adopted in the present invention with the prior art: The speed evaluation metric is represented by the prediction time for a single image. The faster the speed, the higher the segmentation efficiency and the better it meets the requirements of real-time segmentation. The unit of time is milliseconds. The precision evaluation metric is the mean Average Precision (mAP). With P as the vertical axis and R as the horizontal axis, a PR curve is obtained. The value of the Average Precision (AP) is equal to the area under the PR curve. The AP value measures the segmentation accuracy of the model for each category, and mAP is the mean of AP, reflecting the performance of the model for all categories. The larger the mAP value, the higher the segmentation precision of the algorithm.

[0038] Where TP represents the number of samples correctly classified by the model; FP represents the number of samples misclassified by the model; FN represents the number of samples missed by the model.

[0039] To verify the effectiveness of the efficient FracNet natural fracture segmentation network method for tight sandstone core natural fracture datasets, an experimental effect evaluation was conducted. In the following experiments, P, R, mAP@0.5 (the average accuracy of the mask when the IoU threshold is 0.5), Time, and the number of parameters were used as evaluation metrics for the model. In the experiment, the YOLOv8m model was used as the baseline segmentation performance, with epochs = 500 and batch_size = 32 by default. Tensorboard was used to view the training results of the algorithm model.

[0040] The upsampling of the YOLOv8m neck network was improved from upsampling to the CARAFE operator, forming the YOLOv8-C-Seg natural fracture segmentation model. The operation method based on the CARAFE operator greatly reduced the inference speed of the model. To study the impact of the improvement of the CARAFE operator for network upsampling on the performance of natural fracture segmentation of cores. Among them, the resolution of the core fracture dataset images provided for network learning is 2560x1920. One type of fracture was selected and named fracture. There are 341 sample images in the dataset. 273 images are divided into the training set, 34 images are divided into the test set, and 34 images are divided into the validation set, with a ratio of 8:1:1. The prediction results were compared with the basic YOLOv8m network. Table 1 shows the comparison of the model segmentation performance before and after the improvement of the CARAFE operator for upsampling.

[0041] Table 1 Segmentation performance before and after the improvement of the CARAFE operator for the neck network upsampling As shown in Table 1, the segmentation accuracy of the basic YOLOv8m algorithm is 83.4%. After adopting the upsampling CARAFE operator in the neck network, the accuracy drops to 89.6%. However, the recall rate increases by 1.2% and reaches 73.6%. At the same time, the mAP increases by 0.5% and reaches 83.9%. In terms of segmentation time, it is found that after adopting the upsampling CARAFE operator in the neck network, the model reduces memory consumption, and the segmentation time is reduced from 12.2 milliseconds to 9.9 milliseconds, and the segmentation speed is increased by about 18.85%, showing a huge improvement in the segmentation speed.

[0042] After adopting the upsampling CARAFE operator in the neck network, the recall rate and mAP are improved, but the improvement effect of mAP is very small. When the feature information extracted from the neck network is limited, an improvement is proposed for the feature fusion part of the neck to achieve the efficient aggregation of multi-scale feature information output by the backbone network, thereby improving the segmentation performance. Therefore, a new BiFPN structure is proposed. To verify the effectiveness of the improved feature pyramid fusion structure, training and validation are carried out on the same dataset, and comparisons are made on the accuracy, recall rate, segmentation performance, and segmentation time of YOLOv8m, YOLOv8-C-Seg, and YOLOv8-C-BIFPN-Seg (FracNet).

[0043] Table 2 Model parameters and segmentation performance before and after improving feature fusion As shown in the results of Table 2, when PANet is improved to the new BiFPN structure, the experimental results show that the fusion of the new BiFPN and YOLOv8m enables the network structure to learn the importance of different input features at the feature fusion nodes. Without adding too much cost, it strengthens the degree of feature aggregation and improves the segmentation performance of the model. FracNet performs best in terms of recall rate, reaching 77%, indicating that it has the strongest ability to segment all cracks. At the same time, it also leads in the mAP0.5 index, which is 84.8%, reflecting its optimal overall performance in the crack segmentation task. Although the segmentation speed of the FracNet network increases by 0.3 milliseconds compared with the YOLOv8-C-Seg network, it can still meet the real-time segmentation requirements for deployment.

[0044] Compared with YOLOv8m, the recall rate and mAP of FracNet are increased by 4.6% and 1.4% respectively, and the segmentation time is reduced by 16.39%. It still maintains an advantage in terms of segmentation time, enhancing the network's real-time segmentation ability. This shows that adding the upsampling operator CARAFE to the neck network of the benchmark YOLOv8m and improving PANet to the new BiFPN structure can greatly improve the network's performance in capturing all core cracks, ensuring the comprehensiveness and efficiency of the network.

[0045] To further verify the superiority of the improved network in the multi-scale natural fracture segmentation of tight sandstone cores, classical one-stage network models such as the YOLOv8 series, YOLOv9 series, and YOLOv11 series methods were compared. The comparison of the segmentation performance of each model is shown in Table 3.

[0046] Table 3 Comparison of the performance of each model As shown in Table 3 for the comparison of performance. In the fracture segmentation task, the FracNet model demonstrated significant advantages. This model performed well in multiple key performance indicators, especially in terms of precision and mAP0.5. The precision value of FracNet was 86.4%, although slightly lower than 94.3% of YOLOv8s, but it achieved a recall rate of 77%, which was the highest among all the compared models, indicating that it was more comprehensive in segmenting fracture instances.

[0047] In terms of the mAP@0.5 metric, FracNet exceeded YOLOv9e, whose number of parameters was 2.24 times that of FracNet, with an excellent segmentation performance of 84.8%. This reflects the overall superior performance of FracNet in the fracture segmentation task, especially when dealing with complex fracture scenarios.

[0048] The processing time of FracNet was 10.2 milliseconds, comparable to that of YOLOv8m. Considering its excellent performance in precision and recall, this processing time was acceptable. In addition, the number of parameters of FracNet was 24.83, indicating that it had good model complexity and efficiency while maintaining high performance. The final segmentation effect diagram of the efficient FracNet core natural fracture segmentation network is as Figure 8 shown.

[0049] In summary, FracNet performed well in terms of precision, recall, and mAP0.5, making it an ideal choice for the natural fracture segmentation task of tight sandstone cores. Even when dealing with complex tight sandstone core natural fracture segmentation datasets, FracNet could effectively segment the fractures, demonstrating its strong segmentation performance and robustness. These characteristics make FracNet have important application value in practical applications, especially in fracture segmentation scenarios that require high recall and high precision.

[0050] Aiming at the problems of low efficiency and cumbersome steps in the natural fracture segmentation of tight sandstone cores by traditional methods, an efficient FracNet network for natural fracture segmentation of tight sandstone cores is proposed. First, by constructing the lightweight upsampling operator CARAFE module, the CARAFE module is selected to replace the upsampling operation in the feature fusion network to form the YOLOv8-C-Seg network model. On this basis, a new BiFPN structure is proposed as the feature fusion network. The new BiFPN structure is an improvement of the BiFPN structure, and at the same time, combined with the weighted fusion mechanism, the segmentation performance of the algorithm for natural fractures in cores is improved. The results show that the mean average precision of the improved FracNet reaches 84.8%. At the same time, compared with the basic YOLOv8m network, the segmentation time is reduced by 16.39%, which improves the segmentation speed of the network, meets the real-time requirements for deployment to embedded and mobile devices, and at the same time ensures the performance of natural fracture segmentation of cores.

[0051] Obviously, the above embodiments are only examples given for clear illustration and not limitations on the implementation manners. For those of ordinary skill in the art, other different forms of changes or variations can be made based on the above description. It is not necessary and impossible to list all the implementation manners here. And the obvious changes or variations derived therefrom are still within the protection scope of the present invention.

Claims

1. An intelligent segmentation model for natural fractures in dense sandstone cores, based on the FracNet network constructed by improving YOLOv8m, is characterized by: It includes a feature extraction unit, a feature fusion unit and a segmentation unit; The feature extraction unit includes: convolution module, C2f module and SPPF module; The feature fusion unit includes: convolution module, splicing module, CARAFE module and improved Neck network structure BiFPN; The segmentation unit includes: a segmentation head module.

2. The natural fracture intelligent segmentation model of dense sandstone core according to claim 1 is characterized in that: The CARAFE module replaces the upsampling module of the neck network in the YOLOv8m network to improve the segmentation speed of the neck network and the segmentation accuracy of jagged cracks on the edges.

3. The natural fracture intelligent segmentation model of dense sandstone core according to claim 2 is characterized in that: The improved Neck network structure BiFPN is used to enhance the feature extraction capability of the YOLOv8m network, achieve efficient aggregation of multi-scale features, and strengthen the segmentation capability of the YOLOv8m network for natural fractures in complex dense sandstone cores.

4. The natural fracture intelligent segmentation model of dense sandstone core according to claim 3 is characterized in that: The CARAFE module includes an upsampling kernel prediction module and a feature recombination module; The upsampling kernel prediction module is used to reduce the number of channels of the input H×W×C feature map to ; and through convolution operation, the number of channels is further converted into ; Then expand the spatial dimension of the channel and perform softmax normalization on the generated upsampling kernel to ensure that the sum of its weights is 1; The feature reorganization module is used to reversely map the position of each output feature map to the input feature map; The original feature map of the region and the corresponding predicted upsampling kernel are subjected to dot product operation; Different channels share the same upsampling kernel at the same position, producing the final New feature map.

5. The natural fracture intelligent segmentation model of dense sandstone core according to claim 4 is characterized in that: The improved Neck network structure BiFPN is used to connect nodes with the same feature map size; it is also used to establish a new path connecting the original input node and the output node when the original input node and the output node are in the same layer; and a weighted feature fusion mechanism is also introduced.

6. The natural fracture intelligent segmentation model of dense sandstone core according to claim 5 is characterized in that: The learning of weights in the weighted feature fusion mechanism adopts a fast normalization method, and its formula is as follows: In the formula, is the number of fused feature maps at the node; is the feature map input at the node; , To attach weights to the input feature map, randomly select initial weight values ​​between 0 and 1; A constant used to make the denominator non-zero.

7. The natural fracture intelligent segmentation model of dense sandstone core according to claim 6 is characterized in that: The dataset of natural fracture images of tight sandstone cores comes from the area where the constructed intelligent segmentation model of natural fractures in tight sandstone cores is to be applied.

8. A natural fracture intelligent segmentation system for dense sandstone cores, comprising the natural fracture intelligent segmentation model for dense sandstone cores according to any one of claims 1 to 7, characterized in that: Also includes: Data input module; The data input module is used to obtain the natural fracture image of the dense sandstone core to be segmented, and input it into the natural fracture intelligent segmentation model of the dense sandstone core; The intelligent segmentation model of natural fractures in dense sandstone cores is used to receive natural fracture images of dense sandstone cores to be segmented and output natural fracture segmentation results of the cores.

Images