Natural gas hydrate CT image threshold segmentation method and storage medium

Abstract

The invention discloses a natural gas hydrate CT image threshold segmentation method and a storage medium, and belongs to the technical field of natural gas hydrate detection.The method comprises the steps that S1, a natural gas hydrate sample image obtained through synchrotron radiation CT scanning is obtained, the image is preprocessed, and a central area image is obtained; s2, performing edge recognition on the preprocessed central region image by using a convolutional neural network and combining a Sobel operator to obtain an edge feature matrix; s3, clustering the edge feature matrix by adopting a KMeans clustering algorithm, and determining gray boundary values of the solid particles and the methane bubbles; s4, removing gray scale parts of solid particles and methane bubbles in the image according to the gray scale boundary value, and performing maximum and minimum normalization processing on the residual gray scale values of water and hydrates to obtain a gray scale probability density; and S5, drawing a normalized gray probability density map according to the gray probability densities of all the images, and determining a threshold segmentation range of the hydrate according to the probability density peak displacement. The method can automatically extract the hydrate threshold range, and improves the segmentation precision.

Description

Technical Field

[0001] This invention relates to the technical field of natural gas hydrate detection, and in particular to a threshold segmentation method and storage medium for CT images of natural gas hydrates. Background Technology

[0002] Natural gas hydrates are widely found in marine sediments and permafrost, and their exploration and development have attracted considerable attention as a potential clean energy source. In recent years, research on hydrates has mainly focused on their pore structure, permeability, morphological evolution, and mechanical properties. In the study of the microscopic pore structure of hydrates, CT scanning technology has been widely applied due to its ability to provide high-resolution, non-destructive images.

[0003] X-ray CT scanning technology allows direct observation of the growth behavior and detailed pore structure of hydrates within pores, contributing to a better understanding of their pore habits and permeability. Microscopic CT imaging technology enables three-dimensional visualization of the distribution and morphological characteristics of hydrates in sandy sediments under laboratory conditions. These microscopic imaging studies have significant guiding implications for numerical simulations. However, because hydrates and water have similar attenuation coefficients for X-rays, they are difficult to distinguish in CT images. Some studies have used normalized CT image histograms or machine learning algorithms (such as random forests, CNNs, and SVMs) to model saturation, but these methods suffer from complex model design and poor adaptability.

[0004] The information disclosed in this background section is intended only to enhance the understanding of the overall background of the invention and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a threshold segmentation method and storage medium for CT images of natural gas hydrates, which can automatically extract the threshold range of hydrates and improve the accuracy of segmentation.

[0006] To achieve the above objectives, the present invention provides a thresholding method for CT images of natural gas hydrates, comprising the following steps:

[0007] S1: Acquire images of natural gas hydrate samples obtained from synchrotron radiation CT scans, and preprocess the images to obtain images of the central region;

[0008] S2: Use a convolutional neural network combined with the Sobel operator to perform edge recognition on the preprocessed central region image to obtain the edge feature matrix;

[0009] S3: The KMeans clustering algorithm is used to cluster the edge feature matrix to determine the grayscale boundary values ​​between solid particles and methane bubbles;

[0010] S4: Remove the grayscale components of solid particles and methane bubbles from the image based on the grayscale boundary values, and perform maximum and minimum normalization on the remaining grayscale values ​​of water and hydrates to obtain the grayscale probability density.

[0011] S5: Based on the gray-level probability density of all images, draw a normalized gray-level probability density map, and determine the threshold segmentation range of hydrates based on the peak displacement of the probability density.

[0012] In one embodiment of the present invention, in step S1, acquiring the image of the natural gas hydrate sample obtained by synchrotron radiation CT scanning includes setting the scanning frequency to 15 minutes / scan.

[0013] In one embodiment of the present invention, image preprocessing includes cropping the original 2560×2560 pixel CT image to a central region of 1280×1280 pixels.

[0014] In one embodiment of the present invention, the convolutional neural network used in step S2 includes an input layer, a convolutional layer, and a fully connected layer; the input of the input layer is the preprocessed central region image in step S1, the convolutional layer performs convolution operations using edge recognition operators, and the fully connected layer outputs the result matrix after the convolution operation.

[0015] In one embodiment of the present invention, the convolution kernel in the convolutional layer is set to the Sobel operator to perform convolution operation, identify the edges of solid particles and methane bubbles, and perform subsequent numerical determination.

[0016] In one embodiment of the present invention, in step S3, the KMeans algorithm divides the samples into two clusters.

[0017] In one embodiment of the present invention, in step S3, the grayscale boundary values ​​corresponding to the gas and solid are determined by the moving average method. The moving average method is used for calibration, which can improve the accuracy of bubble and solid boundary identification.

[0018] In one embodiment of the present invention, the calculation formula for determining the grayscale boundary value using the moving average method is as follows:

[0019]

[0020] Where t represents time. Represents the original grayscale value, while This represents the value after smoothing using the moving average method.

[0021] In one embodiment of the present invention, in step S5, the threshold segmentation range of the gray value interval corresponding to the hydrate is 0-0.18.

[0022] The present invention also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the natural gas hydrate CT image thresholding method as described above.

[0023] Compared with existing technologies, the natural gas hydrate CT image thresholding segmentation method and storage medium of the present invention obtain the gray value boundaries corresponding to solid particles and methane bubbles based on the unsupervised clustering algorithm of convolutional neural network (CNN) combined with K-Means. Then, the gray value of water and hydrate is extracted separately, and the gray value probability density map is drawn by using max-min normalization. By comparing the difference in gray value changes of CT images before and after hydrate generation, and combining machine learning, the thresholding segmentation range of water and hydrate images is finally obtained, which greatly improves the accuracy of image thresholding segmentation. Attached Figure Description

[0024] Figure 1 This is a flowchart of a natural gas hydrate CT image thresholding method according to an embodiment of the present invention;

[0025] Figure 2 This is a schematic diagram illustrating cropping an original image according to an embodiment of the present invention;

[0026] Figure 3 This is a grayscale distribution image after edge recognition according to an embodiment of the present invention;

[0027] Figure 4a This is a schematic diagram illustrating image recognition using a convolutional neural network with the Roberts operator according to an embodiment of the present invention.

[0028] Figure 4b This is a schematic diagram illustrating image recognition using a convolutional neural network with the Sobel operator according to an embodiment of the present invention;

[0029] Figure 4c This is a schematic diagram illustrating image recognition using a convolutional neural network with the Canny operator according to an embodiment of the present invention.

[0030] Figure 4d A schematic diagram illustrating image recognition using a convolutional neural network with a Laplacian operator according to an embodiment of the present invention.

[0031] Figure 4e This is a schematic diagram illustrating image recognition using a convolutional neural network with the Prewitt operator according to an embodiment of the present invention;

[0032] Figure 5This is a schematic diagram illustrating the relationship between SSE and the number of clusters when using the KMeans algorithm to distinguish particle (gas) edges according to an embodiment of the present invention;

[0033] Figure 6 This is a schematic diagram comparing the identification of bubbles and particles using the method of the present invention at different times according to an embodiment of the present invention;

[0034] Figure 7 This is a schematic diagram showing the relative gray values ​​of natural gas hydrate and water at various times according to an embodiment of the present invention. Detailed Implementation

[0035] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings, but it should be understood that the scope of protection of the present invention is not limited to the specific embodiments.

[0036] Unless otherwise expressly stated, throughout the specification and claims, the term "comprising" or its variations such as "including" or "comprises" shall be understood to include the stated elements or components without excluding other elements or other components.

[0037] like Figure 1 As shown, a thresholding method for CT images of natural gas hydrates according to a preferred embodiment of the present invention includes the following steps:

[0038] S1: Acquire images of natural gas hydrate samples obtained from synchrotron radiation CT scans, and preprocess the images to obtain images of the central region;

[0039] S2: Use a convolutional neural network combined with the Sobel operator to perform edge recognition on the preprocessed central region image to obtain the edge feature matrix;

[0040] S3: The KMeans clustering algorithm is used to cluster the edge feature matrix to determine the grayscale boundary values ​​between solid particles and methane bubbles;

[0041] S4: Remove the grayscale components of solid particles and methane bubbles from the image based on the grayscale boundary values, and perform maximum and minimum normalization on the remaining grayscale values ​​of water and hydrates to obtain the grayscale probability density.

[0042] S5: Based on the gray-level probability density of all images, draw a normalized gray-level probability density map, and determine the threshold segmentation range of hydrates based on the peak displacement of the probability density.

[0043] In this embodiment, synchrotron radiation CT scanning uses an X-ray beam generated by a synchrotron to scan a sample from multiple angles, and then generates a three-dimensional image using a detector array and computer reconstruction technology. The synchrotron accelerates electrons to a high-energy state, uses a magnetic field to make them move along a specific trajectory, and uses an insertion device (such as a oscillator) to radially deflect the electrons, generating an X-ray beam tangential to the trajectory. This technique can produce high-brightness, high-flux X-rays, covering the soft to hard X-ray band, and the energy is adjustable. By combining convolutional neural networks with the Sobel operator for precise edge identification and KMeans clustering for grayscale boundary determination, the method effectively solves the problem of distinguishing hydrates from water due to their similar X-ray attenuation coefficients, significantly improving segmentation accuracy. It eliminates the need for manually setting threshold parameters, automatically adapting to dynamic changes during hydrate formation through max-min normalization and peak shifting of the grayscale probability density map, making it suitable for various experimental conditions. The moving average method is used to calibrate the grayscale boundary values ​​between solids and bubbles, mitigating interference from scanning light variations, operational errors, and algorithm fluctuations, ensuring stable segmentation results. A complete automated process, from image preprocessing to threshold range determination, avoids the subjectivity of manual intervention in traditional methods and offers high single-image processing efficiency, meeting the analysis needs of large-scale time-series CT images. Furthermore, it is specifically designed for the characteristics of natural gas hydrate CT images, first separating solid particles from methane bubbles and then focusing on the grayscale difference analysis between water and hydrates, providing highly targeted and reliable technical support for the study of hydrate microporous structures.

[0044] Specifically, in step S1 of this invention, obtaining images of natural gas hydrate samples obtained by synchrotron radiation CT scanning includes: setting the scanning frequency to 15 minutes / scan; after the CT scan is completed, exporting image format data after CT reconstruction. The image data at each time state is a cross-section of the sand column at intervals of 0.01 mm, and the size of each image is 2560×2560 pixels. For a sample height of 10 mm, each group will have 1000 images.

[0045] Image preprocessing includes: cropping the original 2560×2560 pixel CT image to a central region of 1280×1280 pixels, using the cropped image as a dataset, and eliminating the influence of uneven grayscale at the edges of the field of view. For example... Figure 2 As shown, the image is cropped from its full size on the left to the center portion on the right.

[0046] In this embodiment, considering the slow formation of hydrates (40-50 hours), synchrotron radiation CT scanning is used. The scanning speed is very fast; traditional high-precision CT scans take 3-4 hours or even a day to complete one scan, while synchrotron radiation only requires 10-30 minutes. This allows for high-frequency capture of subtle changes in hydrate formation during critical stages, ensuring no core dynamic process is missed, while reducing scanning frequency during calmer stages, minimizing data storage pressure and redundant information. This achieves a precise balance between data effectiveness and storage / analysis costs. During preprocessing, the original image is cropped to a 1280×1280 pixel central region, directly eliminating the grayscale unevenness at the field of view caused by scanning characteristics. This avoids edge noise interfering with subsequent edge recognition and clustering analysis steps, improving the quality and consistency of the dataset. This lays a reliable foundation for machine learning algorithms to accurately extract features and determine grayscale boundaries, ensuring the accuracy of threshold segmentation from the source.

[0047] Specifically, in step S2 of the present invention, the convolutional neural network (CNN) used includes an input layer, a convolutional layer, and a fully connected layer.

[0048] The input layer takes the preprocessed central region image from step S1 as input, specifically a 1280×1280×1 (1280 pixels long, 1280 pixels wide, 1 channel) CT image. The convolutional layer uses an edge detection operator for convolution. The fully connected layer outputs the matrix resulting from the convolution operation. To preserve image accuracy, no pooling layer is used when the image size is 1280×1280.

[0049] The convolution operation of a convolutional layer involves selecting a matrix of the same size as the kernel on the original image, performing an inner product summation, storing the resulting value in the convolutional layer, then shifting to the next position, selecting a matrix and performing an inner product summation with the kernel, storing the result in the corresponding position, and so on.

[0050] Specifically, in step S2 of the present invention, the convolutional layer uses a 2×2 matrix as the convolutional kernel. The first selected image matrix is No offset The first result obtained by summing the inner products of the convolutional layers is: Then, it moves to the next position by sliding one pixel to the right, and the second selected image matrix is... The second result obtained by summing the inner products of the convolutional layers is: Store it to the right of the first result value, and so on.

[0051] Specifically, in step S2 of the present invention, the convolution kernel is set to the Sobel operator in the convolution layer to perform convolution operation, identify the edges of solid particles and methane bubbles, and perform subsequent numerical determination.

[0052] The Sobel operator can be used to locate the edge coordinates in the original image and extract the corresponding coordinates of the edge portion.

[0053] After convolution with the edge operator, two main colors remain in the image, corresponding to edge and non-edge regions, respectively. It's important to note that because particles / gases themselves blend with water / hydrates to some extent, the edges in CT images are not completely sharp, as shown below. Figure 3 The grayscale distribution after edge recognition is shown. The Sobel operator selected in this invention extracts edges with a certain width, and therefore the corresponding color values ​​also have a certain range.

[0054] Commonly used edge detection operators include the Roberts operator, Sobel operator, Canny operator, Laplacian operator, and Prewitt operator. These operators are all based on the discontinuity of grayscale values ​​at edges, segmenting the target region based on abrupt changes in grayscale. For example... Figures 4a to 4e The figures show convolutional neural networks using Roberts, Sobel, Canny, Laplacian, and Prewitt operators for CT image data at the same location at 13 hours and 25 minutes. Figures 4a to 4e By identifying the left image, we can obtain a comparison of the results of different operators. Among multiple operators, the Sobel operator (… Figure 4b ) performs best. Roberts operator ( Figure 4a ) and Laplacian operator ( Figure 4d Only the outline can be vaguely seen, and the recognition effect is very poor; Canny operator ( Figure 4c The sharp edges of particles are difficult to identify, so the Prewitt operator ( Figure 4e Although the outline and appearance of the particles can be seen, the edge recognition in fine details is not as good as that of the Sobel operator. Figure 4b The Sobel operator is clear, and overall, it is... Figure 4b It can identify more edges and has a higher coverage. Therefore, in this invention, the Sobel operator will be selected to identify the edges of bubbles and sand, and then perform subsequent numerical determination.

[0055] Specifically, in step S3 of this invention, the KMeans clustering algorithm is an unsupervised clustering algorithm, and the sample set is defined as... At that time, the samples are divided into K clusters according to the distance between them, and the clusters are divided into... The goal of the KMeans clustering algorithm is to minimize the squared difference (SSE). ,in, It is a cluster The center point, also known as the centroid of the cluster, is calculated using the following formula: Generally speaking, the smaller the SSE, the better the clustering effect.

[0056] The KMeans algorithm can be used to divide a sample set into K predefined clusters. In this invention, the elbow method is used to evaluate the clustering effect through SSE (Structured Size and Scale). The elbow method involves trying a certain number of K values ​​and observing the changes in SSE. Before reaching a certain K value, SSE changes significantly with the increase of K value. After exceeding this K value, the change in SSE will be relatively gradual. The K value at this point can be considered as the value that is closest to the true number of clusters.

[0057] Figure 5 The graph showing the relationship between SSE and the number of clusters when using the KMeans algorithm to distinguish particle (gas) edges reveals that the SSE curve decreases more slowly when K > 2. Therefore, K = 2 is the optimal value in this invention, indicating that the colors on the graph can indeed be divided into edge and non-edge categories.

[0058] Using the method described above, the color range corresponding to the edge portion can be determined to be [0, c1], where c1 is the centroid value of the smaller cluster extracted by KMeans. This range [0, c1] allows us to delineate the location of the edge on the original image. In this invention, the identified edges include both particles and gases; therefore, c1 is chosen as 3.3, representing the color value corresponding to these two substances.

[0059] Specifically, in step S3 of the present invention, the grayscale boundary values ​​corresponding to the gas and solid are determined by the moving average method, and the moving average method is used for calibration, which can improve the accuracy of bubble and solid boundary identification.

[0060] The formula for determining gray-level boundary values ​​using the moving average method is as follows:

[0061]

[0062] Where t represents time. Represents the original grayscale value, while This represents the value after smoothing using the moving average method.

[0063] Individual outliers in the calculation results can be manually adjusted using the sand volume.

[0064] like Figure 6 As shown, this figure presents a comparison of the identification of bubbles and particles using the method of this invention at different times. The first column of the figure is the original CT image, the second column is the bubble identification result, and the third column is the particle identification result. At the initial stage of hydrate formation (13h25min), Figure 6The first line of the image shows that particle recognition is generally good, but the difficulty increases for recognizing individual blurred boundaries. This is mainly because the changes in edge grayscale are more obvious in CT images, and CT images are only cross-sectional views of a location; not all particle or bubble boundaries will be clearly displayed in the image. Furthermore, comparing the particle recognition results at the same location at different times longitudinally shows that the recognition effect is very good; the particle outlines of the two recognized images are almost identical. Sometimes, differences may occur due to variations in the brightness of the CT images. For the time frame at 22h32min (…),… Figure 6 (In the second row of the image), because some of the methane gas synthesizes methane hydrate, the large bubbles will gradually decompose into small bubbles. As shown in the bubble recognition result image, at the same field of view, as small bubbles continue to emerge, more gas-solid boundaries will appear, further increasing the difficulty of recognition, making it difficult to identify even with the naked eye.

[0065] Specifically, in step S4 of this invention, after the grayscale boundary values ​​of sand and bubbles are determined, the grayscale values ​​of bubbles (grayscale value less than or equal to bubble boundary value) and sand (grayscale value greater than or equal to sand boundary value) are removed from the probability density map. The remaining grayscale values ​​(i.e., grayscale values ​​of water and hydrates) in each time period are then subjected to max-min normalization to obtain the grayscale probability density. This invention performs a linear transformation on the original data, mapping the data to the [0,1] interval. The max-min normalization method can significantly improve the performance of machine learning models.

[0066] like Figure 7 The graph showing the relative grayscale values ​​of natural gas hydrate and water at various times reveals that the probability density distribution for each time period follows a normal distribution. Over time, the peak shifts to the left, indicating the formation of new substances (methane hydrate). Figure 7 It can be seen that after 22h32min, some of the methane gas synthesized methane hydrate. Therefore, the threshold segmentation range can be read from the peak value of the curve after 23h30min in the figure. That is, the threshold segmentation range of hydrate determined in step S5 of the present invention is 0-0.18.

[0067] To analyze the errors of the machine learning method, this invention constructed a validation dataset using a method for manually identifying particles and gases. The dataset consists of 366 randomly selected images from different time periods and locations, with 236 images representing particles and 130 representing gases. Using the natural gas hydrate CT image thresholding segmentation method proposed in this invention, a binary classification machine learning task was constructed on the validation set. This task involved using machine learning to determine whether a location in an image contained gas / particles, and comparing the results with the manually labeled values. The final results are as follows: Using manually labeled values ​​as the true values ​​and machine learning as the predicted values, point-by-point comparisons were performed on each image to obtain TP, TN, FP, and FN, and the Accuracy, Precise, Recall, and F1 scores for each image were calculated. The table below shows the average performance of the proposed method on all samples in the validation set. It can be seen that the method proposed in this invention has high accuracy and reliable results.

[0068] Validation set evaluation results for methane and solid particle identification (Table)

[0069] Average TP Average TN Average FP Average FN Average Accuracy Average Precise Average Recall Average F1 gas 0.041231 0.95679 0.001907 7.12E-05 0.998021 0.949547 0.996718 0.971857477 Particles 0.595519 0.370634 0 0.033847 0.966153 1 0.945388643 0.971767174

[0070] Based on the same inventive concept, one embodiment also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the natural gas hydrate CT image thresholding method as described in the above embodiment.

[0071] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application 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.

[0072] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. 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, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0073] 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, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0074] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0075] The foregoing description of specific exemplary embodiments of the invention is for illustrative and explanatory purposes. These descriptions are not intended to limit the invention to the precise forms disclosed, and it will be apparent that many changes and variations can be made in accordance with the foregoing teachings. The exemplary embodiments were chosen and described in order to explain the specific principles of the invention and its practical application, thereby enabling those skilled in the art to implement and utilize various different exemplary embodiments of the invention, as well as various different choices and variations. The scope of the invention is intended to be defined by the claims and their equivalents.

Claims

1. A threshold segmentation method for CT images of natural gas hydrates, characterized in that, Includes the following steps: S1: Acquire images of natural gas hydrate samples obtained from synchrotron radiation CT scans, and preprocess the images to obtain images of the central region; S2: Use a convolutional neural network combined with the Sobel operator to perform edge recognition on the preprocessed central region image to obtain the edge feature matrix; S3: The KMeans clustering algorithm is used to cluster the edge feature matrix to determine the grayscale boundary values ​​between solid particles and methane bubbles; S4: Remove the grayscale components of solid particles and methane bubbles from the image based on the grayscale boundary values, and perform maximum and minimum normalization on the remaining grayscale values ​​of water and hydrates to obtain the grayscale probability density. S5: Based on the gray-level probability density of all images, draw a normalized gray-level probability density map, and determine the threshold segmentation range of hydrates based on the peak displacement of the probability density.

2. The threshold segmentation method for natural gas hydrate CT images as described in claim 1, characterized in that, In step S1, acquiring images of natural gas hydrate samples obtained by synchrotron radiation CT scanning includes setting the scanning frequency to 15 minutes per scan.

3. The threshold segmentation method for natural gas hydrate CT images as described in claim 2, characterized in that, Image preprocessing includes cropping the original 2560×2560 pixel CT image to a central region of 1280×1280 pixels.

4. The threshold segmentation method for natural gas hydrate CT images as described in claim 1, characterized in that, The convolutional neural network used in step S2 includes an input layer, a convolutional layer, and a fully connected layer; the input of the input layer is the preprocessed central region image in step S1, the convolutional layer performs convolution operations using edge recognition operators, and the fully connected layer outputs the result matrix after the convolution operation.

5. The threshold segmentation method for natural gas hydrate CT images as described in claim 4, characterized in that, In the convolutional layer, the convolution kernel is set to the Sobel operator to perform convolution operations, identify the edges of solid particles and methane bubbles, and then determine the subsequent values.

6. The threshold segmentation method for natural gas hydrate CT images as described in claim 1, characterized in that, In step S3, the KMeans algorithm divides the samples into two clusters.

7. The threshold segmentation method for natural gas hydrate CT images as described in claim 1, characterized in that, In step S3, the grayscale boundary values ​​corresponding to the gas and solid are determined by the moving average method. The moving average method is used for calibration, which can improve the accuracy of bubble and solid boundary identification.

8. The threshold segmentation method for natural gas hydrate CT images as described in claim 7, characterized in that, The formula for determining gray-level boundary values ​​using the moving average method is as follows: ， Where t represents time. Represents the original grayscale value, while This represents the value after smoothing using the moving average method.

9. The threshold segmentation method for CT images of natural gas hydrates as described in claim 8, wherein in step S5, the threshold segmentation range of the gray value interval corresponding to the hydrate is 0-0.

18.

10. A storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the natural gas hydrate CT image thresholding method according to any one of claims 1-9.

Images