Method and system for constructing CT image crack segmentation model

Abstract

The invention provides a method and a system for constructing a CT image crack segmentation model. The method comprises the following steps: acquiring a CT slice image data training set; constructing an initial rock core image crack segmentation model based on a U-Net network, a preset convolutional neural network and a preset U-Net network constructed by a CBMA attention mechanism module; inputting slice images of the CT slice image data training set into the initial rock core image crack segmentation model to output a training segmentation result; calculating a loss value according to the training segmentation result and the corresponding segmentation result, and detecting whether the loss value converges to a preset loss value; and when the loss value converges to a preset loss value, determining a CT image crack segmentation model. According to the method, the CT image crack segmentation model is constructed according to the principle that the U-Net network and the CBMA attention mechanism shrink and amplify the image texture, so that the CT image crack segmentation model outputs a more accurate segmentation result, and the working efficiency of manual crack segmentation is improved.

Description

Technical Field

[0001] The present invention relates to the field of image technology, and particularly to a method and system for constructing a CT image crack segmentation model. Background Art

[0002] The study of sedimentary rock fractures has a certain guiding significance for oil and gas resource exploration. Industrial CT can reconstruct the three-dimensional image of the core and reveal the structure of the cracks in the core. The traditional method for generating CT slice images generally includes three steps: image preprocessing, which focuses on denoising; fracture feature construction, which constructs features for distinguishing fracture and non-fracture patterns through various feature analysis means; and fracture identification and segmentation, which gives an identification model for determining the fracture information in the image through the constructed features.

[0003] Although industrial CT can perform three-dimensional imaging on the core, the contrast between the background and cracks of the core in the finally formed image is very low, and the cracks are manifested at multiple scales in the image. This makes the efficiency of manual crack segmentation unable to well meet the actual production requirements under such imaging conditions. Summary of the Invention

[0004] To solve or partially solve the problems existing in the related art, the present application provides a method and system for constructing a CT image crack segmentation model. By using the principle of shrinking and magnifying image textures in the U-Net network and the CBMA attention mechanism module to construct the CT image crack segmentation model, the accuracy of the CT image crack segmentation result output by the CT image crack segmentation model is higher, thereby promoting the improvement of the work efficiency of manual crack segmentation.

[0005] The first aspect of the present application provides a method for constructing a CT image crack segmentation model, which includes:

[0006] Collecting a training set of CT slice image data, where the training set of CT slice image data includes a first slice image and a first segmentation result;

[0007] Determining a preset U-Net network based on the U-Net network, a preset convolutional neural network, and the CBMA attention mechanism module, and constructing an initial core image crack segmentation model based on the preset U-Net network;

[0008] Inputting the first slice image into the initial core image crack segmentation model, and using the initial core image crack segmentation model to process the first slice image to output a training segmentation result;

[0009] Calculating a loss value according to the training segmentation result and the first segmentation result, and detecting whether the loss value converges to a preset loss value;

[0010] When the loss value converges to the preset loss value, the CT image crack segmentation model is determined.

[0011] In one embodiment, the collecting the CT slice image data training set includes:

[0012] Collecting a CT slice image data set;

[0013] Performing image transformation on each CT slice image in the CT slice image data set through a preset image transformation algorithm to form CT slice images in a preset form, and obtaining a CT slice image training set.

[0014] In one embodiment, after calculating the loss value according to the training segmentation result and the first segmentation result and detecting whether the loss value converges to a preset loss value, the method further includes:

[0015] When the loss value does not converge to the preset loss value, the parameter weights of the initial core image crack segmentation model are updated according to the loss value, and the first slice data is input into the initial core image crack segmentation model with updated parameter weights again.

[0016] In one embodiment, after determining the CT image crack segmentation model, the method further includes:

[0017] Collecting a CT slice image data test set, where the CT slice image data test set includes a second slice image and a second segmentation result;

[0018] Inputting the second slice image into the CT image crack segmentation model, processing the second slice image by using the CT image crack segmentation model, and outputting a test segmentation result;

[0019] Calculating the similarity between the test segmentation result and the second segmentation result;

[0020] Detecting whether the similarity is within a preset similarity range;

[0021] When the similarity is within the preset similarity range, it is determined that the CT image crack segmentation model meets the preset accuracy.

[0022] In one embodiment, calculating the loss value according to the training segmentation result and the first segmentation result includes:

[0023] Inputting the training segmentation result and the first segmentation result into a preset loss function, and calculating the loss value by using the preset loss function.

[0024] In one embodiment, the loss function of the initial core image crack segmentation model is:

[0025]

[0026] Among them, P i is the training segmentation result output by the CT image crack segmentation model at index i, and v i is the segmentation result corresponding to the slice image in the CT slice image data training set.

[0027] In one embodiment, calculating the similarity between the test segmentation result and the second segmentation result includes:

[0028] Calculating the similarity between the test segmentation result and the second segmentation result by using a preset Dice coefficient.

[0029] The second aspect of the present application provides a system for constructing a CT image crack segmentation model, which includes:

[0030] A first acquisition module, configured to acquire a CT slice image data training set, where the CT slice image data training set includes a first slice image and a first segmentation result;

[0031] A construction module, configured to determine a preset U-Net network based on a U-Net network, a preset convolutional neural network, and a CBMA attention mechanism module, and construct an initial core image crack segmentation model based on the preset U-Net network;

[0032] A processing module, configured to input the first slice image into the initial core image crack segmentation model, process the first slice image by using the initial core image crack segmentation model, and output a training segmentation result;

[0033] A first detection module, configured to calculate a loss value according to the training segmentation result and the first segmentation result, and detect whether the loss value converges to a preset loss value;

[0034] A first determination module, configured to determine the CT image crack segmentation model when the loss value converges to the preset loss value.

[0035] The third aspect of the present application provides a computer device, including a memory and a processor, where the memory stores a computer program, and is characterized in that when the processor executes the computer program, the steps of the method for constructing a CT image crack segmentation model in any of the above embodiments are implemented.

[0036] The fourth aspect of the present application provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method for constructing a CT image crack segmentation model in any of the above embodiments are implemented.

[0037] The technical solution provided by this application may include the following beneficial effects:

[0038] Collect a training set of CT slice image data, where the training set of CT slice image data includes a first slice image and a first segmentation result; determine a preset U-Net network based on the U-Net network, a preset convolutional neural network, and a CBMA attention mechanism module, and construct an initial core image crack segmentation model based on the preset U-Net network; input the first slice image into the initial core image crack segmentation model, and use the initial core image crack segmentation model to process the first slice image to output a training segmentation result; calculate a loss value according to the training segmentation result and the first segmentation result, and detect whether the loss value converges to a preset loss value; when the loss value converges to the preset loss value, then determine the CT image crack segmentation model. By using the principle of shrinking and magnifying image textures in the U-Net network and the CBMA attention mechanism module to construct the CT image crack segmentation model, the accuracy of the CT image crack segmentation result output by the CT image crack segmentation model is higher, thereby promoting the improvement of the work efficiency of manual crack segmentation.

[0039] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit this application. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] By describing the exemplary embodiments of this application in more detail in conjunction with the drawings, the above and other objects, features, and advantages of this application will become more obvious. Among them, in the exemplary embodiments of this application, the same reference numerals generally represent the same components.

[0041] Figure 1 It is a schematic flowchart of a method for constructing a CT image crack segmentation model in an embodiment;

[0042] Figure 2 It is a schematic structural diagram of a system for constructing a CT image crack segmentation model in an embodiment;

[0043] Figure 3 It is a schematic internal structure diagram of a computer device in an embodiment;

[0044] Figure 4 It is a schematic diagram of the prior art principle of the CBMA attention mechanism module in an embodiment

[0045] Figure 5 It is a schematic diagram of the principle of this application of the CBMA attention mechanism module in an embodiment;

[0046] Figure 6 It is a schematic diagram of the principle of the VGG-16 convolutional neural network in an embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0047] The embodiment of the present application provides a method and system for constructing a CT image crack segmentation model. By using the principle of shrinking and enlarging image texture in the U-Net network and the CBMA attention mechanism module to construct the CT image crack segmentation model, the accuracy of the CT image crack segmentation result output by the CT image crack segmentation model is higher, thereby promoting the improvement of the work efficiency of manual crack segmentation.

[0048] In order to make the purpose, technical solution and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0049] Embodiment 1

[0050] In view of the difficulty of manual crack segmentation, many researchers have proposed to use computer vision-related technologies to automatically complete crack segmentation in core CT slice images. Traditional vision methods tend to construct rule-based crack segmentation models and complete crack segmentation by analyzing the matching degree between various features in the image and the segmentation model. With the development of deep learning technology, the effect of image segmentation tasks has been greatly improved. A batch of deep learning methods for image segmentation have emerged. Among them, relatively typical ones are image segmentation methods based on fully convolutional networks, semantic image segmentation methods based on encoder-decoder networks, crack segmentation networks based on feature pyramids, and filamentous object segmentation networks based on U-Net. At present, deep learning methods for core CT image crack segmentation methods have not been widely applied. Therefore, it is possible to consider using a deep learning-based crack segmentation method to segment cracks in core CT images.

[0051] Please refer to Figure 1 、 Figure 4 、 Figure 5 and Figure 6 For an embodiment of the method for constructing a CT image crack segmentation model in the embodiment of the present application, it includes:

[0052] Step 10: Collect a training set of CT slice image data, where the training set of CT slice image data includes a first slice image and a first segmentation result;

[0053] In this embodiment, the collected CT slice image data is core CT slice image data. Specifically:

[0054] 1. Prepare core samples: The core samples need to go through a series of processes, including cleaning, drying, cutting, etc., in order to perform CT scans.

[0055] 2. CT Scanning: Place the prepared core samples in a CT scanner and perform high-resolution scanning to obtain the original data. The scanning parameters need to be adjusted according to the experimental requirements, which may include the scanning range, pixel size, scanning layer thickness, etc.

[0056] 3. Data Acquisition: After the CT scanning is completed, process the original data obtained from the scanning through CT image processing software, including operations such as denoising and image enhancement, to obtain high-quality core CT slice image data.

[0057] 4. Data Saving: Save the obtained core CT slice image data for subsequent analysis and processing.

[0058] 5. Use the obtained core CT slice image data as the training set of CT slice image data, that is, complete the acquisition of the training set of CT slice image data.

[0059] After completing the acquisition of the training set of CT slice image data, in order to provide a data basis for the subsequent step 102, it is also necessary to divide the acquired training set of CT slice image data into slice images and corresponding segmentation results.

[0060] For example: The training set of CT slice image data includes x 1 (y 1 , z 1 ), x 2 (y 2 , z 2 ), and x 3 (y 3 , z 3 ). Among them, x 1 is the first group of CT slice image training data, x 2 is the second group of CT slice image training data, x 3 is the third group of CT slice image training data. And y 1 is the slice image of the first group of CT slice image training data, z 1 is the segmentation result of the first group of CT slice image training data. Similarly, the slice images in the training set of CT slice image data can be expressed as: y i , where i ∈ [0, n], n represents that there are n groups of CT slice image training data in the training set of CT slice image data, and y i represents the slice image of the i-th group of CT slice image training data. Similarly, it can also be known that the segmentation images in the training set of CT slice image data can be expressed as: z i , where i ∈ [0, n], n represents that there are n groups of CT slice image training data in the training set of CT slice image data, and z i represents the segmentation result of the i-th group of CT slice image training data.

[0061] Therefore, the second slice image in this embodiment includes y 1 , y 2 and y 3 ; the second segmentation result includes z 1 , z 2 and z 3 .

[0062] In order to make the segmentation result output by the finally constructed CT image crack segmentation model more accurate, in this embodiment, the segmentation result of the CT slice image data training set can be obtained by manual segmentation.

[0063] In this embodiment, the collected core CT slice image data can be stored in the local database; the collected core CT slice image data can also be stored in the cloud, and no specific limitation is made here.

[0064] Step 11: Determine a preset U-Net network based on the U-Net network, the preset convolutional neural network, and the CBMA attention mechanism module, and construct an initial core image crack segmentation model based on the preset U-Net network;

[0065] To improve the usage effect of the U-Net network, the CBMA attention mechanism module is introduced to reduce the model's attention to the background in the slice image, which is beneficial for the model to obtain a more accurate segmentation result. In the decoder part of the model, skip connections are used to enhance the influence of deep features on the shallow network and improve the utilization rate of deep features by the model.

[0066] Before constructing the CT image crack segmentation model, it is necessary to first construct an initial core image crack segmentation model. In this embodiment, a preset U-Net network can be determined based on the preset convolutional neural network and the CBMA attention mechanism module.

[0067] The U-Net network is also a U-Net feature extraction network. The contraction path of U-Net feature extraction uses the VGG-16 convolutional neural network. As Figure 6 shown, the VGG-16 convolutional neural network includes an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer. Among them, the input layer is used to input data, the convolutional layer is used to perform convolutional processing on the data, the max pooling layer is used to extract the corresponding features of the data after convolutional processing, and the fully connected layer is used to classify or regress the features extracted by the convolutional layer and the pooling layer.

[0068] As Figure 4As shown in the figure, the initial U-Net network includes the contraction path and the expansion path of the U-Net network. The contraction path of the U-Net network consists of the first convolutional layer, the second convolutional layer, the third convolutional layer, the fourth convolutional layer, and the fifth pooling layer. The fifth pooling layer directly sends the image data processed by the previous four convolutional layers into the expansion path of the U-Net network.

[0069] Construct a preset U-Net network. Specifically:

[0070] 1. The feature extraction part of the preset U-Net network. In this embodiment, the VGG-16 model is used, and the fully connected layer and the output layer in the VGG-16 model are removed, so that this part of the network focuses on the feature extraction task. The VGG-16 model performs 5 maximum pooling operations on the image, corresponding to 5 groups of convolutional operations. The convolutional kernels of these 5 groups of convolutional operations are all 3x3 in size, with a stride of 1, and the boundary padding method is same to keep the size of the result after convolution consistent with the input. The number of convolutional kernels for the 5 groups of convolutions is 64, 128, 256, 512, 512.

[0071] The result of each group of convolutional operations needs to go through a maximum pooling layer. The pooling size is 2*2, so that the size of the feature map after pooling is half of the input. Due to 5 pooling operations, there are 6 scales of images in the network.

[0072] That is, the fully connected layer and the output layer in the VGG-16 convolutional neural network are removed and used as the contraction path of the U-Net feature extraction network. That is, the feature of the last maximum pooling is directly output to the input of the expansion path of the U-Net feature extraction network. The convolutional layer and the pooling layer with the same output image resolution are regarded as the same layer operation. Then the feature extraction part includes 5 layers.

[0073] 2. The expansion path part of the preset U-Net network. In the present invention, convolutional operations corresponding to each layer in the contraction path part are used. The size and number of convolutional kernels remain the same and the original size of the image is restored by gradually upsampling. The upsampling method used in the present invention is the transposed convolution method of 2*2. The last group of convolutional operations will input a result map with a dimension of 1, because the dimension of the input core CT slice image is also 1.

[0074] Introduce the attention mechanism module CBMA. The attention of CBMA is divided into channel attention and spatial attention. Channel attention aims to judge the image dimension that is most beneficial to crack segmentation. The spatial attention mechanism is used to enhance the model's attention to the crack and its nearby areas and increase the weight of the eigenvalue of the crack area.

[0075] The preset U-Net network in this embodiment retains the original skip connection mechanism in the U-Net.

[0076] As Figure 5 shown, for the upsampling expansion path in the expansion path part of the preset U-Net network, specifically: concatenate the output of the 4th pooling layer and the output feature map after the 5th layer passes through the CBMA attention mechanism module and the upsampling operation by channels as the input of the 6th layer. The 6th layer performs 2 convolutions, outputs the convolution result to the CBMA attention mechanism module, and after concatenating the feature map output after the upsampling operation and the feature map of the 3rd pooling layer by channels, it is used as the input of the 7th layer. The 7th layer performs 2 convolutions, outputs the convolution result to the CBMA attention mechanism module, and after concatenating the feature map output after the upsampling operation and the feature map of the 2nd pooling layer by channels, it is used as the input of the 8th layer. The 8th layer performs 2 convolutions, outputs the convolution result to the CBMA attention mechanism module, and after concatenating the feature map output after the upsampling operation and the feature map of the 1st pooling layer by channels, it is used as the input of the 9th layer. The 9th layer performs 3 convolution operations on the input feature map and outputs a segmentation result map with the same resolution as the input image.

[0077] To make the contraction path of the U-Net use high-order features more effectively, a part of dense connections is added to the upsampling expansion path. Specifically, the output result of the 5th layer passing through the CBMA module is convolved with a 1x1 convolution kernel and then upsampled by 4 times, and then sent to the input layer of the 7th layer, and concatenated with other input images as the input of the 7th layer. The output result of the 7th layer passing through the CBMA module is convolved with a 1x1 convolution kernel and then upsampled by 4 times, and then sent to the input layer of the 9th layer, and concatenated with other input images as the input of the 9th layer.

[0078] 3. After the above two steps, the U-Net network built based on the U-Net network, the preset convolutional neural network, and the CBMA attention mechanism module is determined as the preset U-Net network.

[0079] 4. Build an initial core image crack segmentation model based on the determined preset U-Net network.

[0080] In this embodiment, the CBMA attention mechanism module includes a channel attention mechanism and a spatial attention mechanism.

[0081] Specifically, for a feature map, the channel attention mechanism constructs two feature informations through max pooling and average pooling. The max pooling result is called the first feature information map, and the average pooling result is called the second feature information map. The two feature information maps are sent into a multi-layer perceptron to generate a channel attention map.

[0082] For a feature map, the spatial attention mechanism constructs two feature information maps through max pooling and average pooling along the dimension direction. Subsequently, the two feature information maps undergo a convolution operation to obtain a two-dimensional feature map, representing the attention weights that should be assigned to different positions in the image space.

[0083] For the input feature map of the CBMA attention mechanism module, first multiply the input feature map by the channel attention map bit by bit, and then multiply it by the spatial attention map bit by bit to output the result.

[0084] In this embodiment, upsampling can be achieved by transposed convolution or by interpolation, and no specific limitation is made here.

[0085] In this embodiment, the preset convolutional neural network can be a VGG-16 convolutional neural network or a VGG-19 convolutional neural network, and no specific limitation is made here.

[0086] Step 12: Input the first slice image into the initial core image crack segmentation model, and use the initial core image crack segmentation model to process the first slice image to output a training segmentation result.

[0087] After the initial core image crack segmentation model is constructed, input the first slice image in the CT slice image data training set collected in step 10 into the initial core image crack segmentation model, so that the initial core image crack segmentation model outputs a training segmentation result with the same resolution as the first slice image, and obtain the training segmentation result, that is, complete one training.

[0088] For example, when the CT slice image data training set includes x 1 (y 1 , z 1 ), x 2 (y 2 , z 2 ), and x 3 (y 3 , z 3 ), input y 1 , y 2 and y 3 into the initial core image crack segmentation model in sequence. The initial core image crack segmentation model processes y 1 , y 2 and y 3 through the preset U-Net network in sequence, so that the initial core image crack segmentation model outputs the first training segmentation result corresponding to y 1 , the second training segmentation result corresponding to y 2 and the third training segmentation result corresponding to y 3 in sequence. When the y 1The corresponding first training segmentation result, y 2 The corresponding second training segmentation result and y 3 When the corresponding third training segmentation result is obtained, the sequential training is completed.

[0089] Step 13: Calculate the loss value based on the training segmentation result and the first segmentation result, and detect whether the loss value converges to a preset loss value;

[0090] When the initial core image crack segmentation model outputs a training segmentation result, each time a training segmentation result is obtained, it is compared with the first segmentation result corresponding to the first slice image (in this embodiment, the first segmentation result can be a manual segmentation result).

[0091] For example: When y 1 the corresponding first training segmentation result is obtained, y 1 the corresponding first segmentation result z 1 is obtained, and y 1 the corresponding first training segmentation result is compared with the first segmentation result z 1 to obtain the loss value between y 1 the corresponding first training segmentation result and the first segmentation result z 1

[0092] Detect whether the loss value obtained by the comparison converges to a preset loss value, that is, determine whether the loss value tends to a preset loss value. The closer the loss value is to the preset loss value, the more steps 14 are executed.

[0093] For example, when the preset loss value is set to 1, if the loss value is in the interval range of (0.5, 1), it means that the loss value tends to the preset loss value; if the loss value is not in the interval range of (0.5, 1), it means that the loss value does not tend to the preset loss value.

[0094] In this embodiment, the preset loss value can be set to 1, or can be set to 0, or can be set to any specific value, or can be set to any specific value interval, and no specific limitation is made here.

[0095] Step 14: When the loss value converges to the preset loss value, the CT image crack segmentation model is determined.

[0096] When the loss value tends to the preset loss value, it means that the loss value converges to the preset loss value; and whether the loss value converges to the preset loss value indicates whether the initial CT image crack segmentation model tends to be stable. When the initial CT image crack segmentation model is stable, the stable initial CT image crack segmentation model is determined as the CT image crack segmentation model.

[0097] ​Due to factors such as the system function of the receiver and the distribution law of the crack itself during the CT reconstruction process, there are limitations in the CT slice images of the core, such as low image contrast, low signal-to-noise ratio, and irregular crack distribution. To solve these problems, a crack segmentation model based on the U-Net network is constructed. In the expansion path of the U-Net model, a CBMA attention mechanism module is introduced to change the model's attention to different spatial positions and different channels in the feature map, and a skip connection mechanism is introduced into the expansion path to improve the network's crack segmentation ability. The data set after data augmentation is divided into a training set and a test set. The training set is used to complete the parameter tuning and training of the model to obtain a model with better performance on the training set. Finally, the test set is imported into the trained model to complete the test of the model's performance, and the final crack segmentation model for the core CT image is obtained.

[0098] In this application, by constructing a model based on deep learning, the efficiency of crack segmentation in core CT slice images is improved, and the obtained model maintains a high accuracy rate.

[0099] The model based on the U-Net network used in this application reduces the demand for the number of training images to a certain extent, reduces the difficulty of constructing the core CT data set, and improves the operability of the method.

[0100] In this application, the attention mechanism is introduced into the construction of the U-Net network, which improves the model's resolution ability for specific channels and specific regions. The proportion of pixels occupied by cracks in the core CT slice images is relatively low. The introduction of the CBMA attention mechanism improves the processing efficiency and accuracy of the model to a certain extent.

[0101] In this embodiment, a CT slice image data training set is collected. The CT slice image data training set includes a first slice image and a first segmentation result. A preset U-Net network is determined based on the U-Net network, a preset convolutional neural network, and the CBMA attention mechanism module, and an initial core image crack segmentation model is constructed based on the preset U-Net network. The first slice image is input into the initial core image crack segmentation model, and the initial core image crack segmentation model is used to process the first slice image to output a training segmentation result. The loss value is calculated according to the training segmentation result and the first segmentation result, and it is detected whether the loss value converges to a preset loss value. When the loss value converges to the preset loss value, the CT image crack segmentation model is determined. By using the principle of shrinking and magnifying image textures by the U-Net network and the CBMA attention mechanism module to construct the CT image crack segmentation model, the accuracy rate of the CT image crack segmentation result output by the CT image crack segmentation model is higher, thereby promoting the improvement of the work efficiency of manual crack segmentation.

[0102] Embodiment 2

[0103] Another embodiment of the method for constructing a CT image crack segmentation model in this embodiment includes:

[0104] Step 20: Collect a training set of CT slice image data, where the training set of CT slice image data includes a first slice image and a first segmentation result;

[0105] Since a corresponding training set of CT slice image data is required to construct a CT image crack segmentation model, a training set of CT slice image data needs to be collected. And in order to make the finally constructed CT image crack segmentation model more accurate, a large amount of training set of CT slice image data is required.

[0106] Then, in this embodiment, collecting the training set of CT slice image data may include the following steps 201 - 202.

[0107] Step 201: Collect a CT slice image data set;

[0108] In this embodiment, a CT scan method is also used to scan the core sample to obtain a CT slice image data set. Different from the above embodiment where the core sample is prepared first, then the core sample is CT scanned to obtain the original data, and finally the original data is denoised and image enhanced, and the original data after denoising and image enhancement is used as the CT slice image data set.

[0109] In this embodiment, a CT slice image data set can be obtained by scanning the core sample through a CT scan method; it can also be obtained from a local database; or it can be obtained from the cloud. There is no specific limitation here.

[0110] Step 202: Perform image transformation on each CT slice image in the CT slice image data set through a preset image transformation algorithm to form a CT slice image in a preset form, and obtain a training set of CT slice images.

[0111] In this embodiment, the CT slice images in the preset form include rotated CT slice images, translated CT slice images, scaled CT slice images, and mirrored slice images. And the preset image transformation algorithm includes a rotation algorithm, a translation algorithm, a scaling algorithm, and a mirroring algorithm.

[0112] The rotation algorithm is to rotate a CT slice image. Specifically:

[0113] 1. Determine the center coordinates of the CT slice image as the center position of the rotated image.

[0114] 2. Traverse each pixel of the CT slice image, and map the coordinates of each pixel to the rotated image according to the rotation direction to obtain the rotated image.

[0115] 3. Detect whether the sizes of the CT slice image and the rotated image are the same; when the sizes of the CT slice image and the rotated image are different, adjust the sizes of the rotated image and the CT slice image to be the same.

[0116] The translation algorithm is to translate a CT slice image. Specifically:

[0117] 1. Determine the translation distance and translate the center coordinates of the CT slice image by the translation distance.

[0118] 2. Traverse each pixel of the CT slice image and translate the coordinates of each pixel by the translation distance to obtain the translated image.

[0119] 3. Detect whether the sizes of the CT slice image and the translated image are the same; when the sizes of the CT slice image and the translated image are different, adjust the sizes of the translated image and the CT slice image to be the same.

[0120] The scaling algorithm is to scale a CT slice image. Specifically:

[0121] 1. Select a scaling algorithm, such as nearest neighbor interpolation, bilinear interpolation, or bicubic interpolation.

[0122] 2. Determine the size of the scaled image according to the scaling ratio.

[0123] 3. Use the selected scaling algorithm to map each pixel of the CT slice image to the corresponding position in the scaled image. This involves rotating the pixels of the original image according to the scaling ratio and angle.

[0124] 4. Insert the data of the original image into the corresponding pixel positions of the new image.

[0125] 5. Repeat step 4 until all pixels of the original image are processed.

[0126] After the scaling is completed, the new image will contain an image with the same size as the target image but different resolution.

[0127] In this embodiment, the scaling algorithm can be nearest neighbor interpolation; it can also be bilinear interpolation, or it can be bicubic interpolation, which is not specifically limited here.

[0128] The mirroring algorithm is to perform mirror processing on a CT slice image. Specifically:

[0129] 1. Determine the center coordinates of the CT slice image and set the center coordinates as the center position of the mirrored image.

[0130] 2. Traverse each pixel of the CT slice image, draw a symmetry axis with the center position, and perform pixel conversion on the coordinates of each pixel and the coordinates symmetric to each other with the symmetry axis to obtain the mirrored image.

[0131] 3. Detect whether the sizes of the CT slice image and the mirror image are the same; when the sizes of the CT slice image and the mirror image are different, adjust the size of the mirror image to be the same as that of the CT slice image.

[0132] Perform image transformation on each CT slice image in the CT slice image dataset through a preset image transformation algorithm to form CT slice images in a preset form (corresponding rotated CT slice images, translated CT slice images, scaled CT slice images, and mirrored slice images). Finally, form an image set with the original CT slice images corresponding to the rotated CT slice images, translated CT slice images, scaled CT slice images, and mirrored slice images, and set this image set as the CT slice image training set.

[0133] Use a variety of data augmentation methods to construct a training dataset for CT slice images in each form, expanding the diversity of image data.

[0134] Step 21: Determine a preset U-Net network based on the U-Net network, a preset convolutional neural network, and the CBMA attention mechanism module, and construct an initial core image crack segmentation model based on the preset U-Net network.

[0135] Step 22: Input the first slice image into the initial core image crack segmentation model, and use the initial core image crack segmentation model to process the first slice image to output a training segmentation result.

[0136] In this embodiment, steps 21 - 22 are similar to steps 12 - 13 in the above embodiment. To avoid repetition, they will not be elaborated here.

[0137] Step 23: Calculate the loss value according to the training segmentation result and the first segmentation result, and detect whether the loss value converges to a preset loss value; when it is detected that the loss value does not converge to the preset loss value, execute step 24; when the loss value converges to the preset loss value, execute step 25.

[0138] In this embodiment, step 23 is similar to step 13 in the above embodiment. To avoid repetition, it will not be elaborated here.

[0139] In this embodiment, for the specific calculation method of the loss value in step 23, refer to step 231.

[0140] Step 231: Input the training segmentation result and the first segmentation result into a preset loss function, and use the preset loss function to calculate the loss value.

[0141] In this embodiment, the preset loss function is:

[0142]

[0143] Among them, p i is the training segmentation result output by the CT image crack segmentation model at index i, and v i is the corresponding segmentation result of the slice image in the CT slice image data training set.

[0144] Since the training segmentation result p i output by the CT image crack segmentation model may contain decimal values, this value needs to be binarized and then imported into the loss function for calculation.

[0145] Step 24: When the loss value does not converge to the preset loss value, update the parameter weights of the initial core image crack segmentation model according to the loss value, and input the first slice data into the initial core image crack segmentation model with updated parameter weights again, and execute Step 22.

[0146] When the loss value does not converge to the preset loss value, it means that the initial CT image crack segmentation model has not yet tended to be stable. Then, it is necessary to continue training the initial CT image crack segmentation model. Before continuing to train the initial CT image crack segmentation model, update the parameter weights in the initial CT image crack segmentation model according to the loss value, so that the loss value between the training segmentation result output by the initial CT image crack segmentation model and the first segmentation result converges more to the preset loss value.

[0147] Step 25: When the loss value converges to the preset loss value, determine the CT image crack segmentation model.

[0148] In this embodiment, Step 25 is similar to Step 14 in the above embodiment. To avoid repetition, it will not be elaborated here.

[0149] Step 26: Collect the CT slice image data test set, which includes the second slice image and the second segmentation result;

[0150] In order to verify the accuracy of the determined CT image crack segmentation model, it is necessary to test it using the CT slice image data test set. Therefore, it is also necessary to collect the CT slice image data test set, which includes the second slice image and the second segmentation result.

[0151] In this embodiment, the CT slice image data test set may also include T 1 (A 1 , B 1 ), T 2 (A 2 , B 2 ), and T 3 (A 3 , B 3 ). Among them, T 1is the test data of the first group of CT slice images, T 2 is the test data of the second group of CT slice images, T 3 is the test data of the upper group of CT slice images. And A 1 is the slice image of the training data of the first group of CT slice images, B 1 is the segmentation result of the training data of the first group of CT slice images. Similarly, the slice images in the CT slice image data test set can be expressed as: A i , where i ∈ [0, n], n represents that there are n groups of CT slice image test data in the CT slice image data training set, and A i represents the slice image of the i-th group of CT slice image test data. Similarly, it can also be known that the segmentation images in the CT slice image data test set can be expressed as: B i , where i ∈ [0, n], n represents that there are n groups of CT slice image test data in the CT slice image data training set, and B i represents the segmentation result of the i-th group of CT slice image test data.

[0152] Therefore, the second slice image in this embodiment includes A 1 , A 2 and A 3 ; the second segmentation result includes B 1 , B 2 and B 3 .

[0153] In this embodiment, the CT slice image data test set can be obtained from the local database, or from the cloud, or by scanning the core with a scanning device to obtain the CT slice image data test set, which is not specifically limited here.

[0154] Step 27: Input the second slice image into the CT image crack segmentation model, and use the CT image crack segmentation model to process the second slice image to output the test segmentation result;

[0155] Input the second slice image into the CT image crack segmentation model in sequence. The CT image crack segmentation model processes the second slice image using the preset U-Net network to output the test segmentation result by the CT image crack segmentation model.

[0156] For example, when the CT slice image data test set can also include T 1 (A 1 , B 1 ), T 2 (A 2 , B 2 ) and T 3 (A 3 , B3 ) When, input A 1 into the CT image crack segmentation model, and the CT image crack segmentation model processes A through a preset U-Net network 1 to output the corresponding test segmentation result. Similarly, it can be known that the test segmentation result includes A 1 the corresponding test segmentation result, A 2 the corresponding test segmentation result, and A 3 the corresponding test segmentation result.

[0157] Step 28: Calculate the similarity between the test segmentation result and the second segmentation result;

[0158] In order to evaluate the actual use effect of the CT image crack segmentation model, it is necessary to calculate the similarity between the test segmentation result and the second segmentation result.

[0159] In this embodiment, for the specific method of calculating the similarity between the test segmentation result and the second segmentation result in step 28, refer to step 281.

[0160] Step 281: Calculate the similarity between the test segmentation result and the second segmentation result by using a preset Dice coefficient.

[0161] The Dice coefficient is a method for calculating the similarity between two sets, mainly used to compare the similarity between samples, and is usually used to compare similarities.

[0162] In this embodiment, the calculation process of the preset Dice coefficient is as follows: specifically:

[0163] 1. Determine that both the test segmentation result and the second segmentation result are sets.

[0164] 2. Calculate the Dice coefficient. The formula for the Dice coefficient is: Dice coefficient = 2 * the size of the intersection of the two sets / (the size of set A + the size of set B).

[0165] In actual calculation, first find the intersection of the two sets, that is, the elements that are in both A and B. Then calculate the size of this intersection. Furthermore, calculate the sizes of the two sets A and B respectively. Finally, substitute these values into the formula of the Dice coefficient to obtain the Dice coefficient.

[0166] In this embodiment, three evaluation indicators, namely the Dice coefficient, global accuracy, and recall rate, are used to evaluate the model effect.

[0167] Step 29: Detect whether the similarity is within the preset similarity range;

[0168] It is determined whether the test segmentation result is the same as the second segmentation result to a certain extent by detecting whether the similarity is within a preset similarity range. In the calculation, the similarity range is set to [0, 1], and the larger the value, the more similar the two sets are. In this embodiment, the preset similarity is set to [0.9, 1]. When the similarity value falls within the preset similarity range, it indicates that the test segmentation result is the same as the second segmentation result to a certain extent.

[0169] In this embodiment, the preset Dice coefficient value range is set to [0, 1], and the larger the value, the more similar the two sets are; it can also be that the smaller the value, the more similar the two sets are, and no specific limitation is made here.

[0170] Step 210: When the similarity is within the preset similarity range, it is determined that the CT image crack segmentation model meets the preset accuracy.

[0171] When the similarity is within the preset similarity range, it indicates that the similarity between the test segmentation result and the second segmentation result is relatively high to a certain extent. At this time, the output of the CT image crack segmentation model is also more accurate, that is, the CT image crack segmentation model meets the preset accuracy.

[0172] Step 211: When the similarity is not within the preset similarity range, the CT slice image data training set is collected again, and the CT image crack segmentation model is trained.

[0173] When the similarity is not within the preset similarity range, it indicates that the similarity between the test segmentation result and the second segmentation result is relatively low, which means that the segmentation result output by the CT image crack segmentation model at this time is not very accurate. Then the CT slice image data training set is collected again, and the CT image crack segmentation model is trained.

[0174] Based on the improved U-Net network and CBMA attention mechanism, this embodiment proposes a method for constructing a CT image crack segmentation model. The number of the training set is expanded through data augmentation, and the CBMA mechanism is introduced to reduce the influence of a large number of non-crack regions in the training set on model training. In addition to the original skip connection of the U-Net model, the skip connection mechanism is introduced into the extended path of the U-Net network, improving the network's utilization ability for deep models. The model of the present invention improves the recognition accuracy of cracks in core CT images, which is of great help for subsequent image automatic analysis tasks.

[0175] It should be understood that although steps 20 to 21 are merely one way of constructing the initial core image crack segmentation model, these steps do not necessarily have to be executed in the order indicated by the steps. Unless otherwise clearly stated in this article, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover, steps 20 to 21 do not necessarily have to be completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages does not necessarily have to be sequential, but can be executed alternately or in rotation with at least a part of other steps or sub-steps or stages of other steps.

[0176] Embodiment III

[0177] In this embodiment, as Figure 2 shown, a system for constructing a CT image crack segmentation model is provided, including:

[0178] A first acquisition module 100, configured to acquire a CT slice image data training set, where the CT slice image data training set includes a first slice image and a first segmentation result;

[0179] A construction module 101, configured to determine a preset U-Net network based on a U-Net network, a preset convolutional neural network, and a CBMA attention mechanism module, and construct an initial core image crack segmentation model based on the preset U-Net network;

[0180] A processing module 102, configured to input the first slice image into the initial core image crack segmentation model, process the first slice image by using the initial core image crack segmentation model, and output a training segmentation result;

[0181] A first detection module 103, configured to calculate a loss value according to the training segmentation result and the first segmentation result, and detect whether the loss value converges to a preset loss value;

[0182] A first determination module 104, configured to determine the CT image crack segmentation model when the loss value converges to the preset loss value.

[0183] In this embodiment, the first acquisition module 100 acquires a CT slice image data training set, where the CT slice image data training set includes a first slice image and a first segmentation result, and sends the first slice image of the CT slice image data training set to the processing module 102; sends the first segmentation result of the CT slice image data training set to the first detection module 103.

[0184] The construction module 101 determines a preset U-Net network based on the U-Net network, a preset convolutional neural network, and the CBMA attention mechanism module, constructs an initial core image crack segmentation model based on the preset U-Net network, and finally sends the initial core image crack segmentation model to the processing module 102.

[0185] The processing module 102 inputs the first slice image into the initial core image crack segmentation model, processes the first slice image using the initial core image crack segmentation model, outputs a training segmentation result, and sends the training segmentation result to the first detection module 103. The first detection module 103 calculates a loss value based on the training segmentation result and the first segmentation result, and detects whether the loss value converges to a preset loss value. When the loss value converges to the preset loss value, the first determination module 104 determines the CT image crack segmentation model. By using the principles of image texture reduction and magnification of the U-Net network and the CBMA attention mechanism to construct the CT image crack segmentation model, the CT image crack segmentation model outputs a more accurate segmentation result, promoting the improvement of the work efficiency of manual crack segmentation.

[0186] In one embodiment, the first acquisition module 100 may include an acquisition unit and a conversion unit.

[0187] The acquisition unit is used to acquire a CT slice image dataset;

[0188] The conversion unit is used to perform image transformation on each CT slice image in the CT slice image dataset through a preset image transformation algorithm to form a CT slice image in a preset form, obtaining a CT slice image training set.

[0189] In one embodiment, the system for constructing the CT image crack segmentation model further includes: a second acquisition module, a test module, a calculation module, a second detection module, a second determination module,

[0190] The processing module 102 is further used to, when the loss value does not converge to the preset loss value, update the parameter weights of the initial core image crack segmentation model according to the loss value, and input the first slice data into the initial core image crack segmentation model with updated parameter weights again.

[0191] The second acquisition module is used to acquire a CT slice image data test set, and the CT slice image data test set includes a second slice image and a second segmentation result;

[0192] The test module is used to input the second slice image into the CT image crack segmentation model, process the second slice image using the CT image crack segmentation model, and output a test segmentation result;

[0193] A calculation module, configured to calculate the similarity between the test segmentation result and the second segmentation result;

[0194] A second detection module, configured to detect whether the similarity is within a preset similarity range;

[0195] A second determination module, configured to determine that the CT image crack segmentation model meets the preset accuracy when the similarity is within the preset similarity range.

[0196] The first detection module 103 is further configured to input the training segmentation result and the first segmentation result into a preset loss function, and calculate a loss value by using the preset loss function, where the preset loss function is:

[0197]

[0198] where p i is the training segmentation result output by the CT image crack segmentation model at index i, and v i is the segmentation result corresponding to the slice image in the CT slice image data training set.

[0199] The calculation module is further configured to calculate the similarity between the test segmentation result and the second segmentation result by using a preset Dice coefficient.

[0200] For the specific limitations on the system for constructing the CT image crack segmentation model, reference may be made to the limitations on the method for constructing the CT image crack segmentation model in the foregoing text, which will not be elaborated herein. Each unit in the above system for constructing the CT image crack segmentation model can be implemented in whole or in part by software, hardware, and their combination. The above units can be embedded in or independent of the processor in the computer device in the form of hardware, or stored in the memory of the computer device in the form of software, so as to facilitate the processor to call and execute the operations corresponding to the above units.

[0201] Example 4

[0202] In this embodiment, a computer device is provided. Its internal structure diagram can be as Figure 3As shown in the figure. The computer device includes a processor, a memory, a network interface, a display screen, and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and computer programs, and a database is deployed on the non-volatile storage medium. The database is used for the relevant data required for the method of constructing a CT image crack segmentation model, and the model constructed by the method of constructing a CT image crack segmentation model. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface of the computer device is used to communicate with other computer devices on which application software is deployed. When the computer program is executed by the processor, it realizes a method of constructing a CT image crack segmentation model. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the housing of the computer device, or an external keyboard, a touchpad, or a mouse, etc.

[0203] Those skilled in the art can understand that Figure 3 the structure shown in the figure is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0204] In one embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program. When the processor executes the computer program, it implements the steps of the method of constructing a CT image crack segmentation model described in any of the above embodiments.

[0205] Embodiment Five

[0206] In this embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by the processor, the following steps are implemented:

[0207] Collect a CT slice image data training set, where the CT slice image data training set includes a first slice image and a first segmentation result;

[0208] Based on the U-Net network, a preset convolutional neural network, and the CBMA attention mechanism module, determine a preset U-Net network, and construct an initial core image crack segmentation model based on the preset U-Net network;

[0209] Input the first slice image into the initial core image crack segmentation model, and use the initial core image crack segmentation model to process the first slice image to output a training segmentation result;

[0210] Calculate a loss value based on the training segmentation result and the first segmentation result, and detect whether the loss value converges to a preset loss value;

[0211] When the loss value converges to the preset loss value, determine the CT image crack segmentation model.

[0212] In one embodiment, when the computer program is executed by a processor, it can also implement the steps of the method for constructing a CT image crack segmentation model described in any of the above embodiments.

[0213] When the computer program is executed by a processor, it can also implement the following steps:

[0214] Collect a CT slice image data set;

[0215] Perform image transformation on each CT slice image in the CT slice image data set through a preset image transformation algorithm to form a CT slice image in a preset form, and obtain a CT slice image training set.

[0216] When the loss value does not converge to the preset loss value, update the parameter weights of the initial core image crack segmentation model according to the loss value, and input the first slice data into the initial core image crack segmentation model with updated parameter weights again.

[0217] Collect a CT slice image data test set, where the CT slice image data test set includes a second slice image and a second segmentation result;

[0218] Input the second slice image into the CT image crack segmentation model, and use the CT image crack segmentation model to process the second slice image to output a test segmentation result;

[0219] Calculate the similarity between the test segmentation result and the second segmentation result;

[0220] Detect whether the similarity is within a preset similarity range;

[0221] When the similarity is within the preset similarity range, determine that the CT image crack segmentation model meets the preset accuracy.

[0222] Input the training segmentation result and the first segmentation result into a preset loss function, and use the preset loss function to calculate a loss value.

[0223] The preset loss function is:

[0224]

[0225] Among them, p i is the training segmentation result output by the CT image crack segmentation model at index i, and v i is the segmentation result corresponding to the slice image in the CT slice image data training set.

[0226] Calculate the similarity between the test segmentation result and the second segmentation result by using the preset Dice coefficient.

[0227] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium used in the embodiments provided in the present application can include non-volatile and / or volatile memories. Non-volatile memories can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memories can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0228] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0229] The above-described embodiments only represent several implementation manners of the present application. Their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A method for constructing a CT image crack segmentation model, characterized in that, it includes: Collect a training set of CT slice image data, where the training set of CT slice image data includes a first slice image and a first segmentation result; Determine a preset U-Net network based on the U-Net network, a preset convolutional neural network, and a CBMA attention mechanism module, and construct an initial core image crack segmentation model based on the preset U-Net network; Input the first slice image into the initial core image crack segmentation model, and use the initial core image crack segmentation model to process the first slice image and output a training segmentation result; Calculate a loss value according to the training segmentation result and the first segmentation result, and detect whether the loss value converges to a preset loss value; When the loss value converges to the preset loss value, then determine the CT image crack segmentation model.

2. The method according to claim 1, characterized in that, the collection of the training set of CT slice image data includes: Collect a CT slice image data set; Perform image transformation on each CT slice image in the CT slice image data set through a preset image transformation algorithm to form a CT slice image in a preset form, and obtain a training set of CT slice images.

3. The method according to claim 1, characterized in that, after calculating the loss value according to the training segmentation result and the first segmentation result and detecting whether the loss value converges to a preset loss value, the method further includes: When the loss value does not converge to the preset loss value, then update the parameter weights of the initial core image crack segmentation model according to the loss value, and input the first slice data into the initial core image crack segmentation model with updated parameter weights again.

4. The method according to any one of claims 1-3, characterized in that, after determining the CT image crack segmentation model, the method further includes: Collect a test set of CT slice image data, where the test set of CT slice image data includes a second slice image and a second segmentation result; Input the second slice image into the CT image crack segmentation model, and use the CT image crack segmentation model to process the second slice image and output a test segmentation result; Calculate the similarity between the test segmentation result and the second segmentation result; Detect whether the similarity is within a preset similarity range; When the similarity is within the preset similarity range, then determine that the CT image crack segmentation model meets the preset accuracy.

5. The method according to claim 1, characterized in that, the calculation of the loss value according to the training segmentation result and the first segmentation result includes: Input the training segmentation result and the first segmentation result into a preset loss function, and use the preset loss function to calculate the loss value.

6. The method according to claim 5, characterized in that, the preset loss function is: where p i is the training segmentation result output by the CT image crack segmentation model at index i, and v i is the segmentation result corresponding to the slice image in the CT slice image data training set.

7. The method according to claim 4, characterized in that, the calculation of the similarity between the test segmentation result and the second segmentation result includes: Calculate the similarity between the test segmentation result and the second segmentation result using a preset Dice coefficient.

8. A system for constructing a crack segmentation model of a CT image, characterized in that it includes: A first acquisition module for acquiring a CT slice image data training set, where the CT slice image data training set includes a first slice image and a first segmentation result; A construction module for determining a preset U-Net network based on a U-Net network, a preset convolutional neural network, and a CBMA attention mechanism module, and constructing an initial core image crack segmentation model based on the preset U-Net network; A processing module for inputting the first slice image into the initial core image crack segmentation model, processing the first slice image using the initial core image crack segmentation model, and outputting a training segmentation result; A first detection module for calculating a loss value based on the training segmentation result and the first segmentation result, and detecting whether the loss value converges to a preset loss value; A first determination module for determining a CT image crack segmentation model when the loss value converges to the preset loss value.

9. A computer device, including a memory and a processor, where the memory stores a computer program, characterized in that when the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, on which a computer program is stored, characterized in that when the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images