CN-121998997-A - Rock core CT image crack segmentation method and device based on deep learning

CN121998997ACN 121998997 ACN121998997 ACN 121998997ACN-121998997-A

Abstract

The invention discloses a core CT image crack segmentation method and device based on deep learning, the method comprises the steps of manufacturing a training data set, expanding the number of training samples in the training data set to form a new data set, constructing a deep learning network model, wherein the deep learning network model comprises a U-Net network, a feature extraction network based on a VGG-16 network is adopted as a contraction path of feature extraction in the U-Net network, a CBMA attention module is added to an extension path sampled in the U-Net network, the CBMA attention module is used for changing the attention of the model to different spatial positions and different channels in a feature map, increasing the weight of a region where a core crack is located, training and parameter adjustment are carried out on the deep learning network model by using the training set until the model converges, testing is carried out on the deep learning network model which is completed by using a test set, and crack segmentation effect of the deep learning network model is evaluated. The invention can improve the efficiency of automatic analysis work of the rock core CT image.

Inventors

WANG MINGQIU
DUAN XINBIAO
KANG YONGGAN

Assignees

中国石油化工股份有限公司
中石化石油物探技术研究院有限公司

Dates

Publication Date: 20260508
Application Date: 20241106

Claims (10)

1. The rock core CT image crack segmentation method based on the deep learning is characterized by comprising the following steps of: The method comprises the steps of manufacturing a training data set, wherein the training data set comprises a plurality of training samples, and each training sample comprises a rock core CT slice image and an artificial crack segmentation labeling result corresponding to the rock core CT slice image; expanding the number of training samples in the training data set to form a new data set, and dividing the new data set into a training set and a testing set; Constructing a deep learning network model, wherein the deep learning network model is used for identifying crack characteristics in an input rock core CT image and obtaining crack segmentation results, the deep learning network model comprises a U-Net network, a contraction path of characteristic extraction in the U-Net network adopts a characteristic extraction network based on a VGG-16 network, a CBMA attention module is added to an up-sampling expansion path in the U-Net network, and the CBMA attention module is used for changing attention of the model to different spatial positions and different channels in a characteristic diagram and improving the weight of an area where a rock core crack is located; Training and parameter adjustment are carried out on the deep learning network model by utilizing the training set until the model converges, and training is completed; and testing the training-completed deep learning network model by using the test set, and evaluating the crack segmentation effect of the deep learning network model.
2. The depth learning based core CT image crack segmentation method as recited in claim 1, wherein the expanding the number of training samples in the training dataset comprises: And performing data enhancement operation on each core CT slice image in the training data set to acquire more training data, and simultaneously adjusting the corresponding crack segmentation result to adapt to the enhanced image data.
3. The deep learning based core CT image crack segmentation method according to claim 1, wherein the U-Net network comprises a 9-layer structure of a contraction path of feature extraction and an up-sampling propagation path, wherein the encoder section for feature extraction comprises layers 1 to 5 and the decoder section for up-sampling comprises layers 6 to 9; The contraction path of the feature extraction in the U-Net network adopts a VGG-16 network with a full connection layer and an output layer removed, the feature of the largest pooling of the last layer of the VGG-16 network is directly output to the input of the expansion path part of the U-Net network, and a convolution layer and a pooling layer with the same resolution of an output image are regarded as the same layer of operation; The up-sampling extension path in the U-Net network is that the output of the layer 4 pooling layer and the output characteristic diagram of the layer 5 after CBMA attention module and up-sampling operation are spliced according to the channel to be used as the input of the layer 6; the 6 th layer outputs the convolution result to the CBMA attention module through 2 times of convolution, and the characteristic diagram output after the up-sampling operation and the characteristic diagram output by the 3 rd layer pooling are spliced according to the channel and then are used as the input of the 7 th layer; The 7 th layer outputs the convolution result to the CBMA attention module through 2 times of convolution, and the feature map output after the up-sampling operation and the feature map output by the 2 nd layer pooling are spliced according to the channel to be used as the input of the 8 th layer; the 8 th layer outputs a convolution result to the CBMA attention module through 2 times of convolution, and the feature map output after the up-sampling operation and the feature map output by the 1 st layer pooling are spliced according to channels and then serve as the input of the 9 th layer; and the 9 th layer carries out 3 times of convolution operation on the input characteristic image and outputs a segmentation result image consistent with the resolution of the input image.
4. The deep learning based core CT image crack segmentation method according to claim 3, wherein the upsampled propagation path in the U-Net network further comprises: After the output result of the 5 th layer through the CBMA module is subjected to convolution of 1x1 convolution kernels and up-sampling by 4 times, the output result is transmitted to the input layer of the 7 th layer, and is spliced with other input images to serve as input of the 7 th layer; And (3) the output result of the 7 th layer through the CBMA module is subjected to convolution of 1x1 convolution kernels and 4 times of upsampling and then is transmitted to the input layer of the 9 th layer, and is spliced with other input images to be used as the input of the 9 th layer.
5. The deep learning based core CT image crack segmentation method of claim 1, wherein training and parametrizing the deep learning network model using the training set comprises: Training the constructed deep learning network model by using the training set, outputting a segmentation image with the same resolution as the input image by the model for each input image, obtaining crack segmentation results of all images in the training set, and completing one-time training; Comparing the obtained crack segmentation result with the manual segmentation labeling result in the training set, calculating the error of the crack segmentation result and the manual segmentation labeling result through a loss function, adjusting the parameter setting of the deep learning network model according to the calculated error, and retraining the model by utilizing each group of images in the training set; repeating the training process until the model parameters are converged, and obtaining the deep learning network model after training.
6. The depth learning based core CT image crack segmentation method according to claim 1, wherein the indices that evaluate the crack segmentation effect of the depth learning network model include a Dice coefficient, a global accuracy, and a recall.
7. The method for segmenting the core CT image cracks based on the deep learning according to claim 1, wherein, The CBMA attention module comprises a channel attention unit and a space attention unit; For a feature map, the channel attention unit constructs two feature information through maximum pooling and average pooling, wherein the maximum pooling result is a first special diagnosis information map, the average pooling result is a second feature information map, and the two feature information maps are sent to a multi-layer perceptron to generate a channel attention map; For a feature map, the spatial attention unit constructs two feature information maps through maximum pooling and average pooling along the dimension direction, and then the two feature information maps are subjected to convolution operation to obtain a two-dimensional feature map which represents attention weights which should be given to different positions in the image space; For the input feature map, the CBMA attention module firstly multiplies the input feature map by the channel attention map by bits, then multiplies the input feature map by the space attention map by bits, and outputs a result.
8. The depth learning based core CT image crack segmentation method of claim 1, further comprising: inputting a core CT image to be processed; And carrying out crack characteristic recognition on the input rock core CT image by using the deep learning network model which is subjected to training and testing, and outputting a crack segmentation result.
9. Core CT image crack segmentation device based on degree of depth study, characterized by comprising: the data input module is used for inputting a core CT image to be processed; The crack segmentation module is used for carrying out crack characteristic recognition on the input rock core CT image by utilizing the deep learning network model which completes training and testing and outputting crack segmentation results; The deep learning network model comprises a U-Net network, a feature extraction network based on a VGG-16 network is adopted as a contraction path of feature extraction in the U-Net network, a CBMA attention module is added to an extension path sampled in the U-Net network, and the CBMA attention module is used for changing attention of the model to different spatial positions and different channels in a feature map and improving weight of an area where a core crack is located.
10. An electronic device, the electronic device comprising: At least one processor, and A memory communicatively coupled to the at least one processor, wherein, The memory stores instructions executable by the at least one processor to enable the at least one processor to perform the depth learning based core CT image crack segmentation method of any one of claims 1-8.

Description

Rock core CT image crack segmentation method and device based on deep learning Technical Field The invention relates to the field of image processing in seismic exploration, in particular to a rock core CT image crack segmentation method and device based on deep learning. Background The sedimentary rock gap research has guiding significance for oil and gas resource exploration work. Industrial CT can reconstruct three-dimensional images of a core revealing the structure of cracks in the core. While industrial CT can image a core in three dimensions, the contrast between the background of the core and the cracks in the final formed image is very low and the cracks appear in multiple dimensions of the image. The efficiency of the artificial segmentation crack working under such imaging conditions cannot be well adapted to the actual production requirements. Aiming at the difficulty of manually dividing cracks, many researchers propose to automatically complete crack division in a rock core CT slice image by using a computer vision correlation technology. Conventional visual methods tend to build rule-based crack segmentation models, which accomplish the crack segmentation work by analyzing the degree of matching of various features in the image with the segmentation model. The traditional method generally comprises three steps, namely image preprocessing, denoising, crack characteristic construction, crack recognition and segmentation, wherein the characteristics are used for distinguishing crack and non-crack modes through various characteristic analysis means, and a recognition model for determining crack information in an image is given through the constructed characteristics. In recent years, with the development of deep learning techniques, the effect of image segmentation tasks has been greatly improved. A batch of deep learning methods aiming at image segmentation are emerging at home and abroad. Among them, typical image segmentation methods such as full convolution network-based image segmentation methods, encoder-decoder network-based semantic image segmentation methods, feature pyramid-based slit segmentation networks, U-Net-based filamentary target segmentation networks, and the like are used. At present, a deep learning method aiming at a rock core CT image crack segmentation method has not been applied in a large scale. Accordingly, it may be considered to perform crack segmentation on the core CT image using a crack segmentation method based on depth learning. Disclosure of Invention The invention aims to provide a core CT image crack segmentation method and device based on deep learning, which can improve the efficiency of core CT image automatic analysis work. In order to achieve the above object, in a first aspect, the present invention provides a core CT image crack segmentation method based on deep learning, including: The method comprises the steps of manufacturing a training data set, wherein the training data set comprises a plurality of training samples, and each training sample comprises a rock core CT slice image and an artificial crack segmentation labeling result corresponding to the rock core CT slice image; expanding the number of training samples in the training data set to form a new data set, and dividing the new data set into a training set and a testing set; Constructing a deep learning network model, wherein the deep learning network model is used for identifying crack characteristics in an input rock core CT image and obtaining crack segmentation results, the deep learning network model comprises a U-Net network, a contraction path of characteristic extraction in the U-Net network adopts a characteristic extraction network based on a VGG-16 network, a CBMA attention module is added to an up-sampling expansion path in the U-Net network, and the CBMA attention module is used for changing attention of the model to different spatial positions and different channels in a characteristic diagram and improving the weight of an area where a rock core crack is located; Training and parameter adjustment are carried out on the deep learning network model by utilizing the training set until the model converges, and training is completed; and testing the training-completed deep learning network model by using the test set, and evaluating the crack segmentation effect of the deep learning network model. Optionally, the expanding the number of training samples in the training dataset includes: And performing data enhancement operation on each core CT slice image in the training data set to acquire more training data, and simultaneously adjusting the corresponding crack segmentation result to adapt to the enhanced image data. Optionally, the U-Net network includes a 9-layer structure of a contracted path for feature extraction and an extended path for upsampling, wherein the encoder section for feature extraction includes layers 1 through 5 and the decoder section for upsampling includes l