Basic substance identification method of rock stratum image based on deep learning, program product and equipment

Abstract

The invention discloses a deep learning-based rock stratum image base material identification method, program product and equipment, and the method comprises the steps: obtaining a to-be-detected rock stratum image which comprises rock stratum images under all energy channels; based on the rock stratum image to be measured, a rock stratum input image of a trained base material recognition model is formed, the base material recognition model comprises a decoder and a base material prediction network, and each decoding layer in the decoder comprises an attention layer and an up-sampling layer; based on the rock stratum input image, acquiring a rock stratum multi-attention feature map output by an attention layer; on the basis of the rock stratum input image, obtaining up-sampling rock stratum features output by an up-sampling layer; fusing the rock stratum multi-attention feature map and the up-sampling rock stratum features to obtain rock stratum fusion features corresponding to the decoding layer; and based on the rock stratum fusion features corresponding to the decoding layer, predicting base material data in the rock stratum image to be detected through a base material prediction network.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to a rock layer image base substance identification method based on deep learning, a computer program product and a computing device. BACKGROUND

[0002] With the extension of unconventional oil and gas exploration and development to deep reservoirs, and the increasing demand for accurate early warning in coal mine safety monitoring, micro-scale characterization technology has become a key means to crack the characteristics of coal and rock. Coal and rock, as a complex porous medium, its internal structure is composed of organic matter, various mineral substances (quartz, kaolinite, calcite, pyrite, etc.) and pores and cracks of different shapes. The micro characteristics of this multi-component system directly determine the physical and chemical properties of coal and rock, which has a significant impact on coalbed methane development and mine safety utilization. In the field of coalbed methane development, pore connectivity and tortuosity control gas adsorption / desorption efficiency and permeability. In the field of mine safety, the dissolution and recrystallization of minerals caused by water-rock interaction in micro-pores can weaken the mechanical properties of coal and rock, significantly increasing the risk of coal and gas outburst.

[0003] In order to study the internal structure of coal and rock, the existing solutions include traditional CT scanning technology, deep learning model, threshold segmentation, etc. to characterize the internal composition of coal and rock. However, traditional CT data is only used for simple material decomposition, without deeply mining the energy-material attenuation correlation characteristics, making it difficult to distinguish the composition of complex materials; the deep learning model of the conventional convolutional neural network has a fixed receptive field, which is difficult to capture micro-pore or macro-pore features; the threshold segmentation relies on manual threshold selection, and only a rough division of pores and minerals leads to a deviation in the results. The above methods are difficult to solve the material distribution rule and pore evolution mechanism of coal and rock at the micro-scale, and when dividing the minerals, the entire coal and rock body is divided into a kind of mineral substance, without dividing it according to the actual mineral composition. Therefore, a more accurate coal and rock mineral component identification and higher precision pore extraction method is needed. SUMMARY

[0004] In order to solve the existing technical problems, the present application provides a rock layer image base substance identification method based on deep learning, a computer program product and a computing device, which can enhance attention features in pore-dense areas, thus retaining details, enhance up-sampling features in homogeneous rock layer areas, and maintain continuity, thereby improving the accuracy of pore identification.

[0005] In a first aspect, a rock layer image base substance identification method based on deep learning is provided, comprising: acquiring a rock layer image to be measured, the rock layer image to be measured comprising rock layer images under each energy channel;

[0006] Form an input image of the rock stratum to a trained base material recognition model based on the to-be-tested rock stratum image, wherein the base material recognition model comprises a decoder and a base material prediction network, and each decoding layer in the decoder comprises an attention layer and an up-sampling layer;

[0007] Obtain a rock stratum multi-attention feature map output by the attention layer based on the input image of the rock stratum;

[0008] Obtain an up-sampled rock stratum feature output by the up-sampling layer based on the input image of the rock stratum;

[0009] Fuse the rock stratum multi-attention feature map and the up-sampled rock stratum feature to obtain a rock stratum fusion feature corresponding to the decoding layer;

[0010] Predict the base material data in the to-be-tested rock stratum image through the base material prediction network based on the rock stratum fusion feature corresponding to the decoding layer.

[0011] In a second aspect, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the base material recognition method for a rock stratum image based on deep learning according to any of the embodiments of the present application.

[0012] In a third aspect, a computing device is provided, comprising a memory and a processor, wherein the memory stores a computer program, and the computer program, when executed by the processor, causes the processor to perform the base material recognition method for a rock stratum image based on deep learning according to any of the embodiments of the present application.

[0013] In the present application, the attention layer is configured in the decoder, the rock stratum multi-attention feature map is obtained through the attention layer, the rock stratum multi-attention feature map can provide accurate position information of the base material, the up-sampled rock stratum feature is obtained through the up-sampling layer in the decoder, and the up-sampled rock stratum feature can provide high-level semantic understanding. Through fusion of the two, the attention feature can be enhanced in the dense pore area, details are retained, the up-sampled feature is enhanced in the homogeneous rock stratum area, and continuity is maintained, so as to improve the accuracy of pore recognition. BRIEF DESCRIPTION OF DRAWINGS

[0014] Figure 1 An application environment diagram of the base material recognition method for a rock stratum image based on deep learning in an embodiment;

[0015] Figure 2 A flowchart of the base material recognition method for a rock stratum image based on deep learning in an embodiment;

[0016] Figure 3 A network structure diagram of the base material recognition model in an embodiment;

[0017] Figure 4A schematic diagram of a network structure of an attention layer in an embodiment;

[0018] Figure 5 A schematic diagram of a network structure of a feature enhancement network in an embodiment;

[0019] Figure 6 A schematic diagram of training a base material recognition model in an embodiment;

[0020] Figure 7 A schematic diagram of a network structure of training a base material recognition model in an embodiment;

[0021] Figure 8 A schematic diagram of a base material recognition device based on a rock stratum image in an embodiment;

[0022] Figure 9 A schematic diagram of a computing device in an embodiment. DETAILED DESCRIPTION

[0023] The technical solutions of the present application are further described in detail below in combination with the accompanying drawings and specific embodiments.

[0024] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in the description of the application herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the present application. As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items.

[0025] In the following description, expressions relating to "some embodiments" describe a subset of all possible embodiments, but it should be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict.

[0026] Reference is made to Figure 1Fig. 1 is an application environment diagram of a rock stratum image-based base substance identification method based on deep learning according to an embodiment. The rock stratum image-based base substance identification method based on deep learning is applied to a computing device 10 that can acquire projection data of a rock stratum under each energy channel. The projection data is acquired in the following manner: X-ray tube voltages of 45, 55, and 65 KVp are set respectively, and corresponding tube current parameters are set. Sequential scanning is performed under different tube voltage settings to acquire X-ray multi-spectral attenuation data. The acquired data is preprocessed, and a blind source separation algorithm of non-negative matrix factorization (NMF) is used to solve the preprocessed multi-group data. By setting appropriate iteration numbers and convergence thresholds, the separation effect is optimized, and narrow-spectrum projection is acquired. That is, projection data of the rock stratum under multiple energy channels, such as [35, 45) KeV, [45, 55) KeV, and [55, 65) KeV, is obtained. The reconstructed algorithm is used to reconstruct the projection data of each energy channel after blind separation to generate the rock stratum image to be measured under each different energy channel. During the reconstruction process, the reconstruction algorithm parameters are adjusted according to the imaging target requirements to improve the decomposition image quality and the detail resolution and reduce the subsequent experimental error. The computing device 10 identifies various base substances in the rock stratum image to be measured by using a trained base substance identification model. The base substances include but are not limited to pores and various mineral components, and the mineral components can be CaCO3, SiO2, and the like. The rock stratum image-based base substance identification method based on deep learning provided in the embodiments of the present application can be applied to various rock stratum environments, such as coal rock strata.

[0027] Referring to Figure 2 Fig. 2 is a flowchart of a rock stratum image-based base substance identification method based on deep learning provided in an embodiment of the present application. The rock stratum image-based base substance identification method based on deep learning is applied to a computing device, and the rock stratum image-based base substance identification method based on deep learning includes the following steps:

[0028] S11, acquiring a rock stratum image to be measured, the rock stratum image to be measured including rock stratum images under each energy channel.

[0029] In the embodiment, the step of reconstructing the projection data under each energy channel to obtain the rock stratum image to be measured can be performed in the computing device or in other devices and then imported into the computing device.

[0030] S12, forming a rock stratum input image to a trained base substance identification model based on the rock stratum image to be measured.

[0031] In the embodiment, the base substance identification model is trained based on a training data set, as shown in Figure 3As shown, the base material recognition model includes an encoder, a decoder and a base material prediction network, the encoder includes a feature extraction network and a feature enhancement network, wherein the feature extraction network includes encoding layers at different scales, and each decoding layer in the decoder includes an attention layer and an up-sampling layer. The encoder is used to extract different scale features of the to-be-tested rock layer image through the encoding layers at different scales. The feature enhancement network can be a depth separable dilated convolution pyramid structure, which is used to expand the receptive field of the base material features extracted by the encoding layer without losing spatial resolution or increasing the amount of convolution kernel calculation, which is crucial for obtaining the global distribution, connectivity and distinguishing true and false pores of different size pores. In this embodiment, the to-be-tested rock layer image can be used as a rock layer input image, or an image processed from the to-be-tested rock layer image can be used as a rock layer input image.

[0032] In this embodiment, the feature extraction network part adopts an Xception module for feature extraction. The Xception module mainly includes a normal convolution layer, a separable convolution layer, a BN layer, a ReLU activation function and a maximum pooling layer. The feature map output by the Xception feature extraction network is introduced into the feature enhancement network to enhance the semantic information of the feature map.

[0033] S13, based on the rock layer input image, obtaining the rock layer multi-attention feature map output by the attention layer.

[0034] In this embodiment, the encoding features at different direction positions are extracted through the attention layer, which can improve the expression ability of the position sensitive features of the base material. As shown in Figure 3 As shown, the feature data output by the encoding layer at the same scale layer is taken as the input of the attention layer at the same scale layer.

[0035] S14, based on the rock layer input image, obtaining the up-sampling rock layer feature output by the up-sampling layer.

[0036] In this embodiment, as shown in Figure 3 As shown, the output of the previous decoding layer of the up-sampling layer is taken as the input of the up-sampling layer. For example, the input of the second scale up-sampling layer is the output of the third scale decoding layer. The up-sampling rock layer feature is output by the up-sampling layer at each scale layer.

[0037] S15, fusing the rock layer multi-attention feature map and the up-sampling rock layer feature to obtain the rock layer fusion feature corresponding to the decoding layer.

[0038] In the embodiment, after obtaining the stratum multi-attention feature map under the same scale layer and the up-sampled stratum feature, the two are fused to obtain a stratum fusion feature. The stratum multi-attention feature map can provide accurate position information of the base material, and the up-sampled stratum feature can provide high-level semantic understanding. Through the fusion of the two, the attention feature can be enhanced in the dense pore area, thus retaining details, and the up-sampled feature can be enhanced in the homogeneous stratum area, thus maintaining continuity.

[0039] S16, based on the stratum fusion feature corresponding to the decoding layer, predicting base material data in the to-be-measured stratum image through a base material prediction network.

[0040] In the embodiment, when the decoding layer is not the last decoding layer connected with the base material prediction network, for example, is not the first scale decoding layer, the stratum fusion feature corresponding to the decoding layer is taken as the input of the up-sampling layer of the next layer until the stratum fusion feature output by the last decoding layer is obtained as the input of the base material prediction network, and the base material data in the to-be-measured stratum image is output. The base material data includes but is not limited to pore data, mineral composition data, and the like.

[0041] In the above embodiment, by configuring an attention layer in the decoder, a stratum multi-attention feature map is obtained through the attention layer, which can provide accurate position information of the base material; an up-sampled stratum feature is obtained through the up-sampling layer in the decoder, which can provide high-level semantic understanding. Through the fusion of the two, the attention feature can be enhanced in the dense pore area, thus retaining details, and the up-sampled feature can be enhanced in the homogeneous stratum area, thus maintaining continuity, thereby improving the accuracy of pore recognition.

[0042] In some embodiments, the base material recognition model includes an encoder including a plurality of encoding layers, the encoding layers being configured to output a stratum feature map including a feature map under each energy channel, and the attention layer includes a channel attention layer and a coordinate attention layer. The obtaining of the stratum multi-attention feature map output by the attention layer based on the stratum input image includes:

[0043] obtaining a stratum feature map belonging to the same scale layer as the attention layer as the input of the attention layer, and outputting a channel attention feature under each energy channel through the channel attention layer;

[0044] taking the channel attention feature under the energy channel as the input of the coordinate attention layer, respectively outputting horizontal pooling features and vertical pooling features through the pooling layer in the coordinate attention layer, and splicing the horizontal pooling features and the vertical pooling features under the energy channel to obtain a stratum pooling splicing feature under the energy channel;

[0045] Based on the rock stratum pooling splicing feature, a rock stratum multi-attention feature map under the energy channel is output through a dynamic convolution kernel in the coordinate attention layer, and the rock stratum multi-attention feature map is obtained based on the rock stratum multi-attention feature map under each energy channel.

[0046] In the embodiment, introducing the coordinate channel attention can provide more accurate position coding of the pore features. Based on the characteristics of the channel attention mechanism and the coordinate attention mechanism, a dynamic kernel prediction module is added, compared with the traditional attention module, the local and global feature information captured by the dynamic kernel prediction module can obtain more and more accurate feature view across channels, so that the model can more accurately locate and identify the target of the microstructure. As shown in Figure 4 Figure 4 is a network structure schematic diagram of the attention layer in an embodiment. The attention layer includes a channel attention layer and a coordinate attention layer. The channel attention features are respectively input into a horizontal average pooling layer and a vertical average pooling layer, and the horizontal pooling features are output through the horizontal average pooling layer, and the vertical pooling features are output through the vertical average pooling layer. The dynamic convolution kernel generator generates adaptive convolution kernels according to the rock stratum pooling splicing features, so that larger convolution kernels are generated for large pores to capture the overall morphology, and slender convolution kernels are generated for microcracks to enhance the linear features. As can be seen from Figure 4 , the rock stratum feature maps are respectively output into a maximum pooling layer and an average pooling layer for processing, the output of the maximum pooling layer and the output of the average pooling layer are respectively input into a shared MLP, and the data output by the shared MLP is respectively input into a first activation layer and a second activation layer for processing. The output of the first activation layer and the output of the second activation layer are spliced to obtain the channel attention features.

[0047] Optionally, the encoder includes a feature extraction network and a feature enhancement network, and the rock stratum feature map belonging to the same scale layer as the attention layer includes:

[0048] Based on the rock stratum input image, the rock stratum feature map corresponding to each encoding layer is output through the feature extraction network, and the rock stratum feature map includes the feature map under each energy channel;

[0049] The rock stratum feature map output by the encoding layer connected with the feature enhancement network is taken as the input of the feature enhancement network, and the rock stratum enhancement feature is output through the feature enhancement network.

[0050] For example, as shown in Figure 3 , three encoding layers of different scales are configured, and the first scale encoding layer, the second scale encoding layer and the third scale encoding layer respectively output the corresponding rock stratum feature maps. For example, the rock stratum images under three energy channels are input, and the rock stratum feature maps also include the feature maps under three energy channels. ​

[0051] Optionally, the step of outputting rock layer enhancement features through the feature enhancement network includes:

[0052] The inputs of each energy channel in the feature enhancement network are convolved using dilated convolution with different dilation ratios to obtain dilated convolution features for each energy channel.

[0053] Based on the void convolution features under each energy channel, point convolution processing is performed to obtain the rock layer enhancement features.

[0054] In this embodiment, the feature enhancement network depth includes a separable dilated convolution pyramid composed of two parts: depthwise atrous convolution and pointwise convolution, as shown in the diagram below. Figure 5 As shown, dilated convolution expands the receptive field of feature extraction without sacrificing spatial resolution or increasing the computational cost of the convolution kernel. This is crucial for obtaining the global distribution, connectivity, and differentiation of true and false pores of different sizes. Adding this module improves the extraction of micropore features while maintaining the model's computational efficiency. This module performs dilated convolution on the rock layer feature map under each input energy channel using different dilation rates. The convolution results are then concatenated and followed by a 1×1 convolution to obtain the final feature map. Point convolution is a 1×1 convolution, unifying the pore representation under different energy channels, while simultaneously performing channel fusion on the feature map output by dilated convolution. Different dilation rates are used for the rock layer feature maps under different energy channels to calculate the dilated convolution features for each energy channel. This ensures that micropores correspond to small dilation rates, preserving more details of micropores and preventing feature omissions due to uniform small dilation rates; and that macropores correspond to large dilation rates, achieving global perception and avoiding computational waste due to uniform large dilation rates.

[0055] In some embodiments, Figure 6 This is a schematic diagram of training a base material recognition model in one embodiment; the method further includes:

[0056] S61. Obtain the actual sample dataset.

[0057] In this embodiment, each actual sample in the actual sample dataset includes a sample rock layer image corresponding to the sample rock layer. The sample rock layer image includes sample images under each energy channel, and the sample images are reconstructed based on the projection data of the corresponding energy channel. The actual samples are reconstructed based on projection data collected from actual rock layers.

[0058] The X-ray tube voltage of the CT configuration is three different energies, for example, 45, 55, and 65KVp, and the corresponding tube current parameters are sequentially scanned under different tube voltage settings. The X-ray multispectral attenuation data of the actual sample is obtained by using the sequence scanning mode. The collected data is preprocessed, and the blind source separation algorithm of non-negative matrix factorization (NMF) is used to solve the preprocessed multiple sets of data. By setting appropriate iteration times and convergence thresholds, the separation effect is optimized, and the narrow energy spectrum projection is obtained, that is, the projection data of the actual rock formation under the energy channel of [35, 45) KeV, [45, 55) KeV, and [55, 65) KeV. The reconstruction algorithm is used to reconstruct the projection data under each energy channel after blind separation, and the sample image under each energy channel is generated. In the reconstruction process, the reconstruction algorithm parameters are adjusted according to the imaging target requirements, the decomposition image quality and the detail resolution are improved, and the subsequent experimental error is reduced.

[0059] S62, obtain a simulation sample data set.

[0060] In this embodiment, the simulation sample data set is generated based on a three-dimensional physical model representing the rock formation and the base material. Each simulation sample in the simulation sample data set includes a simulation rock formation image and base material label data in the simulation rock formation image. The base material label data includes, but is not limited to, pore label data, various mineral label data, and the like.

[0061] The determination of the rock formation components is to measure the rock formation using an X-ray diffractometer to obtain the mineral composition (such as CaCO3, SiO2) of the rock formation sample. The material and non-material are divided to determine the location of the pores in the rock formation. A three-dimensional physical model is built according to the composition of the rock formation and the pores to construct the relationship between the attenuation coefficient of the material and the energy channel. The rock formation can be a coal rock formation, a shale formation, and the like. For example, the simulation sample data set can be obtained by simulation software based on a three-dimensional physical model.

[0062] S63, in the first iteration process, based on the simulation sample data set, training the base material recognition model to obtain a pre-trained base material recognition model.

[0063] In this embodiment, the base material recognition model is first trained based on the base material label data, which can enable the base material recognition model to learn the rock formation features corresponding to the base material label to accelerate the convergence of the base material recognition model. In the first iteration process, after a preset number of iterations, the first iteration is completed. In the first iteration process, the base material prediction loss value of each iteration is calculated, the gradient of the parameter adjustment is calculated by backward propagation based on the base material prediction loss value, and the parameters in the base material recognition model are adjusted according to the gradient.

[0064] S64. In the second round of iteration, based on the actual sample dataset and the simulation sample dataset, calculate the total loss value for each iteration in the second round of iteration, and train the base material recognition model based on the total loss value.

[0065] In this embodiment, as Figure 7 As shown, the network structures of the encoder and decoder are similar to... Figure 3 The network structures shown are identical, except one is a pre-trained network and the other is an untrained network. The training network for the base material recognition model includes the encoder, the decoder connected to the encoder, an adversarial training network, and a total loss calculation network. The adversarial training network is used to learn the differences in base material features between the actual sample dataset and the simulated sample dataset through adversarial training. The adversarial training network includes a domain adversarial discriminator. The encoder includes a feature extraction network and a feature enhancement network. The feature extraction network outputs a rock layer sample feature map corresponding to the input sample image, and the feature enhancement network outputs enhanced rock layer sample features corresponding to the input sample image based on the rock layer sample feature map.

[0066] During network training, the actual sample dataset is an unlabeled sample set. To reduce the deviation between real and simulated data, which leads to decreased cross-domain performance and low accuracy, an adversarial training network (e.g., a feature-level adaptive domain adversarial discriminator (DANN)) is added after the feature enhancement network output. This optimizes the network's task loss (mineral classification, pore extraction) and domain adversarial loss in parallel. Simultaneously, data augmentation (image rotation, Gaussian noise addition) is applied to both the simulated and actual sample datasets, achieving domain invariance learning of the feature space between the simulated and actual sample datasets of the rock strata. First, the domain discriminator is frozen, and the network is pre-trained on the simulated sample dataset, i.e., the first iteration is performed. Then, the actual sample dataset is added for adversarial fine-tuning. This involves fixing the feature extractor and training the domain adversarial discriminator. The domain adversarial discriminator is fixed, and a gradient reversal layer (GRL) is used to multiply the gradient by a negative coefficient during backpropagation, making the optimization objectives of the feature extractor and the domain adversarial discriminator opposite. This optimizes the feature extractor and dynamically adjusts the parameters.

[0067] In this embodiment, in the second round of iteration process, a plurality of losses are calculated, including but not limited to discriminant loss, base material prediction loss value. In this embodiment, the discriminant loss represents the core supervision signal in domain adversarial training that drives the domain discriminator to distinguish the features to the actual sample data set or the simulation sample data set, and its essence is binary cross entropy loss. The role of the discriminant loss is: optimizing the domain discriminator: by minimizing the loss, forcing the discriminator to accurately identify that the features come from the simulation sample data set (such as laboratory simulation rock formation) or the actual sample data set (such as real downhole data); guide the feature extractor: through the gradient reversal layer (GRL), the loss gradient is taken as negative after the backward propagation, which drives the feature extractor to generate domain invariant features (such as stripping equipment noise, retaining consistent pore structure across domains), thereby narrowing the distribution difference between simulation and real data, and improving the generalization ability of the model in the real scene. The base material prediction loss is the core supervision signal in dual-energy CT rock formation analysis, which functions to constrain the neural network to accurately decompose the material composition (such as minerals, pore fluid) in the CT image. The base material prediction loss calculates the pixel-level difference (commonly using L1 / L2 loss) between the predicted base material density map (such as calcite, quartz, pore) and the real material distribution, which drives the model to learn the energy absorption characteristics of the material to X-ray, thereby accurately quantifying the mineral composition and porosity of the rock formation, and providing key physical parameters for reservoir evaluation.

[0068] Optionally, the calculating, in the second round of iteration process, a total loss value of each iteration in the second round of iteration process based on the actual sample data set and the simulation sample data set comprises:

[0069] inputting the rock sample feature map into the decoder and the adversarial training network as input respectively;

[0070] outputting, by the domain adversarial discriminator, a discriminant result corresponding to the rock sample feature map based on the rock sample feature map, and calculating a discriminant loss under the current iteration based on the discriminant result;

[0071] outputting, by the decoder, a base material prediction result corresponding to the rock sample feature map based on the rock sample feature map, and calculating a base material prediction loss value under the current iteration based on the base material prediction result and a base material label data corresponding to the rock sample feature map;

[0072] calculating, by the total loss calculation network, a total loss value under the current iteration based on the discriminant loss and the base material prediction loss value.

[0073] Optionally, the discriminant loss comprises a base material sample domain loss and a base material feature confusion loss, and the outputting, by the domain adversarial discriminator, a discriminant result corresponding to the rock sample feature map based on the rock sample feature map, and calculating a discriminant loss under the current iteration based on the discriminant result comprises:

[0074] outputting, by the domain adversarial discriminator, a first probability that the rock sample feature map belongs to the simulation sample dataset and a second probability that the rock sample feature map belongs to the actual sample dataset based on the rock sample feature map;

[0075] calculating the base substance sample domain loss based on the first probability and the second probability;

[0076] calculating the base substance feature confusion loss based on the second probability;

[0077] weighting the base substance sample domain loss and the base substance feature confusion loss to obtain the discrimination loss.

[0078] In the embodiment, the base substance prediction result includes multiple base substance images decomposed, such as a pore image, a CaCO3 image, and a SiO2 image. The total loss value L TOTAL = λ1×L TASK + λ2×L DANN , wherein λ1 and λ2 are weight parameters, L TASK is a base substance prediction loss value, and L DANN represents a discrimination loss. The calculation formula of the base substance prediction loss value is as follows:

[0079] L TASK = L MSE + αL CE

[0080]

[0081] , wherein L MSE represents a base substance classification loss, T k represents a predicted class of the kth decomposed image, T′ k represents a base substance label data corresponding to the kth decomposed image, and K represents the total number of decomposed images. α represents a coefficient. LCE represents a pixel loss, and the calculation formula is as follows:

[0082]

[0083] T k,j represents a pixel value of the jth pixel in the kth decomposed image in the base substance label data, J represents the total number of pixels in the kth decomposed image, and T′ k,j represents a predicted pixel value of the jth pixel in the kth decomposed image.

[0084] The calculation formula of the discrimination loss is as follows:

[0085] L DANN = L disc + βL adv

[0086] L disc denotes the base substance sample domain loss, L adv denotes the base substance feature confusion loss, and β denotes the adversarial loss weight. The calculation formula of the base substance sample domain loss is as follows:

[0087]

[0088] wherein n s denotes the sample quantity of the input simulation sample in the simulation sample data set, n T denotes the total sample quantity of the input actual sample in the actual sample data set. denotes the i th sample in the input simulation sample, denotes the i th sample in the input actual sample. denotes the discrimination probability that the sample belongs to the actual sample, denotes the discrimination probability that the sample belongs to the actual sample.

[0089] In some embodiments, after predicting the base substance data in the to-be-tested rock stratum image, i.e., each base substance image, topological analysis of the pore image, porosity calculation, and pore size analysis are performed, which is of great significance to the research and practice in the fields of energy, materials, geology, and the like, and plays a key role in multiple actual scenarios. Connectivity analysis is performed on the decomposed pores, and the obtained pore image is converted into a binary image, wherein the pore region is assigned a value of 1, and the matrix region is assigned a value of 0. A connected domain labeling algorithm is used to label the pore region in the binary image.

[0090] Porosity Φ is an important indicator for measuring the development degree of coal rock pores, and is defined as the ratio of pore volume to total coal rock volume. In two-dimensional image analysis, porosity can be calculated by pixel statistics, and the formula is as follows:

[0091]

[0092] wherein N p is the pixel quantity of the pore region, and N total is the total pixel quantity of the image.

[0093] One or more embodiments of the present application have the following beneficial effects:

[0094] (1) The present application uses conventional micro-focus CT to obtain multi-energy spectrum projection data, and through multiple tube voltages, X-ray multi-spectrum CT imaging based on multi-energy projection blind separation can calculate and separate the multi-energy projection in a blind scene where the X-ray energy spectrum is unknown, thereby obtaining narrow energy spectrum projections of multiple energy channels under different voltages.

[0095] (2) The present application uses a deep separable dilated convolution pyramid network model to decompose each component material of a multi-component coal rock CT image. Advanced deep learning methods such as deep separable dilated convolution and improved attention mechanism are used to extract image features, expand the receptive field, and obtain more and more accurate feature views, thereby improving the accuracy and precision of the base material and micro-pore decomposition. At the same time, during the data training and testing process, the feature level domain discriminator and the adversarial loss are introduced, so that the model is continuously migrated and optimized to conform to the real data, and the model can be better generalized to the actual data set.

[0096] (3) Topological analysis and porosity calculation are performed on the extracted pores to realize the research and practical application of pores in coal rocks in the fields of energy, materials, geology, etc.

[0097] (4) The present application can realize the leap from "fuzzy recognition" to "precise quantification" of coal rock mineral and pore micro-characterization. This technology has significant academic value and industrial application potential in the research of some complex porous media in the field of spectral CT material decomposition.

[0098] In another aspect, the present application provides a computer program product comprising a computer program which, when executed by a processor, implements the base material identification method of the rock image based on deep learning according to any one of the embodiments of the present application.

[0099] In the computer program product, the optional implementation form of the program module architecture of the computer program implementing each step of the target identification method can be a base material identification device for rock images based on deep learning.

[0100] Please refer to Figure 8 An embodiment of the present application provides a base material identification device for rock images based on deep learning, comprising: an acquisition module 80 configured to acquire a rock image to be measured, the rock image to be measured comprising rock images under each energy channel; a prediction module 81 configured to form a rock input image to a trained base material identification model based on the rock image to be measured, wherein the base material identification model comprises a decoder and a base material prediction network, each decoding layer in the decoder comprising an attention layer and an up-sampling layer; the prediction module 81 is further configured to acquire a rock multi-attention feature map output by the attention layer based on the rock input image; the prediction module 81 is further configured to acquire an up-sampled rock feature output by the up-sampling layer based on the rock input image; the prediction module 81 is further configured to fuse the rock multi-attention feature map and the up-sampled rock feature to obtain a rock fusion feature corresponding to the decoding layer; and the prediction module 81 is further configured to predict base material data in the rock image to be measured based on the rock fusion feature corresponding to the decoding layer through the base material prediction network.

[0101] Optionally, the base material recognition model comprises an encoder, the encoder comprises a plurality of encoding layers, the encoding layers are configured to output a rock formation feature map, the rock formation feature map comprises feature maps under respective energy channels, the attention layer comprises a channel attention layer and a coordinate attention layer, and the prediction module 81 is further configured to:

[0102] obtain a rock formation feature map belonging to a same scale layer as the attention layer as input of the attention layer, output channel attention features under respective energy channels through the channel attention layer;

[0103] take the channel attention features under the energy channels as input of the coordinate attention layer, respectively output horizontal pooling features and vertical pooling features through a pooling layer in the coordinate attention layer, and splice the horizontal pooling features and the vertical pooling features under the energy channels to obtain rock formation pooling splicing features under the energy channels;

[0104] based on the rock formation pooling splicing features, output rock formation multi-attention feature maps under the energy channels through a dynamic convolution kernel in the coordinate attention layer, and obtain the rock formation multi-attention feature map based on the rock formation multi-attention feature maps under the respective energy channels.

[0105] Optionally, the encoder comprises a feature extraction network and a feature enhancement network, and the prediction module 81 is further configured to:

[0106] based on the rock formation input image, output rock formation feature maps corresponding to respective encoding layers through the feature extraction network, the rock formation feature maps comprising feature maps under respective energy channels;

[0107] take the rock formation feature maps output by the encoding layers connected with the feature enhancement network as input of the feature enhancement network, and output rock formation enhanced features through the feature enhancement network.

[0108] Optionally, the prediction module 81 is further configured to:

[0109] respectively perform convolution calculation on inputs under respective energy channels in the feature enhancement network by using a dilated convolution with different dilated rates to obtain dilated convolution features under the respective energy channels;

[0110] based on the dilated convolution features under the respective energy channels, perform point convolution processing to obtain the rock formation enhanced features.

[0111] Optionally, the training module 82 is further configured to:

[0112] obtain an actual sample dataset, each actual sample in the actual sample dataset comprising a sample rock layer corresponding sample rock layer image, wherein the sample rock layer image comprises sample images under respective energy channels, the sample images being reconstructed based on projection data of the corresponding energy channels;

[0113] obtain a simulation sample dataset, the simulation sample dataset being generated based on a three-dimensional physical model representing a rock layer and a base material, each simulation sample in the simulation sample dataset comprising a simulation rock layer image and base material label data in the simulation rock layer image;

[0114] in a first iteration process, based on the simulation sample dataset, train the base material recognition model to obtain a pre-trained base material recognition model;

[0115] in a second iteration process, based on the actual sample dataset and the simulation sample dataset, calculate a total loss value of each iteration in the second iteration process based on the pre-trained base material recognition model, and continue to train the base material recognition model based on the total loss value.

[0116] Optionally, the training network of the base material recognition model comprises the encoder, the decoder and an adversarial training network connected to the encoder respectively, and a total loss calculation network, wherein the adversarial training network is used to learn the difference in base material features between the actual sample dataset and the simulation sample dataset through adversarial training, wherein the adversarial training network comprises a domain adversarial discriminator, the encoder comprises a feature extraction network and a feature enhancement network, the feature extraction network is used to output a rock layer sample feature map corresponding to an input sample image, and the feature enhancement network is used to output a rock layer enhanced sample feature corresponding to the input sample image based on the rock layer sample feature map.

[0117] Optionally, the training module 82 is further configured to:

[0118] input the rock layer sample feature map into the decoder and the adversarial training network respectively;

[0119] based on the rock layer sample feature map, output a discrimination result corresponding to the rock layer sample feature map through the domain adversarial discriminator, and calculate a discrimination loss under the current iteration based on the discrimination result;

[0120] based on the rock layer sample feature map, output a base material prediction result corresponding to the rock layer sample feature map through the decoder, and calculate a base material prediction loss value under the current iteration based on the base material prediction result and base material label data corresponding to the rock layer sample feature map;

[0121] Based on the discriminant loss and the base substance prediction loss value, a total loss value in the current iteration is calculated by the total loss calculation network.

[0122] Optionally, the discriminant loss includes a base substance sample domain loss and a base substance feature confusion loss, and the training module 82 is further configured to:

[0123] Based on the rock stratum sample feature map, the domain adversarial discriminator outputs a first probability that the rock stratum sample feature map belongs to the simulation sample dataset and a second probability that the rock stratum sample feature map belongs to the actual sample dataset;

[0124] Based on the first probability and the second probability, the base substance sample domain loss is calculated.

[0125] Based on the second probability, the base substance feature confusion loss is calculated.

[0126] The base substance sample domain loss and the base substance feature confusion loss are weighted to obtain the discriminant loss.

[0127] Referring to Figure 9 In another aspect of the embodiments of the present application, a computing device 10 is provided, which includes a memory 3011 and a processor 3012. The memory 3011 stores a computer program, and the computer program is executed by the processor to cause the processor 3012 to perform the steps of the base substance identification method for rock stratum images based on deep learning provided by any of the embodiments of the present application. The computing device 10 is, for example, a desktop computer, a laptop computer, a tablet computer, a handheld computer, a smart speaker, a server, etc., a mobile phone (for example, a smart phone, a wireless phone, etc.), a wearable device (for example, a pair of smart glasses or a smart watch), or the like.

[0128] The processor 3012 is a control center, which connects various parts of the computer device through various interfaces and lines, performs various functions of the computer device and processes data by running or executing software programs and / or modules stored in the memory 3011, and calling data stored in the memory 3011. Optionally, the processor 3012 can include one or more processing cores; preferably, the processor 3012 can integrate an application processor and a modem processor, wherein the application processor mainly processes the operating system, user pages, and application programs, etc., and the modem processor mainly processes wireless communication. It can be understood that the above-mentioned modem processor can also not be integrated into the processor 3012.

[0129] The memory 3011 can be used to store software programs and modules, and the processor 3012 executes various function applications and data processing by running the software programs and modules stored in the memory 3011. The memory 3011 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, application programs required by at least one function (such as a sound playing function, an image playing function, etc.), and the like; and the data storage area can store data created according to the use of the computer device, etc. In addition, the memory 3011 can include a high-speed random access memory, and can also include a nonvolatile memory, for example, at least one magnetic disk storage device, a flash memory device, or other volatile solid-state memory device. Accordingly, the memory 3011 can also include a memory controller to provide the processor 3012 with access to the memory 3011.

[0130] In another aspect, the embodiment of the present application also provides a storage medium storing a computer program, and the computer program is executed by a processor to make the processor execute the steps of the method for identifying base matter of a rock stratum image based on deep learning provided by any one of the embodiments of the present application.

[0131] Those skilled in the art can understand that all or part of the processes in the method provided by the above-mentioned embodiments can be completed by a computer program instructing related hardware, and the program can be stored in a non-volatile computer readable storage medium, and when the program is executed, the processes of the above-mentioned embodiments of the method can be included. Any reference to memory, storage, database or other medium used in the embodiments provided by the present application can include non-volatile and / or volatile memory. The non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. The volatile memory can include random access memory (RAM) or external cache memory. As an illustration but 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 (Synchl ink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0132] The above merely illustrates the specific embodiments of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of the changes or replacements within the technical range disclosed by the present application, which should be covered in the protection scope of the present application. The protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for identifying the matrix material of rock strata images based on deep learning, characterized in that, The method comprises: obtaining a to-be-detected rock stratum image, the to-be-detected rock stratum image comprising rock stratum images under respective energy channels; forming, based on the to-be-detected rock stratum image, a rock stratum input image to a trained base substance recognition model, wherein the base substance recognition model comprises a decoder and a base substance prediction network, each decoding layer in the decoder comprising an attention layer and an up-sampling layer; obtaining, based on the rock stratum input image, a rock stratum multi-attention feature map output by the attention layer; obtaining, based on the rock stratum input image, an up-sampled rock stratum feature output by the up-sampling layer; fusing the rock stratum multi-attention feature map and the up-sampled rock stratum feature to obtain a rock stratum fusion feature corresponding to the decoding layer; predicting, based on the rock stratum fusion feature corresponding to the decoding layer, base substance data in the to-be-detected rock stratum image through the base substance prediction network. 2.The method of claim 1, wherein, The base substance recognition model comprises an encoder, the encoder comprising a plurality of encoding layers, the encoding layers being configured to output a rock stratum feature map, the rock stratum feature map comprising feature maps under respective energy channels, the attention layer comprising a channel attention layer and a coordinate attention layer, and the obtaining, based on the rock stratum input image, of a rock stratum multi-attention feature map output by the attention layer comprising: obtaining, as input of the attention layer, a rock stratum feature map belonging to a same scale layer as the attention layer, outputting, through the channel attention layer, channel attention features under respective energy channels; taking the channel attention features under the energy channels as input of the coordinate attention layer, outputting, through a pooling layer in the coordinate attention layer, horizontal pooling features and vertical pooling features respectively, splicing the horizontal pooling features and the vertical pooling features under the energy channels to obtain rock stratum pooling splicing features under the energy channels; outputting, based on the rock stratum pooling splicing features, rock stratum multi-attention feature maps under the energy channels through a dynamic convolution kernel in the coordinate attention layer, and obtaining the rock stratum multi-attention feature map based on the rock stratum multi-attention feature maps under the respective energy channels. 3.The method of claim 2, wherein, The encoder comprises a feature extraction network and a feature enhancement network, and the obtaining, as input of the attention layer, of a rock stratum feature map belonging to a same scale layer as the attention layer comprises: outputting, based on the rock stratum input image, rock stratum feature maps corresponding to respective encoding layers through the feature extraction network, the rock stratum feature maps comprising feature maps under respective energy channels; taking, as input of the feature enhancement network, the rock stratum feature maps output by the encoding layers connected to the feature enhancement network, and outputting, through the feature enhancement network, rock stratum enhanced features. 4.The method of claim 2, wherein, The outputting, through the feature enhancement network, of rock stratum enhanced features comprises: performing convolution calculation on input under each energy channel in the feature enhancement network through a respective hollow convolution with a different hollow rate to obtain hollow convolution features under the respective energy channels; performing point convolution processing based on the hollow convolution features under the respective energy channels to obtain the rock stratum enhanced features. 5.The method of claim 1, wherein, The method further comprises: obtain an actual sample dataset, each actual sample in the actual sample dataset comprising a sample rock layer corresponding sample rock layer image, wherein the sample rock layer image comprises sample images under respective energy channels, the sample images being reconstructed based on projection data of the corresponding energy channels; obtain a simulation sample dataset, the simulation sample dataset being generated based on a three-dimensional physical model representing a rock layer and a base material, each simulation sample in the simulation sample dataset comprising a simulation rock layer image and base material label data in the simulation rock layer image; in a first iteration process, train the base material recognition model based on the simulation sample dataset to obtain a pre-trained base material recognition model; in a second iteration process, based on the actual sample dataset and the simulation sample dataset, calculate a total loss value of each iteration in the second iteration process, and continue to train the base material recognition model based on the total loss value. 6.The method of claim 5, wherein, The training network of the base material recognition model comprises the encoder, the decoder and the adversarial training network connected with the encoder respectively, and a total loss calculation network, wherein the adversarial training network is used to learn the difference in base material features between the actual sample dataset and the simulation sample dataset through adversarial training, wherein the adversarial training network comprises a domain adversarial discriminator, the encoder comprises a feature extraction network and a feature enhancement network, the feature extraction network is used to output a rock layer sample feature map corresponding to the input sample image, and the feature enhancement network is used to output a rock layer enhanced sample feature corresponding to the input sample image based on the rock layer sample feature map. 7.The method of claim 6, wherein, The total loss value of each iteration in the second iteration process is calculated based on the actual sample dataset and the simulation sample dataset, and the total loss value of each iteration in the second iteration process comprises: the rock layer sample feature map is input into the decoder and the adversarial training network respectively; based on the rock layer sample feature map, the domain adversarial discriminator outputs a discrimination result corresponding to the rock layer sample feature map, and based on the discrimination result, a discrimination loss in the current iteration is calculated; based on the rock layer sample feature map, the decoder outputs a base material prediction result corresponding to the rock layer sample feature map, and based on the base material prediction result and the base material label data corresponding to the rock layer sample feature map, a base material prediction loss value in the current iteration is calculated; based on the discrimination loss and the base material prediction loss value, the total loss calculation network calculates a total loss value in the current iteration. 8.The method of claim 7, wherein, The discrimination loss comprises a base material sample domain loss and a base material feature confusion loss, and the discrimination loss in the current iteration is calculated based on the discrimination result, which comprises: based on the rock layer sample feature map, the domain adversarial discriminator outputs a first probability that the rock layer sample feature map belongs to the simulation sample dataset and a second probability that the rock layer sample feature map belongs to the actual sample dataset; calculating the base material sample domain loss based on the first probability and the second probability; calculating the base material feature confusion loss based on the second probability; weighting the base material sample domain loss and the base material feature confusion loss to obtain the discrimination loss.

9. A computer program product comprising a computer program, characterized in that, The computer program is executed by a processor to implement the base material identification method for rock stratum images based on deep learning according to any one of claims 1 to 8.

10. A computing device, comprising: A computer device comprising a memory and a processor, wherein the memory stores a computer program, and the computer program is executed by the processor to enable the processor to execute the base material identification method for rock stratum images based on deep learning according to any one of claims 1 to 8.

Images