A method and device for identifying and classifying regional fractures

Abstract

The present invention relates to a method and device for regional fracture identification and occurrence classification, comprising: obtaining an image to be detected; inputting the image to be detected into a fully trained fracture identification model to obtain a predicted fracture binary image, wherein the fracture identification model includes a backbone feature extraction layer for pixel differential edge detection based on a residual network structure; clipping the predicted fracture binary image, fitting at least one fracture according to the adjacent pixel information of the fracture pixel points of the clipped image block, and classifying the at least one fracture. The present invention constructs an edge detection model based on pixel differential to realize the identification of fractures in the outcrop area in the field, and calculates the angle of the identified fracture to realize fracture classification, which solves the low accuracy of rock fracture identification and classification problems in the current reservoir fracture research in the outcrop area, and improves the accuracy of rock fracture identification and category visibility.

Description

Technical Field

[0001] The present invention relates to the field of image recognition technology, and in particular to a method and device for regional crack recognition and occurrence classification. Background Art

[0002] Fractured oil and gas reservoirs are an important research object in the field of petroleum geology. It has always been difficult to intelligently identify fractures in outcrop areas in the field. The main reason is that the degree of fracture development and the occurrence of fractures in outcrop areas in the field are very diverse. Traditional manual identification is inefficient and statistical measurement methods are insufficient, which brings inconvenience to geological applications.

[0003] Therefore, how to carry out accurate, efficient and fast intelligent field crack identification method is an urgent problem to be solved. Summary of the invention

[0004] In view of this, it is necessary to provide a method and device for identifying and classifying zone fractures, so as to overcome the problem that the identification of fractures in the exposed zone in the prior art is not intelligent and efficient enough.

[0005] In order to solve the above technical problems, the present invention provides a method for regional fracture identification and occurrence classification, comprising:

[0006] Acquire the image to be detected;

[0007] Inputting the image to be detected into a well-trained crack recognition model to obtain a predicted crack binary image, wherein the crack recognition model includes a backbone feature extraction layer for pixel differential edge detection based on a residual network structure;

[0008] The predicted crack binary image is clipped, and at least one crack is fitted according to the adjacent pixel information of the crack pixel points of the clipped image block, and the at least one crack is classified.

[0009] Furthermore, the training process of the crack identification model includes:

[0010] Acquire an outcrop area fracture sample set containing annotation information, wherein the annotation information includes a real fracture binary image corresponding to a sample image in the outcrop area fracture sample set;

[0011] After the outcrop area fracture sample set is augmented, it is input into the constructed fracture identification model, and a predicted fracture binary map is output;

[0012] Determining a loss function according to an error between the predicted crack binary image and the real crack binary image;

[0013] According to the loss function, the constructed crack identification model is trained until convergence to obtain the fully trained crack identification model.

[0014] Furthermore, the network model structure of the constructed crack recognition model includes an image input layer, a backbone feature extraction layer, and a crack recognition layer, wherein:

[0015] The image input layer is used to normalize the size of the sample image;

[0016] The backbone feature extraction layer is used to extract features from the sample image and output valid features;

[0017] The crack identification layer is used to judge and predict the effective features to obtain the predicted crack binary image.

[0018] Furthermore, the backbone feature extraction layer includes a central difference convolution layer, a radial difference convolution layer, an angular difference convolution layer and a base convolution layer connected in sequence, wherein:

[0019] The center difference convolution layer is used to replace the original pixels of the local feature block extracted by the convolution kernel with the center pixel difference by using the center difference method, generate a center feature map, and perform a pooling operation;

[0020] The radial difference convolution layer is used to replace the original pixels of the local feature block extracted by the convolution kernel with radial pixel differences in a radial difference manner, generate a radial feature map, and perform a pooling operation;

[0021] The angular difference convolution layer is used to replace the original pixels of the local feature block extracted by the convolution kernel with angular pixel differences by using angular difference, generate an angular feature map, and perform a pooling operation;

[0022] The basic convolution layer is used to perform a convolution operation on the angular feature map after the pooling operation to generate a basic convolution feature map.

[0023] Furthermore, the backbone feature extraction layer further includes a first convolution sampling layer, a second convolution sampling layer, a third convolution sampling layer and a fourth convolution sampling layer, wherein:

[0024] The first convolution sampling layer sequentially performs convolution, bilinear difference upsampling, and error calculation operations on the central feature map;

[0025] The second convolution sampling layer sequentially performs convolution, bilinear difference upsampling, and error calculation operations on the radial feature map;

[0026] The third convolution sampling layer sequentially performs convolution, bilinear difference upsampling, and error calculation operations on the angular feature map;

[0027] The fourth convolution sampling layer sequentially performs convolution, bilinear difference upsampling, and error calculation operations on the basic convolution feature map.

[0028] Further, the step of calculating the error from the first convolution sampling layer to the fourth convolution sampling layer includes:

[0029] Determine a first value and a second value respectively according to a positive sample set, a negative sample set and a balance factor for balancing positive and negative samples in the outcrop area fracture sample set;

[0030] Comparing a preset threshold with a first probability that the crack is a true value, and determining at least one situation;

[0031] For the at least one situation, a side loss function is determined according to the first value, the second value and a second probability that the pixel value is a true crack.

[0032] Furthermore, the crack identification layer includes a feature fusion layer, a first convolution layer and an error calculation layer, wherein:

[0033] The feature fusion layer is used to perform feature fusion on the sampling feature graph obtained after bilinear difference upsampling from the first convolution sampling layer to the fourth convolution sampling layer to obtain a feature fusion graph;

[0034] The first convolution layer is used to perform a convolution operation on the feature fusion map to obtain a convolution feature map;

[0035] The error calculation layer is used to determine the total loss function according to the sum of the side loss functions obtained by calculating the errors from the first convolution sampling layer to the fourth convolution sampling layer, wherein the total loss function is used to perform error back propagation and adjust the network structure.

[0036] Furthermore, the adjacent pixel information includes spatial proximity, directional consistency and alignment consistency. The step of clipping the predicted crack binary image and fitting at least one crack according to adjacent pixel information of crack pixels of the clipped image block includes:

[0037] Cut out corresponding image blocks from the predicted crack binary image, and characterize the image blocks by crack region frames, crack classification probabilities, and crack directions to obtain image block expressions;

[0038] According to the image block expression, performing an adjacent detection in a preset field to determine the spatial proximity, the directional consistency and the alignment consistency;

[0039] Determining, according to the spatial proximity, the directional consistency and the alignment consistency, a mapping probability that different pixel points belong to the same crack;

[0040] According to the mapping probability, a set of pixel points of the same crack is determined, and pixel points with preset values ​​are extracted, and a fitting straight line of at least one crack is fitted by a least square method;

[0041] For the fitting straight line of the at least one crack, a coordinate set in the X-axis direction and a coordinate set in the Y-axis direction are formed, and the straight line probability is fitted in the coordinate set in the X-axis direction and the coordinate set in the Y-axis direction to determine the final optimal straight line.

[0042] Further, the classifying the at least one crack comprises:

[0043] Determining the direction of the at least one crack according to the angle of the optimal straight line;

[0044] The type of the at least one crack is determined according to the direction of the at least one crack.

[0045] The present invention also provides a device for identifying and classifying regional fractures, comprising:

[0046] An acquisition unit, used for acquiring an image to be detected;

[0047] An extraction unit, used for inputting the image to be detected into a well-trained crack recognition model to obtain a predicted crack binary image, wherein the crack recognition model includes a backbone feature extraction layer for performing pixel differential edge detection based on a residual network structure;

[0048] The classification unit is used to clip the predicted crack binary image, fit at least one crack according to the adjacent pixel information of the crack pixel points of the clipped image block, and classify the at least one crack.

[0049] Compared with the prior art, the beneficial effects of the present invention include: first, effectively acquiring the image to be detected; then, using the backbone feature extraction layer based on the residual network structure for convolution to form an efficient recognition fracture recognition model, avoiding the problem of overfitting and gradient explosion or disappearance in the network, effectively making judgments and predictions, and determining the predicted fracture binary map; finally, the predicted fracture binary map is successively clipped and fitted, at least one fracture is identified, and the fracture is classified and judged, and the corresponding fracture type is intelligently identified. In summary, the present invention constructs an edge detection model based on pixel difference to realize the recognition of fractures in the outcrop area in the field, and calculates the angle of the identified fracture to realize the classification of fractures, which solves the low accuracy and classification problems of rock fracture recognition in the outcrop area in the current reservoir fracture research, and improves the accuracy of rock fracture recognition and category visibility. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1A schematic diagram of a flow chart of an embodiment of a method for regional fracture identification and occurrence classification provided by the present invention;

[0051] Figure 2 The present invention provides Figure 1 A flow chart of an embodiment of a training process of a crack recognition model in step S102;

[0052] Figure 3 A structural schematic diagram of an embodiment of a crack identification model provided by the present invention;

[0053] Figure 4 A schematic diagram of the structure of an embodiment of the first convolution sampling layer to the fourth convolution sampling layer provided by the present invention;

[0054] Figure 5 The present invention provides Figure 1 A schematic diagram of a flow chart of an embodiment of step S103;

[0055] Figure 6 The present invention provides Figure 1 A schematic diagram of another embodiment of step S103;

[0056] Figure 7 A schematic diagram for describing the positions of adjacent pixels according to an embodiment of the present invention;

[0057] Figure 8 A detection schematic diagram of an embodiment of a crack identification result provided by the present invention;

[0058] Fig. 9 A schematic diagram of the structure of an embodiment of a device for identifying and classifying regional fractures provided by the present invention;

[0059] Fig.10 A schematic structural diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0060] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not used to limit the scope of the present invention.

[0061] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0062] In the description of the present invention, reference to an "embodiment" means that a particular feature, structure, or characteristic described in conjunction with the embodiment may be included in at least one embodiment of the present invention. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the described embodiments may be combined with other embodiments.

[0063] The present invention provides a method and device for regional fracture identification and occurrence classification, constructs a fracture identification model based on pixel difference, fits multiple fractures, and performs corresponding fracture type identification, providing a new idea for further improving the intelligence and efficiency of fracture identification.

[0064] Before describing the embodiments, the following definitions are given for the relevant terms:

[0065] Cracks in field outcrop areas: Cracks existing in field outcrop areas indicate the possibility of fractured oil and gas reservoirs.

[0066] Based on the description of the above technical terms, in the prior art, the field outcrop area fractures are often identified by manual identification or statistical methods, which has problems such as low efficiency and insufficient statistics, causing various inconveniences to the actual exploration. Therefore, the present invention aims to propose an efficient and accurate method for regional fracture identification and occurrence classification to solve the above problems.

[0067] The specific embodiments are described in detail below:

[0068] The embodiment of the present invention provides a method for identifying and classifying regional fractures. Figure 1 FIG. 1 is a flow chart of an embodiment of a method for regional fracture identification and occurrence classification provided by the present invention, including steps S101 to S104, wherein:

[0069] In step S101, an image to be detected is obtained;

[0070] In step S102, the image to be detected is input into a well-trained crack recognition model to obtain a predicted crack binary image, wherein the crack recognition model includes a backbone feature extraction layer that performs convolution based on a residual network structure;

[0071] In step S103, the predicted crack binary image is clipped, and at least one crack is fitted according to the adjacent pixel information of the crack pixel points of the clipped image block, and the at least one crack is classified.

[0072] In an embodiment of the present invention, first, the image to be detected is effectively acquired; then, a high-efficiency crack recognition model is formed by using a backbone feature extraction layer based on a residual network structure for convolution, so as to avoid the problems of overfitting and gradient explosion or disappearance in the network, effectively perform judgment and prediction, and determine a predicted crack binary map; finally, the predicted crack binary map is successively trimmed and fitted to identify at least one crack, and the crack is classified and judged, and the corresponding crack type is intelligently identified.

[0073] As a preferred embodiment, Figure 2 Come and see, Figure 2 The present invention provides Figure 1 Step S102 is a flow chart of an embodiment of a training process of a crack recognition model, specifically including steps S201 to S204, wherein:

[0074] In step S201, an outcrop area fracture sample set containing annotation information is obtained, wherein the annotation information includes a real fracture binary image corresponding to a sample image in the outcrop area fracture sample set;

[0075] In step S202, the fracture sample set in the outcrop area is augmented and then input into the constructed fracture identification model to output a predicted fracture binary map;

[0076] In step S203, a loss function is determined according to an error between the predicted crack binary image and the real crack binary image;

[0077] In step S204, the constructed crack identification model is trained to convergence according to the loss function to obtain the fully trained crack identification model.

[0078] In the embodiment of the present invention, the crack identification model is effectively trained.

[0079] As a more specific embodiment, step S201 specifically includes:

[0080] Obtain multiple real outcrop area crack samples and mark the outcrop crack samples and non-outcrop crack samples. There is a certain regularity in the cracks in the field outcrop area. According to the different crack directions, select about 120 image samples, use the image marking tool labelme to mark the collected samples for cracks, and generate a binary image of the cracks, with cracks as 1 and non-cracks as 0. It is used for loss function calculation in step 4;

[0081] The labeled samples are augmented by mirroring, rotating, blurring, denoising, etc. A variety of augmentation processes are performed on the labeled samples to make the data set as diverse as possible, so that the trained model has stronger generalization ability;

[0082] The augmented image samples are divided into training set, validation set and test set according to the ratio of 7:2:1. The number of augmented image samples is 1338, including 936 training sets, 266 validation sets and 146 test sets.

[0083] In the embodiment of the present invention, the richness of the fracture samples in the outcrop area is ensured through data augmentation.

[0084] As a preferred embodiment, Figure 3 Come and see, Figure 3 This is a schematic diagram of the structure of an embodiment of a crack recognition model provided by the present invention. The network model structure of the constructed crack recognition model includes an image input layer, a backbone feature extraction layer, and a crack recognition layer, wherein:

[0085] The image input layer is used to normalize the size of the sample image;

[0086] The backbone feature extraction layer is used to extract features from the sample image and output valid features;

[0087] The crack identification layer is used to judge and predict the effective features to obtain the predicted crack binary image.

[0088] In an embodiment of the present invention, the improved RCF network is trained using the data set until the network converges, and the converged network parameters are saved as a crack recognition model.

[0089] In a specific embodiment of the present invention, improving the RCF network includes:

[0090] Image input module, used to normalize the image size;

[0091] The improved VGG16 (16-layer VGG convolutional network, VGG stands for Visual Geometry Group) backbone feature extraction network is used to extract features from crack images and output effective features;

[0092] The crack identification module judges and predicts the effective features to obtain the expected results.

[0093] As a preferred embodiment, the backbone feature extraction layer includes a central difference convolution layer, a radial difference convolution layer, an angular difference convolution layer and a base convolution layer connected in sequence, wherein:

[0094] The center difference convolution layer is used to replace the original pixels of the local feature block extracted by the convolution kernel with the center pixel difference by using the center difference method, generate a center feature map, and perform a pooling operation;

[0095] The radial difference convolution layer is used to replace the original pixels of the local feature block extracted by the convolution kernel with radial pixel differences in a radial difference manner, generate a radial feature map, and perform a pooling operation;

[0096] The angular difference convolution layer is used to replace the original pixels of the local feature block extracted by the convolution kernel with angular pixel differences by using angular difference, generate an angular feature map, and perform a pooling operation;

[0097] The basic convolution layer is used to perform a convolution operation on the angular feature map after the pooling operation to generate a basic convolution feature map.

[0098] In the embodiment of the present invention, edge detection is performed using multi-angle differences.

[0099] As a preferred embodiment, the backbone feature extraction layer further includes a first convolution sampling layer, a second convolution sampling layer, a third convolution sampling layer and a fourth convolution sampling layer, wherein:

[0100] The first convolution sampling layer sequentially performs convolution, bilinear difference upsampling, and error calculation operations on the central feature map;

[0101] The second convolution sampling layer sequentially performs convolution, bilinear difference upsampling, and error calculation operations on the radial feature map;

[0102] The third convolution sampling layer sequentially performs convolution, bilinear difference upsampling, and error calculation operations on the angular feature map;

[0103] The fourth convolution sampling layer sequentially performs convolution, bilinear difference upsampling, and error calculation operations on the basic convolution feature map.

[0104] In an embodiment of the present invention, corresponding convolution, bilinear difference upsampling, and error calculation operations are performed on multiple feature maps obtained by differentials through a first convolution sampling layer, a second convolution sampling layer, a third convolution sampling layer, and a fourth convolution sampling layer.

[0105] As a preferred embodiment, the step of calculating the error from the first convolution sampling layer to the fourth convolution sampling layer includes:

[0106] Determine a first value and a second value respectively according to a positive sample set, a negative sample set and a balance factor for balancing positive and negative samples in the outcrop area fracture sample set;

[0107] Comparing a preset threshold with a first probability that the crack is a true value, and determining at least one situation;

[0108] For the at least one situation, a side loss function is determined according to the first value, the second value and a second probability that the pixel value is a true crack.

[0109] In the embodiment of the present invention, a side loss function is effectively constructed.

[0110] In a specific embodiment of the present invention, Figure 4 Come and see, Figure 4 The schematic diagram of the structure of the first convolution sampling layer to the fourth convolution sampling layer provided by the present invention is a schematic diagram of the structure of an embodiment of the present invention, the core PDCblock (Pixel differential convolution block) in the VGG16 backbone feature extraction network module, the structure is as follows Figure 4 As shown in (a), compared with the traditional VGG16 that uses basic convolution, each convolution block in PDCblock adopts the ResNet (Deep Residual Learning for Image Recognition, residual network) structure, as shown in Figure 4 (b) is used to avoid overfitting and gradient explosion or disappearance in the network. The image is input to the crack recognition module, and four PDCblock operations and three 2×2 maximum pooling operations are performed in sequence. After each PDCblock extracts features, it enters the side structure, and is output as a feature layer after one 1×1 convolution and one bilinear difference upsampling to restore the image size. The specific steps are as follows:

[0111] In the first step, the training data set is preprocessed in the input module; for example, the preprocessed crack data is uniformly adjusted to 3×256×256;

[0112] In the second step, the processed data set is input into the backbone feature extraction module in the improved RCF network for feature extraction and obtains effective feature layers of 32×256×256, 64×128×128, 128×64×64, and 128×32×32;

[0113] The core of the backbone feature extraction module is PDCblock, which has the following structure: Figure 4 As shown in (a), it consists of a central difference convolution layer, a radial difference convolution layer, an angular difference convolution layer and a basic convolution layer. When performing convolution operation, each difference convolution layer replaces the original pixels in the local feature block extracted by the convolution kernel with pixel differences. The operation formula is: where x i 、x' i is the input pixel, wi is the convolution kernel weight, p={(x1, x'1),(x2, x'2)…(xn 、x' n )} is the pixel pair in the current feature block. This can better obtain the gradient information of the image, thus facilitating the detection of crack edges;

[0114] Among them, the pixel pair (x i 、x' i ) Use central difference, radial difference, and angular difference to obtain pixel pairs. For example, for a 3×3 feature block, the corresponding pixel points described by row are x1, x2, x3...x9, the center point pixel is x5, and the pixel pairs are (x1, x5), (x2, x5)...(x9, x5);

[0115] In the third step, the valid feature map obtained after each PDCblock is subjected to one 1×1 convolution, one upsampling operation, and one loss function calculation in sequence;

[0116] Among them, each time after PDCblock feature extraction, the image size becomes 1 / 2 of that before extraction, and the number of channels becomes twice. First, a 1×1 convolution is performed to retain the image size and change the channel to 1. Then, a bilinear interpolation upsampling operation is performed to restore the image to the original image size. Finally, the loss function is calculated for the restored image and the true label value.

[0117] As a preferred embodiment, the crack identification layer includes a feature fusion layer, a first convolution layer and an error calculation layer, wherein:

[0118] The feature fusion layer is used to perform feature fusion on the sampling feature graph obtained after bilinear difference upsampling from the first convolution sampling layer to the fourth convolution sampling layer to obtain a feature fusion graph;

[0119] The first convolution layer is used to perform a convolution operation on the feature fusion map to obtain a convolution feature map;

[0120] The error calculation layer is used to determine the total loss function according to the sum of the side loss functions obtained by calculating the errors from the first convolution sampling layer to the fourth convolution sampling layer, wherein the total loss function is used to perform error back propagation and adjust the network structure.

[0121] In the embodiment of the present invention, the network structure is efficiently adjusted through the total loss function.

[0122] In a specific embodiment of the present invention, the restored four single-channel effective feature maps are subjected to a feature fusion operation, followed by a 1×1 convolution operation and a loss function calculation. The specific steps are as follows:

[0123] Step 1: Calculate the total loss function, L = ∑i,jlij, which is the sum of each side loss function. The side loss function is expressed by the following formula:

[0124]

[0125]

[0126]

[0127] Where Y+ and Y- represent the positive sample set and the negative sample set respectively. λ is the factor to balance the positive and negative samples, and is a constant. l(Xi;W) is the loss function of the i-th feature pixel Xi, yi is the probability that the crack is a true value, and p(Xi) is the probability that Xi is a true crack. η is the threshold. If the probability that a pixel is an edge pixel is less than the threshold η, then the pixel is considered to be a non-edge pixel;

[0128] Step 2, error back propagation, the improved RCF edge detection adjusts its own structural parameters based on gradient descent;

[0129] Step 3: Repeat the above steps until the prediction error converges to complete the network training, and save the converged network as a crack recognition model.

[0130] As a preferred embodiment, the adjacent pixel information includes spatial proximity, directional consistency and alignment consistency, combined with Figure 5 Come and see, Figure 5 The present invention provides Figure 1 The flowchart of step S103 in an embodiment of the present invention specifically includes steps S501 to S505, wherein:

[0131] In step S501, the predicted crack binary image is cut out into corresponding image blocks, and the image blocks are characterized by crack region frames, crack classification probabilities, and crack directions to obtain image block expressions;

[0132] In step S502, according to the image block expression, a neighbor detection in a preset field is performed to determine the spatial proximity, the directional consistency and the alignment consistency;

[0133] In step S503, the mapping probability that different pixel points belong to the same crack is determined according to the spatial proximity, the directional consistency and the alignment consistency;

[0134] In step S504, according to the mapping probability, a set of pixel points of the same crack is determined, and pixel points with preset values ​​are extracted, and a fitting straight line of at least one crack is fitted using the least square method;

[0135] In step S505, for the fitting straight line of the at least one crack, a coordinate set in the X-axis direction and a coordinate set in the Y-axis direction are formed, and the straight line probability is fitted in the coordinate set in the X-axis direction and the coordinate set in the Y-axis direction to determine the final optimal straight line.

[0136] In the embodiment of the present invention, a fitting straight line of at least one crack is efficiently fitted.

[0137] As a preferred embodiment, Figure 6 Come and see, Figure 6 The present invention provides Figure 1 Step S103 is a flow chart of another embodiment, wherein step S103 specifically includes steps S601 to S602, wherein:

[0138] In step S601, the direction of the at least one crack is determined according to the angle of the optimal straight line;

[0139] In step S602, the type of the at least one crack is determined according to the direction of the at least one crack.

[0140] In the embodiment of the present invention, the type of at least one crack is efficiently determined by the angle.

[0141] In a specific embodiment of the present invention, Figure 7 Come and see, Figure 7 This is a schematic diagram of the position description of an embodiment of adjacent pixels provided by the present invention. The corresponding image block is cut from the binary image of the identified crack binary image. The adjacent information detection is performed on the sampled crack pixel points to determine the same crack information. Then, 10% of the pixels on the same crack are randomly obtained and fitted into a straight line using the least squares method. The angle of the straight line is obtained to determine the direction of the crack, and then the crack is classified according to the direction, wherein:

[0142] Step 01, the identified crack block is described as: di = {Bi, pi, θi}, where: Bi is the crack region box; pi is the crack classification probability; θi is the estimated crack direction, α i,j The angle formed by connecting two center points. The adjacent detection is described as follows Figure 4 , let di be the detected crack patch centered on xi, and dj be the adjacent detection within the neighborhood Ni of di centered on xj. The adjacent information of di and dj is defined as follows:

[0143] Among them, the spatial proximity between dj and di is expressed as:

[0144] (σ x=rswi, where rθ is a constant and wi is the pixel block width)

[0145] Among them, the directional consistency of dj and di is expressed as:

[0146] (σ θ =rθwi, where rθ is a constant)

[0147] Among them, the alignment consistency of dj and di is expressed as:

[0148] (σ x = rαwi, where rα is a constant)

[0149] Then, the probability that xj and xi belong to the same crack is:

[0150]

[0151] Among them, the sigmoid function is calculated for P, and the value of P is mapped to [0,1]. If P>0.5, xj and xi belong to the same crack;

[0152] Step 02: After the judgment in step 01, a set of pixel points of the same crack is obtained, 10% of the pixel points are randomly selected, and the least square method is used to fit the pixel points into a straight line, and the angle of the straight line is obtained to determine the direction of the crack.

[0153] Among them, the number of pixels on the crack l is m, and the coordinate set in the X-axis direction is The coordinate set in the Y-axis direction is Where l = 1, 2..., n = 0.1 × m. The slope of the straight line is continuously fitted to the values ​​in the X and Y sets, and finally the optimal straight line is obtained;

[0154] In step 03, according to the crack directions obtained in step 6.2, the cracks are classified into different crack categories, such as horizontal cracks, low-angle cracks, oblique angle cracks, and vertical cracks, and visualized as different colors.

[0155] Combine the following Figure 8 Come and see, Figure 8 A detection schematic diagram of an embodiment of a crack identification result provided by the present invention is used to better illustrate the technical solution of the present invention with a specific application example:

[0156] Step 1: Obtain multiple real outcrop area fracture samples and mark the outcrop fracture samples and non-outcrop fracture samples;

[0157] Step 2: Perform augmentation processing such as mirroring, rotation, blurring, and denoising on the labeled samples;

[0158] Step 3: Divide the augmented image samples into training set, validation set, and test set according to the ratio of 7:2:1;

[0159] Step 4: using the data set to train an improved RCF (Richer Convolutional Features for Edge Detection) edge detection network until the network converges, and saving the converged network parameters as a crack recognition model;

[0160] Step 5: inputting the collected crack image to be detected into the crack recognition model, and outputting the recognition information of the cracks in the crack image to be detected;

[0161] Step 6: Input the identified cracks into the crack classification module and output color cracks representing different categories.

[0162] The embodiment of the present invention also provides a device for identifying and classifying regional fractures. Fig. 9 Come and see, Fig. 9 The schematic diagram of the structure of an embodiment of a device for identifying and classifying fracture zones provided by the present invention is as follows. The device for identifying and classifying fracture zones 900 comprises:

[0163] An acquisition unit 901 is used to acquire an image to be detected;

[0164] An extraction unit 902 is used to input the image to be detected into a well-trained crack recognition model to obtain a predicted crack binary image, wherein the crack recognition model includes a backbone feature extraction layer that performs convolution based on a residual network structure;

[0165] The classification unit 903 is used to clip the predicted crack binary image, fit at least one crack according to the adjacent pixel information of the crack pixel points of the clipped image block, and classify the at least one crack.

[0166] The more specific implementation of each unit of the zone fracture identification and occurrence classification device can refer to the description of the above-mentioned zone fracture identification and occurrence classification method, and has similar beneficial effects, which will not be repeated here.

[0167] The embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the method for identifying zone fractures and classifying occurrences as described above is implemented.

[0168] Generally speaking, the computer instructions for implementing the method of the present invention can be carried by any combination of one or more computer-readable storage media. Non-transitory computer-readable storage media can include any computer-readable media except the signal itself that is temporarily propagating.

[0169] The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more conductors, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, apparatus, or device.

[0170] Computer program code for performing the operation of the present invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages, in particular, Python suitable for neural network computing and platform frameworks based on TensorFlow, PyTorch, etc. can be used. The program code can be executed entirely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (for example, using an Internet service provider to connect through the Internet).

[0171] The embodiment of the present invention further provides an electronic device, Fig.10 Come and see, Fig.10 This is a schematic diagram of the structure of an embodiment of an electronic device provided by the present invention. The electronic device 1000 includes a processor 1001, a memory 1002, and a computer program stored in the memory 1002 and executable on the processor 1001. When the processor 1001 executes the program, the method for identifying and classifying the zone fractures and / or ...

[0172] As a preferred embodiment, the electronic device 1000 further includes a display 1003 for displaying the processor 1001 executing the above-mentioned method for identifying zone fractures and classifying their occurrence.

[0173] Exemplarily, the computer program may be divided into one or more modules / units, one or more modules / units are stored in the memory 1002, and are executed by the processor 1001 to complete the present invention. One or more modules / units may be a series of computer program instruction segments that can complete specific functions, and the instruction segments are used to describe the execution process of the computer program in the electronic device 1000. For example, the computer program may be divided into the acquisition unit 901, the extraction unit 902, and the classification unit 903 in the above embodiment. The specific functions of each unit are as described above, and are not repeated here.

[0174] The electronic device 1000 may be a desktop computer, a notebook, a PDA or a smart phone with an adjustable camera module.

[0175] Among them, the processor 1001 may be an integrated circuit chip with signal processing capabilities. The above-mentioned processor 1001 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The disclosed methods, steps and logic block diagrams in the embodiments of the present invention can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.

[0176] The memory 1002 may be, but is not limited to, a random access memory (RAM), a read only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable read-only memory (EEPROM), etc. The memory 1002 is used to store programs, and the processor 1001 executes the program after receiving the execution instruction. The process definition method disclosed in any of the above embodiments of the present invention can be applied to the processor 1001, or implemented by the processor 1001.

[0177] The display 1003 may be an LCD display screen or an LED display screen, for example, a display screen on a mobile phone.

[0178] Understandably, Fig.10 The structure shown is only a schematic diagram of a structure of the electronic device 1000. The electronic device 1000 may also include Fig.10 More or fewer components as shown. Fig.10 Each component shown in the figure can be implemented by hardware, software or a combination thereof.

[0179] The computer-readable storage medium and electronic device provided according to the above embodiments of the present invention can be implemented with reference to the specific description of the method for zone fracture identification and occurrence classification as described above according to the present invention, and have similar beneficial effects as the method for zone fracture identification and occurrence classification as described above, which will not be repeated here.

[0180] The present invention discloses a method and device for regional crack identification and occurrence classification. First, the image to be detected is effectively acquired; then, a high-efficiency crack identification model is formed by using a backbone feature extraction layer based on a residual network structure for convolution, so as to avoid the problems of overfitting and gradient explosion or disappearance in the network, effectively perform judgment and prediction, and determine a predicted crack binary map; finally, the predicted crack binary map is clipped and fitted in turn, at least one crack is identified, and the crack is classified and judged, and the corresponding crack type is intelligently identified.

[0181] The technical solution of the present invention constructs an edge detection model based on pixel difference to realize the identification of cracks in the outcrop area in the field, and calculates the angles of the identified cracks to realize the classification of cracks, which solves the problems of low accuracy in the identification and classification of rock cracks in the outcrop area in the current reservoir fracture research, and improves the accuracy of rock fracture identification and category visibility.

[0182] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for identifying and classifying regional fractures, characterized in that: include: Acquire the image to be detected; Inputting the image to be detected into a well-trained crack recognition model to obtain a predicted crack binary image, wherein the crack recognition model includes a backbone feature extraction layer for performing pixel differential edge detection based on a residual network structure; Clipping the predicted crack binary image, fitting at least one crack according to adjacent pixel information of crack pixels of the clipped image block, and classifying the at least one crack; The adjacent pixel information includes spatial proximity, directional consistency and alignment consistency, and the predicted crack binary image is clipped to fit at least one crack according to the adjacent pixel information of the crack pixel point of the clipped image block, including: Cut out corresponding image blocks from the predicted crack binary image, and characterize the image blocks by crack region frames, crack classification probabilities, and crack directions to obtain image block expressions; According to the image block expression, performing an adjacent detection in a preset field to determine the spatial proximity, the directional consistency and the alignment consistency; Determining, according to the spatial proximity, the directional consistency and the alignment consistency, a mapping probability that different pixel points belong to the same crack; According to the mapping probability, a set of pixel points of the same crack is determined, and pixel points with preset values ​​are extracted, and a fitting straight line of at least one crack is fitted by a least square method; For the fitting straight line of the at least one crack, a coordinate set in the X-axis direction and a coordinate set in the Y-axis direction are formed, and a straight line probability is fitted in the coordinate set in the X-axis direction and the coordinate set in the Y-axis direction to determine a final optimal straight line; The at least one crack is classified, comprising: Determining the direction of the at least one crack according to the angle of the optimal straight line; The type of the at least one crack is determined according to the direction of the at least one crack.

2. The method for regional fracture identification and occurrence classification according to claim 1, characterized in that: The training process of the crack identification model includes: Acquire an outcrop area fracture sample set containing annotation information, wherein the annotation information includes a real fracture binary image corresponding to a sample image in the outcrop area fracture sample set; After the fracture sample set in the outcrop area is augmented, it is input into the constructed fracture identification model, and a predicted fracture binary map is output; Determining a loss function according to an error between the predicted crack binary image and the real crack binary image; According to the loss function, the constructed crack identification model is trained until convergence to obtain the fully trained crack identification model.

3. The method for regional fracture identification and occurrence classification according to claim 1, characterized in that: The network model structure of the constructed crack recognition model includes an image input layer, a backbone feature extraction layer, and a crack recognition layer, wherein: The image input layer is used to normalize the size of the sample image; The backbone feature extraction layer is used to extract features from the sample image and output valid features; The crack identification layer is used to judge and predict the effective features to obtain the predicted crack binary image.

4. The method for regional fracture identification and occurrence classification according to claim 3 is characterized in that: The backbone feature extraction layer includes a central difference convolution layer, a radial difference convolution layer, an angular difference convolution layer and a base convolution layer connected in sequence, wherein: The center difference convolution layer is used to replace the original pixels of the local feature block extracted by the convolution kernel with the center pixel difference by using the center difference method, generate a center feature map, and perform a pooling operation; The radial difference convolution layer is used to replace the original pixels of the local feature block extracted by the convolution kernel with radial pixel differences in a radial difference manner, generate a radial feature map, and perform a pooling operation; The angular difference convolution layer is used to replace the original pixels of the local feature block extracted by the convolution kernel with angular pixel differences by using angular difference, generate an angular feature map, and perform a pooling operation; The basic convolution layer is used to perform a convolution operation on the angular feature map after the pooling operation to generate a basic convolution feature map.

5. The method for regional fracture identification and occurrence classification according to claim 4, characterized in that: The backbone feature extraction layer further includes a first convolution sampling layer, a second convolution sampling layer, a third convolution sampling layer and a fourth convolution sampling layer, wherein: The first convolution sampling layer sequentially performs convolution, bilinear difference upsampling, and error calculation operations on the central feature map; The second convolution sampling layer sequentially performs convolution, bilinear difference upsampling, and error calculation operations on the radial feature map; The third convolution sampling layer sequentially performs convolution, bilinear difference upsampling, and error calculation operations on the angular feature map; The fourth convolution sampling layer sequentially performs convolution, bilinear difference upsampling, and error calculation operations on the basic convolution feature map.

6. The method for regional fracture identification and occurrence classification according to claim 5, characterized in that: The step of calculating the error from the first convolution sampling layer to the fourth convolution sampling layer comprises: Determine a first value and a second value respectively according to a positive sample set, a negative sample set and a balance factor for balancing positive and negative samples in the outcrop area fracture sample set; Comparing a preset threshold with a first probability that the crack is a true value, and determining at least one situation; For the at least one situation, a side loss function is determined according to the first value, the second value and a second probability that the pixel value is a true crack.

7. The method for regional fracture identification and occurrence classification according to claim 3, characterized in that: The crack identification layer includes a feature fusion layer, a first convolution layer and an error calculation layer, wherein: The feature fusion layer is used to perform feature fusion on the sampling feature graphs obtained after bilinear difference upsampling from the first convolution sampling layer to the fourth convolution sampling layer to obtain a feature fusion graph; The first convolution layer is used to perform a convolution operation on the feature fusion map to obtain a convolution feature map; The error calculation layer is used to determine the total loss function according to the sum of the side loss functions obtained by calculating the errors from the first convolution sampling layer to the fourth convolution sampling layer, wherein the total loss function is used to perform error back propagation and adjust the network structure.

8. A device for identifying and classifying regional fractures, characterized in that: include: An acquisition unit, used for acquiring an image to be detected; An extraction unit, used for inputting the image to be detected into a well-trained crack recognition model to obtain a predicted crack binary image, wherein the crack recognition model includes a backbone feature extraction layer for performing pixel differential edge detection based on a residual network structure; A classification unit, used for clipping the predicted crack binary image, fitting at least one crack according to adjacent pixel information of crack pixels of the clipped image block, and classifying the at least one crack; The adjacent pixel information includes spatial proximity, directional consistency and alignment consistency, and the predicted crack binary image is clipped to fit at least one crack according to the adjacent pixel information of the crack pixel point of the clipped image block, including: Cut out corresponding image blocks from the predicted crack binary image, and characterize the image blocks by crack region frames, crack classification probabilities, and crack directions to obtain image block expressions; According to the image block expression, performing an adjacent detection in a preset field to determine the spatial proximity, the directional consistency and the alignment consistency; Determining, according to the spatial proximity, the directional consistency and the alignment consistency, a mapping probability that different pixel points belong to the same crack; According to the mapping probability, a set of pixel points of the same crack is determined, and pixel points with preset values ​​are extracted, and a fitting straight line of at least one crack is fitted by a least square method; For the fitting straight line of the at least one crack, a coordinate set in the X-axis direction and a coordinate set in the Y-axis direction are formed, and a straight line probability is fitted in the coordinate set in the X-axis direction and the coordinate set in the Y-axis direction to determine a final optimal straight line; The at least one crack is classified, comprising: Determining the direction of the at least one crack according to the angle of the optimal straight line; The type of the at least one crack is determined according to the direction of the at least one crack.

Images