A rock mass fracture intelligent identification and digital reconstruction method

Abstract

The present application relates to the technical field of rock mass engineering, and particularly relates to a rock mass fracture intelligent identification and digital reconstruction method, comprising: collecting a rock mass fracture image; manually marking a color block; based on transfer learning, introducing an already-labeled medical image cell membrane sample and a road fracture sample, and jointly training a fracture identification model; inputting the rock mass fracture image into the fracture identification model to segment the fracture, and building a snake convolution network to extract multi-scale features of the image; digitally reconstructing a binary image to obtain a box set and a polyline set data; and reducing the digital data from a pixel scale to an engineering scale. The present application realizes full-process automation from intelligent identification of a rock mass fracture image to digital reconstruction, and can be widely applied to fields such as geotechnical engineering design, slope disaster identification, underground space stability analysis and the like.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of rock mass engineering, in particular to a rock mass fracture intelligent identification and digital reconstruction method. BACKGROUND

[0002] In the stability analysis of rock mass engineering, the widely existing fractures in the rock mass are the key control factors, and it is crucial to accurately obtain their geometric parameters and spatial distribution characteristics.

[0003] On the one hand, in the current mainstream fracture parameter acquisition technology, although the non-contact image recognition technology based on deep learning has made progress in efficiency and accuracy, it still faces a core bottleneck in development and application: the lack of high-quality and professional fracture image dataset, which leads to poor model training effect, weak generalization ability, insufficient engineering robustness and stability, and restricts the practical process of the technology. On the other hand, in the numerical simulation (such as using FLAC, PFC, etc.) technology of rock mass fractures, the current fracture modeling mainly relies on artificial methods (such as Fish code or CAD drawing), which has inherent defects such as strong subjectivity, low efficiency, unstable accuracy, difficulty in fully expressing complex fracture spatial form and geometric details, and difficulty in guaranteeing model consistency and reliability.

[0004] In summary, the current technology has two key shortcomings: the lack of high-quality fracture dataset at the data level limits the performance and practicality of intelligent identification models; and there is a huge gap between the identification results and the input of numerical simulation at the application level, and the low-efficiency and subjective artificial modeling hinders the automation and standardization of the analysis process. SUMMARY

[0005] The purpose of the present application is to provide a fast and accurate rock mass fracture intelligent identification and digital reconstruction method, which realizes the full-process automation from rock mass fracture image to intelligent identification and then to digital reconstruction.

[0006] In order to achieve the above purpose, the present application provides a rock mass fracture intelligent identification and digital reconstruction method, which comprises the following steps:

[0007] S10, collecting rock mass fracture images;

[0008] S20, selecting a plurality of spatial feature points in the real rock mass surface and the rock mass fracture images, adding a marked color block at the spatial feature point position in the rock mass fracture images in an artificial marking manner, obtaining the coordinate value of the marked color block in the physical space, recording the color corresponding to the hexadecimal code of the marked color block and pairing and recording the image coordinates and the actual physical coordinates;

[0009] S30, based on transfer learning, introducing labeled medical image cell membrane samples and road crack samples, jointly training the crack identification model;

[0010] S40, inputting the rock mass crack image into the crack identification model to segment the crack, and building a snake convolution network to extract multi-scale features of the image;

[0011] S50, digitizing and reconstructing the binary image to obtain a box set and a polyline set data;

[0012] S60, based on the color block calibration method, establishing the corresponding relationship between the pixels and the physical space, selecting a preset calibration color block in the image as a reference area, calculating the pixel-physical scale conversion factor of the image through the known actual size of the marker color block and the pixel length in the image, and combining the actual physical coordinates of the marker color block to construct a mapping model between the pixels and the physical space coordinates, realizing the restoration of the image scale to the actual physical scale.

[0013] Further, after the rock mass crack image is collected by the high-resolution camera in S10, distortion correction, brightness adjustment and noise suppression preprocessing are further performed.

[0014] Further, the sample introduced in S30 for jointly training the crack identification model based on the labeled medical image cell membrane samples and road crack samples is represented as:

[0015] ;

[0016] Wherein, represents the joint training sample, represents the crack data set sample, represents the medical image cell membrane sample.

[0017] Further, the snake convolution network in S40 adopts a learnable sampling grid extending along the curve and additionally predicts the sampling offset layer by layer, explicitly encodes the edge direction and width change of the crack in the feature extraction stage, outputs a crack probability map for each pixel, and the probability map is mapped by Sigmoid and thresholded to obtain a final crack binary image.

[0018] Further, in S50, the binary image of the crack identification result is converted into structured data, and the crack pixels are converted into a plurality of box data.

[0019] Further, in S50, the crack region is refined, and the current pixel is set as , the eight-domain pixels are , the pixel value is 0 representing the background or 1 representing the foreground, is located in The other pixels are numbered in a clockwise direction, and the number of foreground pixels in the defined area is defined As follows:

[0020] ;

[0021] Wherein, The value range of the pixel to physical world scale is 0 to 8;

[0022] The number of jumps from the background pixel to the foreground pixel in the defined area The specific calculation formula is:

[0023] ;

[0024] Wherein, ;

[0025] In the pixel deletion process, each iteration includes two sub-steps, in the first sub-step, when the pixel Satisfies the following conditions at the same time, it is marked as deletable:

[0026] ;

[0027] In the second sub-step, when the pixel Satisfies the following conditions at the same time, it is marked as deletable:

[0028] ;

[0029] In each iteration process, first, the condition shown in the formula of the first sub-step is judged, and the pixel satisfying the condition is deleted; then, the condition shown in the formula of the second sub-step is judged, and the pixel satisfying the condition is deleted; the above iteration is repeated until there is no new pixel to be deleted, so as to obtain a single-pixel-width line, and the thinning result is connected as a plurality of polyline data according to the pixel adjacency relationship, reflecting the spatial path and geometric morphology of the crack.

[0030] Further, when the color block calibration method is adopted in S60, the actual physical distance between the marked color blocks is divided by the pixel distance between the marked color blocks to obtain a pixel to physical world scale, the pixel coordinates of the crack in the binary image are multiplied by the pixel to physical world scale to obtain physical world scale crack coordinates, and the physical world scale crack coordinates are multiplied by the corresponding rotation matrix to obtain physical world crack coordinates.

[0031] The above scheme of the present application has the following beneficial effects:

[0032] The rock mass crack intelligent identification and digital reconstruction method provided by the application can establish the corresponding relationship between pixels and physical space through the color block calibration method, realize the reduction of image scale to actual physical scale, train the crack identification model through transfer learning and the introduction of medical image cell membrane samples and road crack samples, can significantly improve the generalization ability and accuracy in crack image identification, can adapt to the identification requirements of complex crack morphology, uneven contrast and other actual engineering scenes, realize the intelligent identification of rock mass crack image, and realize the full-process automation of digital reconstruction, and can be widely applied to the fields of geotechnical engineering design, slope disaster identification, underground space stability analysis and the like.

[0033] Other benefits of the application will be described in detail in the subsequent specific embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 The step flow chart of the application is shown in the figure;

[0035] Figure 2 The rock mass crack image and the marked color block schematic diagram obtained in the embodiment of the application are shown in the figure;

[0036] Figure 3 The binary image schematic diagram after the crack identification model identification in the embodiment of the application is shown in the figure;

[0037] Figure 4 The digital reconstruction block set data schematic diagram in the embodiment of the application is shown in the figure;

[0038] Figure 5 The digital reconstruction fold line set data schematic diagram in the embodiment of the application is shown in the figure. DETAILED DESCRIPTION

[0039] The embodiments of the present disclosure will be described below through specific specific examples, and those skilled in the art can easily understand other advantages and effects of the present disclosure from the content disclosed in the specification. Obviously, the described embodiments are only part of the embodiments of the present disclosure, not all. The present disclosure can also be implemented or applied by other different specific embodiments, and the details in the specification can be modified or changed based on different views and applications without departing from the spirit of the present disclosure. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict. Based on the embodiments in the present disclosure, all other embodiments obtained by those skilled in the art without creative labor belong to the scope of protection of the present disclosure.

[0040] It is important to note that the various aspects described herein can be implemented in a wide variety of forms, and that any particular structure and / or function described herein is merely illustrative. Based on the teachings herein one skilled in the art should appreciate that an aspect described herein can be implemented independently of any other aspects and that an aspect described herein can be implemented both as any number of software and / or hardware structures and as any number of processes and / or operations.

[0041] It is also important to note that the present disclosure can be carried out in many different forms and have a wide range of applicability therefore the specific exemplification of the embodiments provided here are not to be construed as limiting. Although specific terms can be employed herein, these are used in a generic and descriptive sense only and not for purposes of limitation.

[0042] As shown in Figure 1 Embodiments of the present application provide a rock mass fracture intelligent identification and digital reconstruction method, which first improves the rock mass fracture identification accuracy and robustness based on medical image transfer learning, then digitizes the fracture geometric information into data of box set and polyline set, and completes pixel-physical size calibration based on color block marking method, completes digital reconstruction, and realizes the engineering scale reduction of the fracture. The method specifically comprises:

[0043] S10, collecting rock mass fracture image.

[0044] In this embodiment, the image data of the rock mass surface or borehole with fractures is collected by a high-resolution camera, and pretreatments such as distortion correction, brightness adjustment and noise suppression are performed to meet the subsequent identification accuracy requirements.

[0045] S20, manually marking color blocks.

[0046] In this embodiment, a plurality of, for example, two representative spatial feature points are selected in the real rock mass surface and rock mass fracture image, and a color block is added at the spatial feature point position in the image by manual marking. The coordinates of the color block in the physical space are obtained by total station, three-dimensional laser scanner or manual measurement means, the color of the color block is recorded as a hexadecimal code, and the image coordinates and actual physical coordinates are paired and recorded.

[0047] S30, training the crack identification model with the aid of transfer learning.

[0048] In this embodiment, based on the transfer learning method, the labeled medical image cell membrane samples and road crack samples are introduced to jointly train the crack identification model:

[0049] ;

[0050] Among them, represents the joint training sample, represents the crack data set sample, represents the ISBI-2012 sample (i.e. medical image cell membrane sample). By introducing medical image cell membrane samples and road crack samples, the generalization ability and accuracy of the crack identification model in crack image recognition can be significantly improved, so that it can adapt to the recognition requirements of complex crack morphology, uneven contrast and other actual engineering scenes.

[0051] It should be noted that transfer learning is a machine learning technique, and its core idea is to transfer the knowledge and model parameters learned in one field (source field) to another related but different field (target field) to help improve the learning efficiency and performance of the latter. For the case of serious lack of crack image data set, transfer learning allows the use of large general data set to pre-train the model, and then fine-tune it with relatively small target field data to achieve the purpose of accurate identification of rock mass cracks and strong generalization ability.

[0052] For medical image cell membrane samples and road crack samples, the two are similar to rock mass cracks to some extent in terms of underlying visual patterns, including: cell membranes, road cracks and rock mass cracks usually present as elongated, winding, branching, closed linear or curved structures; have similar image processing problems such as low contrast, fuzzy boundary, complex and variable morphology, background noise interference, etc. Therefore, by training the crack identification model with medical image cell membrane samples and road crack samples, the crack identification model can better learn how to enhance weak edges, connect broken line segments, resist background noise, etc., which is directly related and transferable to rock mass crack identification.

[0053] S40, inputting the rock mass crack image into the crack identification model to obtain a crack identification result binary image.

[0054] In this embodiment, the trained crack identification model is applied to crack segmentation of collected images, and a snake convolutional network is built to extract multi-scale features of the images. Specifically, The snake convolution can be expressed as:

[0055] ;

[0056] The snake convolution can be expressed as:

[0057] ;

[0058] wherein, represents the two-dimensional coordinates of the convolution kernel at the , , represents the two-dimensional coordinates of the convolution kernel at the , , and are index variables, representing the center point position of the current sampling point of the snake convolution, represents the offset, representing the position of how many steps forward / backward from the center position, and respectively represent the initial offset in the horizontal direction and the vertical direction, and respectively represent the offset in the horizontal direction and the vertical direction in each iteration, which controls the offset of the snake convolution sampling point in the horizontal direction and the vertical direction.

[0059] The snake convolution network explicitly encodes the edge direction and width variation of the crack in the feature extraction stage by layer-by-layer using a learnable sampling grid extending along the curve and additionally predicting the sampling offset, thereby outputting a crack probability map for each pixel; the probability map is mapped by Sigmoid and thresholded to obtain the final crack binary image. Unlike the conventional U-Net, the snake convolution replaces the fixed square convolution kernel with dynamic geometric sampling, and maintains the topology and boundary alignment of slender and bifurcated structures through offset regression, continuity / curvature regularization and branchable sampling, thereby performing better in pixel-level positioning and connectivity preservation of narrow ridge-shaped cracks.

[0060] S50, digitizing reconstruction is performed on the binary image to obtain box set and polyline set data.

[0061] In this embodiment, the binary image of the crack identification result is converted into structured data, and the crack pixels are converted into multiple groups of box data. The box can be stored in the format, reflecting the spatial distribution and local geometric features of the crack, wherein represents a two-dimensional coordinate, represents the side length of the box. As a preferred embodiment, the Zhang-Suen thinning algorithm is also used in this embodiment to thin the crack region. Specifically, let the current pixel be , and the eight-field pixels are , and the pixel value is 0 (background) or 1 (foreground). Among them, Located above, other pixels are numbered in clockwise direction. The number of foreground pixels in the defined region is defined as follows:

[0062] ;

[0063] wherein, the value range of is 0 to 8.

[0064] The number of jumps from background pixels to foreground pixels in the defined region is defined as follows:

[0065] ;

[0066] wherein, .

[0067] In the pixel deletion process, each iteration includes two sub-steps. In the first sub-step, when a pixel satisfies the following conditions at the same time, it is marked as deletable:

[0068] ;

[0069] In the second sub-step, when a pixel satisfies the following conditions at the same time, it is marked as deletable:

[0070] ;

[0071] In each iteration process, first, the condition shown in the formula of the first sub-step is judged, and the pixels satisfying the condition are deleted; then, the condition shown in the formula of the second sub-step is judged, and the pixels satisfying the condition are deleted; the above iteration is repeated until there is no new pixel to be deleted, so as to obtain the skeleton representation of the input image (binary image), that is, the single-pixel-width line, and connect the thinning result into multiple groups of polyline data according to the pixel adjacency relationship. The polylines are stored in the format of , reflecting the spatial path and geometric morphology of the fissure.

[0072] S60, the digitized data is restored from the pixel scale to the engineering scale.

[0073] In this embodiment, the corresponding relationship between the pixels and the physical space is established based on the color block calibration method. A preset calibration color block in the image is selected as a reference area. The pixel-physical scale conversion factor of the image is calculated through the known actual size of the calibration color block and the pixel length in the image. Combined with the real coordinates of the marker color block recorded in S20, a mapping model between the pixels and the spatial coordinates is further constructed, so as to realize the restoration of the image scale to the actual physical scale. ​​

[0074] Specifically, the actual physical distance between the marked color blocks is divided by the pixel distance between the marked color blocks to obtain the pixel-to-physical-world scale. The pixel coordinates of the fracture in the binary image are multiplied by the pixel-to-physical-world scale to obtain the physical-world scale fracture coordinates. The physical-world scale fracture coordinates are then multiplied by the corresponding rotation matrix to obtain the physical-world fracture coordinates, thus completing the digital reconstruction of rock mass fractures.

[0075] As described above, the intelligent identification and digital reconstruction method for rock mass fractures provided in this embodiment has advantages such as high identification efficiency and strong modeling accuracy. It realizes full-process automation from rock mass fracture image to intelligent identification and then to digital reconstruction, and can be widely used in geotechnical engineering design, slope disaster identification, underground space stability analysis and other fields.

[0076] The following case further illustrates the effectiveness of this method. In this case, after acquiring images of rock fractures using a high-resolution camera, two color blocks are manually marked, such as... Figure 2 As shown, the coordinates of two color patches in the real world (-2.5m, 5m) and (2.5m, 5m) are recorded. The hexadecimal color values ​​corresponding to the two color patches and their real coordinates are input. After being recognized by a trained crack recognition model, a binary image is obtained, as shown. Figure 3 As shown; the obtained binary image is digitally reconstructed and restored to the engineering scale. A schematic diagram of the digital reconstruction bounding box set data is shown below. Figure 4 As shown in the diagram, the data of the digitally reconstructed polyline set is illustrated below. Figure 5 As shown in the diagram. It should be noted that these data are schematic diagrams displayed by the numerical simulation software after being imported. The diagrams demonstrate that this method has a very high accuracy rate in identifying rock mass cracks, and it matches the actual rock mass crack images very well.

[0077] Based on the same inventive concept, this embodiment also provides hardware for intelligent identification and digital reconstruction of rock mass fractures, including: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of the intelligent identification and digital reconstruction method of rock mass fractures as described above.

[0078] The processor can include one or more processing cores, such as a 4-core processor, an 8-core processor, and the like. The processor can be implemented in at least one of a hardware form of a digital signal processing (DSP), a field-programmable gate array (FPGA), a programmable logic array (PLA), and the like. The processor can also include a main processor and a coprocessor. The main processor is a processor for processing data in an awake state, also referred to as a central processing unit (CPU). The coprocessor is a low-power processor for processing data in a standby state. In some embodiments, the processor can be integrated with a graphics processor (GPU) for rendering and drawing content required to be displayed on the display screen. In some embodiments, the processor can further include an artificial intelligence (AI) processor for processing machine learning-related computing operations.

[0079] The memory can include one or more computer-readable storage media that can be non-transitory. The memory can also include a high-speed random access memory, and a nonvolatile memory such as one or more disk storage devices, flash storage devices. In the present embodiment, the memory is at least used to store the following computer programs, wherein the computer programs are loaded and executed by the processor, and can implement the related steps of the method for determining the modal and vibration characteristics of the composite tail shaft of the helicopter. In addition, the resources stored by the memory can also include an operating system and data, and the storage mode can be temporary storage or permanent storage. The operating system can include Windows, Unix, Linux, and the like. The data can include, but is not limited to, related data of the composite tail shaft of the helicopter, and the like.

[0080] The embodiment also provides a computer readable storage medium, and the computer readable storage medium stores a computer program. The computer program is executed by a processor to implement the steps of the rock mass crack intelligent identification and digital reconstruction method. It can be understood that if the method is implemented in the form of a software function unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application, essentially or the part that contributes to the prior art, or all or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and executes all or part of the steps of the method described in each embodiment of the present application. The foregoing storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0081] The technical features of the above embodiments can be combined in any manner. To make the description concise, all possible combinations of the technical features in the above embodiments are not described, but as long as the combinations of the technical features do not contradict, they should be considered as falling within the scope of the present disclosure.

[0082] The above embodiments only express several implementation manners of the present application, and the description is specific and detailed, but it should not be understood as a limitation on the scope of the application. It should be pointed out that for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which all belong to the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A rock mass fracture intelligent identification and digital reconstruction method, characterized in that, Comprise the following steps: S10, collect the rock mass fracture image; S20, select a plurality of spatial feature points in the real rock mass surface and the rock mass fracture image, add a mark color block to the spatial feature point position in the rock mass fracture image in an artificial marking manner, obtain the coordinate value of the mark color block in the physical space, record the color corresponding to the hexadecimal code of the mark color block, and pair the image coordinates with the actual physical coordinates for recording; S30, based on transfer learning, introduce the labeled medical image cell membrane sample and the road crack sample, and jointly train the crack identification model; S40, input the rock mass fracture image into the crack identification model to segment the crack, and build a snake convolution network to extract multi-scale features of the image; The snake convolution network explicitly encodes the edge direction and width change of the crack in the feature extraction stage by using a learnable sampling grid extending along the curve and additionally predicting the sampling offset, outputs a crack probability map for each pixel, and the probability map is mapped by Sigmoid and thresholded to obtain a final crack binary image; S50, digitally reconstruct the binary image to obtain a box set and a polyline set data; The binary image of the crack identification result is converted into a structured data, and the crack pixels are converted into a plurality of box data; In S50, the crack region is refined, and the current pixel is , and the eight-domain pixels are , and the pixel value of 0 represents the background or 1 represents the foreground, , which is located above, and the other pixels are numbered in a clockwise direction, and the number of domain foreground pixels is defined as follows: ； wherein, the value of R1 is in the range of 0 to 8; Defining the number of jumps from background pixels to foreground pixels in a field The specific formula is: ； wherein ; In the pixel deletion process, each iteration includes two sub-steps, in the first sub-step, when a pixel It is marked as deletable when it simultaneously satisfies the following conditions: ； In a second sub-step, when the pixel It is marked as deletable when the following conditions are met simultaneously: ； In each iteration process, first, the condition judgment shown in the first sub-step formula is executed, and the pixels meeting the condition are deleted; then, the condition judgment shown in the second sub-step formula is executed, and the pixels meeting the condition are deleted; the above iteration is repeated until there is no new pixel to be deleted, thereby obtaining a single-pixel-width line, and connecting the thinning result into a plurality of polyline data according to the pixel adjacency relationship, reflecting the spatial path and geometric morphology of the crack; S60, based on the color block calibration method, the corresponding relationship between the pixels and the physical space is established, a preset calibration color block in the image is selected as a reference area, a pixel-physical scale conversion factor is calculated by the known actual size of the mark color block and the pixel length in the image, a mapping model between the pixels and the physical space coordinates is constructed combined with the actual physical coordinates of the mark color block, and the restoration of the image scale to the actual physical scale is realized; When the color block calibration method is used, the actual physical distance between the mark color blocks is divided by the pixel distance between the mark color blocks to obtain the scale of the pixels to the physical world, the pixel coordinates of the crack in the binary image are multiplied by the scale of the pixels to the physical world to obtain the physical world scale crack coordinates, and the physical world scale crack coordinates are multiplied by the corresponding rotation matrix to obtain the physical world crack coordinates.

2. The method according to claim 1, wherein, In S10, after the rock mass fracture image is collected by the high-resolution camera, distortion correction, brightness adjustment and noise suppression preprocessing are performed.

3. The method according to claim 1, characterized in that, In S30, the labeled medical image cell membrane sample and the road crack sample are introduced to jointly train the crack identification model, and the sample is represented as: ； wherein, denotes a joint training sample, denotes a fissure dataset sample, denotes a medical image cell membrane sample.

Images