A method and system for identifying and extracting three-dimensional soil fissures considering vegetation shielding

Abstract

The present application relates to the technical field of soil engineering monitoring, and discloses a method and system for identifying and extracting three-dimensional fissures of soil considering vegetation shielding, which acquires three-dimensional point cloud and multispectral image of soil, and completes preprocessing; identifies the vegetation area and generates a mask, removes the vegetation area through the mask, and obtains soil surface images without vegetation interference; performs spatial registration on the soil surface images without vegetation interference and the three-dimensional point cloud of soil, and obtains registered point cloud; constructs a three-dimensional voxel field, performs three-dimensional rough segmentation of fissures, infers fissures and completes fractures in the blind area of vegetation shielding in combination with the spatial characteristics of multispectral images, and obtains three-dimensional initial area of fissures; performs topological learning on the fissure point cloud in the three-dimensional initial area of fissures, and generates three-dimensional fissure distribution; calculates and outputs indexes including fissure length, width, depth, volume, volume rate and spatial inclination based on the three-dimensional fissure distribution, and integrates to obtain the three-dimensional fissure identification result of soil; and the present application improves the fissure identification efficiency and precision.

Description

Technical Field

[0001] This invention relates to the field of soil engineering monitoring technology, specifically to a method and system for identifying and extracting three-dimensional soil cracks considering vegetation obstruction. Background Technology

[0002] In geotechnical engineering and geological disaster monitoring, soil fissures are a key indicator of soil stability. In vegetation-covered scenarios, the shading of vegetation stems and leaves creates blind spots in the image and point cloud, making it impossible to fully acquire information about the fissure area. Changes in natural lighting cause unstable imaging quality and blurred fissure edges. Traditional models struggle to accurately separate vegetation from soil, easily misidentifying vegetation as fissures or masking fissures. Existing methods mostly focus on two-dimensional feature extraction, failing to acquire three-dimensional indicators such as depth, volume, and spatial distribution. The spatial registration accuracy between point clouds and images is low, resulting in large errors in three-dimensional reconstruction, making it difficult to meet the needs of refined engineering monitoring. Summary of the Invention

[0003] The purpose of this invention is to solve the above-mentioned problems by designing a method and system for identifying and extracting three-dimensional soil cracks that takes into account vegetation obstruction.

[0004] The first aspect of this invention provides a method for identifying and extracting three-dimensional soil cracks considering vegetation obstruction, the method comprising the following steps: The three-dimensional point cloud and multispectral image of the soil are acquired, and preprocessing is completed through illumination adjustment, noise reduction, distortion correction, timestamp alignment and coordinate unification. The preprocessed multispectral image is input into the ME Mask R-CNN model to identify vegetation areas and generate a mask. The vegetation areas are then removed using the mask to obtain a soil surface image without vegetation interference. The ICP algorithm and RANSAC algorithm are combined to perform spatial registration of soil surface images and 3D point clouds of soil without vegetation interference, and the registered point cloud is obtained. After registration, the point cloud is constructed into a three-dimensional voxel field and input into the 3D-UNet model for coarse three-dimensional segmentation of the crack. Combined with multispectral spatial features, crack inference and fracture completion are performed in the vegetation shading blind area to obtain the initial three-dimensional crack region. The DGCNN model is used to perform topological learning on the fracture point cloud in the initial three-dimensional region of the fracture to construct the spatial relationship between nodes, edges and surfaces. The fracture wall morphology is optimized by B-spline surface fitting to generate a three-dimensional fracture distribution. Based on the calculation of three-dimensional crack distribution, the system outputs indicators including crack length, width, depth, volume, volume fraction, and spatial dip angle, and integrates them to obtain the three-dimensional crack identification results of the soil.

[0005] Optionally, in the first implementation of the first aspect of the present invention, the preprocessing of acquiring the three-dimensional point cloud and multispectral image of the soil, including illumination adjustment, noise reduction, distortion correction, timestamp alignment and coordinate unification, includes: Outlier filtering and noise smoothing are performed on the 3D point cloud of soil, and salt-and-pepper noise and Gaussian noise removal are performed on the multispectral image. Lens distortion correction is performed on the multispectral image according to the camera calibration parameters, and the processed 3D point cloud of soil is aligned with the processed multispectral image according to the acquisition timestamp. Using the scanning platform's reference coordinate system, the aligned 3D point cloud of soil and the multispectral image are subjected to a unified coordinate transformation to complete the preprocessing operation.

[0006] Optionally, in a second implementation of the first aspect of the present invention, the step of inputting the preprocessed multispectral image into the ME Mask R-CNN model, identifying vegetation areas and generating a mask, and removing vegetation areas through the mask to obtain a soil surface image without vegetation interference includes: The preprocessed multispectral image is input into the ME Mask R-CNN model, and the backbone network of the ME Mask R-CNN model extracts multi-scale features of the image to generate an image feature map. The ME Mask R-CNN model is used to generate candidate regions through its region proposal network. Potential target regions, including vegetation, are then selected to obtain the candidate regions. Candidate regions are classified and masked for prediction, and vegetation region segmentation mask is output. Based on the vegetation segmentation mask, a mask removal operation is performed on the multispectral image to remove vegetation region pixels, resulting in a soil surface image without vegetation interference.

[0007] Optionally, in the third implementation of the first aspect of the present invention, the step of combining the ICP algorithm and the RANSAC algorithm to spatially register the soil surface image without vegetation disturbance and the three-dimensional point cloud of the soil to obtain the registered point cloud includes: First, extract feature points from the soil surface image without vegetation interference, and then extract the three-dimensional feature points from the corresponding three-dimensional point cloud. Using the unified coordinate system as the initial value, an initial correspondence between image feature points and 3D point cloud feature points is established to obtain initial corresponding point pairs; The RANSAC algorithm is used to filter the initial corresponding point pairs, and erroneous matching point pairs are removed. The filtered matching point pairs are then input into the ICP algorithm to iteratively calculate the rotation matrix and translation vector between the point cloud and the image, thereby obtaining the optimal rotation and translation parameters. Perform coordinate transformation on the 3D point cloud according to the optimal rotation and translation parameters, and perform spatial registration to obtain the registered point cloud.

[0008] Optionally, in the fourth implementation of the first aspect of the present invention, the step of constructing a three-dimensional voxel field from the registered point cloud and inputting it into a 3D-UNet model for coarse three-dimensional segmentation of the crack, and combining multispectral spatial features to infer cracks and fill in fractures in the vegetation-shaded blind areas to obtain the initial three-dimensional crack region, includes: The registered point cloud is divided into three-dimensional voxels according to spatial scale to construct a three-dimensional voxel field of soil. The three-dimensional voxel field is input into the 3D-UNet model. The deep features of the voxels are extracted through the encoding network and the three-dimensional spatial scale is recovered through the decoding network. The crack probability voxels are output to determine the preliminary distribution range of the cracks. Spatial features of multispectral images are extracted, cracks are inferred in the locations of vegetation-covered blind spots, and point cloud completion and morphological extension are performed on the crack fracture locations to obtain the initial three-dimensional region of the crack.

[0009] Optionally, in the fifth implementation of the first aspect of the present invention, the step of using a DGCNN model to perform topological learning on the crack point cloud in the initial three-dimensional region of the crack, constructing the spatial relationships of nodes, edges, and surfaces, and optimizing the crack wall morphology through B-spline surface fitting to generate a three-dimensional crack distribution includes: The point cloud within the initial 3D region of the crack is input into the DGCNN model to construct the local neighborhood relationship of the point cloud. The global topological structure of the fracture point cloud is learned through the DGCNN model to determine the spatial relationships between nodes, edges and faces. The fracture point cloud is grouped according to its topological structure to distinguish the main body, branches and edge locations of the fractures. B-spline surface fitting is performed on the grouped fracture point cloud to smooth the fracture wall morphology, fill voids and missing areas, remove redundant points and pseudo-structures after fitting, and generate a three-dimensional fracture distribution.

[0010] Optionally, in the sixth implementation of the first aspect of the present invention, the step of calculating and outputting indices including crack length, width, depth, volume, volume fraction, and spatial dip angle based on the three-dimensional crack distribution, and integrating these indices to obtain the three-dimensional crack identification result of the soil, includes: Perform three-dimensional connected domain analysis on the three-dimensional fracture distribution, divide it into independent fracture elements and label them with numbers; Three-dimensional skeleton extraction is performed on each independent fracture element to obtain the fracture centerline and branch node positions; Calculate the cross-sectional width at each location along the normal to the centerline of the fracture, and calculate the fracture depth at each location along the depth direction; Calculate the volume of a single fracture and the total volume fraction of the region based on the three-dimensional distribution of fractures, and calculate the spatial orientation and dip angle of the fractures. By integrating indicators including crack length, width, depth, volume, volume fraction, and spatial dip angle, a three-dimensional crack identification result for soil is formed.

[0011] A second aspect of the present invention provides a three-dimensional soil crack identification and extraction system that takes into account vegetation obstruction, the system comprising: The preprocessing module is used to acquire 3D point cloud and multispectral images of soil, and completes the preprocessing by adjusting illumination, denoising, distortion correction, timestamp alignment and coordinate unification. The recognition module is used to input the preprocessed multispectral image into the ME Mask R-CNN model, identify the vegetation area and generate a mask, and remove the vegetation area through the mask to obtain a soil surface image without vegetation interference. The spatial registration module is used to perform spatial registration of soil surface images and 3D point clouds of soil without vegetation disturbance by combining the ICP algorithm and the RANSAC algorithm, so as to obtain the registered point cloud. The completion module is used to construct a three-dimensional voxel field from the registered point cloud and input it into the 3D-UNet model for coarse three-dimensional segmentation of the crack. It combines multispectral spatial features to infer cracks and complete the fracture in the vegetation shading blind area to obtain the initial three-dimensional crack region. The fitting and optimization module is used to perform topological learning on the crack point cloud in the initial three-dimensional region of the crack using the DGCNN model, construct the spatial relationship between nodes, edges and surfaces, optimize the crack wall morphology through B-spline surface fitting, and generate a three-dimensional crack distribution. The calculation module is used to calculate and output indicators including crack length, width, depth, volume, volume fraction, and spatial dip angle based on the three-dimensional crack distribution, and integrate them to obtain the three-dimensional crack identification results of the soil.

[0012] A third aspect of the present invention provides a soil three-dimensional crack identification and extraction device that takes into account vegetation shading. The soil three-dimensional crack identification and extraction device that takes into account vegetation shading includes a memory and at least one processor. The memory stores instructions. The at least one processor invokes the instructions in the memory to cause the soil three-dimensional crack identification and extraction device that takes into account vegetation shading to perform the steps of the soil three-dimensional crack identification and extraction method that takes into account vegetation shading as described in any of the preceding claims.

[0013] A fourth aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed by a processor, implement the steps of the method for identifying and extracting three-dimensional soil fissures considering vegetation shading as described in any of the preceding claims.

[0014] The technical solution provided by this invention involves acquiring a three-dimensional point cloud and multispectral image of the soil, performing preprocessing such as illumination adjustment, denoising, distortion correction, timestamp alignment, and coordinate unification; inputting the preprocessed multispectral image into the ME Mask R-CNN model to identify vegetation areas and generate a mask; removing vegetation areas using the mask to obtain a soil surface image free from vegetation interference; using a combination of ICP and RANSAC algorithms to spatially register the soil surface image free from vegetation interference and the three-dimensional point cloud of the soil to obtain a registered point cloud; constructing a three-dimensional voxel field from the registered point cloud and inputting it into a 3D-UNet model for coarse three-dimensional segmentation of cracks; combining multispectral spatial features to infer cracks and complete fracture edges in vegetation-occluded blind areas to obtain the initial three-dimensional crack region; and employing... The DGCNN model performs topological learning on the crack point cloud in the initial 3D region of the crack, constructs the spatial relationships between nodes, edges, and surfaces, and optimizes the crack wall morphology through B-spline surface fitting to generate a 3D crack distribution. Based on the 3D crack distribution, it calculates and outputs indices including crack length, width, depth, volume, volume fraction, and spatial tilt angle, and integrates them to obtain the 3D crack identification results of the soil. This invention effectively solves the problems of vegetation shading and light interference, realizes fully automatic extraction of 3D features of soil cracks, and improves crack identification efficiency, completeness, and 3D measurement accuracy. Attached Figure Description

[0015] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.

[0016] Figure 1 A flowchart of a method for identifying and extracting three-dimensional soil fissures considering vegetation obstruction, provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a three-dimensional soil crack identification and extraction system considering vegetation obstruction provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a soil three-dimensional crack identification and extraction device that takes into account vegetation obstruction, provided in an embodiment of the present invention. Detailed Implementation

[0017] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, apparatus, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0018] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 The flowchart of the method for identifying and extracting three-dimensional soil fissures considering vegetation obstruction provided in this embodiment of the invention includes the following steps: Step 101: Acquire the three-dimensional point cloud and multispectral image of the soil, and complete the preprocessing by adjusting the illumination, denoising, correcting distortion, aligning the timestamps and unifying the coordinates; In this embodiment, outlier filtering and noise smoothing are performed on the 3D point cloud of the soil, and salt-and-pepper noise and Gaussian noise removal are performed on the multispectral image. Lens distortion correction is performed on the multispectral image according to the camera calibration parameters, and the processed 3D point cloud of the soil and the processed multispectral image are aligned one-to-one according to the acquisition timestamp. Using the scanning platform's reference coordinate system, the aligned 3D point cloud of the soil and the multispectral image are uniformly transformed to complete the preprocessing operation.

[0019] Step 102: Input the preprocessed multispectral image into the ME Mask R-CNN model to identify vegetation areas and generate a mask. Remove vegetation areas through the mask to obtain a soil surface image without vegetation interference. In this embodiment, the preprocessed multispectral image is input into the ME Mask R-CNN model. The backbone network of the ME Mask R-CNN model extracts multi-scale features of the image to generate an image feature map. The region proposal network of the ME Mask R-CNN model is used to generate candidate regions, and potential target regions including vegetation are screened to obtain candidate regions. The candidate regions are classified and masked for prediction, and vegetation region segmentation masks are output. The multispectral image is then masked for removal based on the vegetation segmentation masks to remove pixels in the vegetation region, resulting in a soil surface image without vegetation interference.

[0020] In this embodiment, for soil surfaces not covered by vegetation, multispectral images can provide richer texture and humidity information than ordinary RGB images. Cracks are usually accompanied by changes in soil moisture or mineral exposure, which helps the ME Mask R-CNN model to more accurately identify the crack boundaries in the exposed areas. For blind areas completely covered by vegetation, the principle of spatial continuity is used to analyze the crack direction and texture extension trend around the covered area. Combined with the known crack network topology, logical inference and break-point completion are performed on the cracks in the blind area.

[0021] In this embodiment, the region proposal network of the ME Mask R-CNN model generates candidate regions mainly based on the texture gradient, edge features, and multispectral band response of the image to lock potential targets, such as the high reflectivity of vegetation in the near-infrared band. After obtaining candidate regions, classification and mask prediction are performed. Based on the vegetation feature library learned during model training, including color space distribution, shape regularity, and contextual information, only those regions whose feature vectors highly match the vegetation category will be finally selected as vegetation regions and masks will be generated.

[0022] Step 103: Using the combination of ICP and RANSAC algorithms, spatial registration is performed on the soil surface image and the three-dimensional point cloud of the soil without vegetation interference to obtain the registered point cloud. In this embodiment, feature points of the soil surface image without vegetation interference are first extracted, and then three-dimensional feature points of the corresponding three-dimensional point cloud are extracted. Using the unified coordinate system as the initial value, an initial correspondence between the image feature points and the three-dimensional point cloud feature points is established to obtain initial corresponding point pairs. The RANSAC algorithm is used to filter the initial corresponding point pairs, and incorrect matching point pairs are removed. The filtered matching point pairs are input into the ICP algorithm to iteratively calculate the rotation matrix and translation vector between the point cloud and the image to obtain the optimal rotation and translation parameters. Based on the optimal rotation and translation parameters, coordinate transformation is performed on the three-dimensional point cloud to perform spatial registration and obtain the registered point cloud.

[0023] In this embodiment, a two-dimensional feature point detection algorithm is activated on the soil surface image without vegetation interference obtained through vegetation removal and image preprocessing. The algorithm iterates through the image pixel information and, based on texture gradient, corner response, and edge features, locates and extracts two-dimensional feature points that can stably characterize the soil surface structure. Simultaneously, in the three-dimensional point cloud data of the corresponding monitoring area, three-dimensional point cloud feature points with the same physical location as the two-dimensional feature points are extracted according to spatial location mapping relationships. Using the global measurement coordinate system after previous coordinate unification as the initial reference, the extracted two-dimensional feature points of the soil surface image and the three-dimensional feature points of the three-dimensional point cloud are associated and paired according to spatial location association rules and the calibration parameters of the acquisition equipment to establish a preliminary correspondence, forming initial corresponding point pairs containing multiple sets of candidate matching relationships. The RANSAC random sampling consensus algorithm is used to screen the initial corresponding point pairs round by round. A hypothesis model is constructed through random sampling and consistency verification is performed. Incorrect matching point pairs caused by noise interference, positioning deviation, or local occlusion are identified and eliminated, while highly reliable matching point pairs that meet spatial geometric constraints are retained. The screened reliable matching point pairs are then input into the ICP (Internet Protocol version). The iterative nearest-point algorithm iteratively optimizes the calculations multiple times to solve for the rotation matrix and translation vector between the 3D point cloud and the 2D image until the algorithm converges, obtaining the optimal rotation and translation parameters that meet the registration accuracy requirements. Based on the optimal rotation and translation parameters obtained through iterative calculation, a unified coordinate transformation operation is performed on the original 3D point cloud data. The spatial coordinates of all point clouds are adjusted according to the rotation matrix and translation vector, so that the 3D point cloud and the soil surface image are completely aligned in the same coordinate system. This eliminates the spatial offset and angular deviation between multi-source data, completes high-precision spatial registration processing, and obtains a registered point cloud that can be directly used for 3D topology reconstruction.

[0024] Step 104: After registration, the point cloud is constructed into a three-dimensional voxel field and input into the 3D-UNet model for coarse three-dimensional segmentation of the crack. Combined with multispectral spatial features, crack inference and fracture completion are performed on the vegetation shading blind area to obtain the initial three-dimensional crack region. In this embodiment, the registered point cloud is divided into three-dimensional voxels according to the spatial scale to construct a three-dimensional voxel field of soil. The three-dimensional voxel field is input into the 3D-UNet model, and the deep features of the voxels are extracted through the encoding network. The three-dimensional spatial scale is restored through the decoding network, and the crack probability voxels are output to determine the initial distribution range of cracks. The spatial features of the multispectral image are extracted, and cracks are inferred in the location of the vegetation occlusion blind area. Point cloud completion and morphological extension are performed on the location of the crack fracture to obtain the initial three-dimensional region of cracks.

[0025] In this embodiment, the 3D-UNet model, through its encoder-decoder structure, is able to capture the deep spatial features of the voxel field and infer unknown regions based on the known distribution probability of the surrounding crack point cloud. It also uses the spatial features of multispectral images to infer blind areas, thus compensating for the lack of simple point cloud data.

[0026] In this embodiment, the 3D point cloud data, after high-precision spatial registration, is divided into grids according to a preset spatial scale and resolution standard. The continuously distributed point cloud data is discretized into regularly arranged 3D voxel units. Based on the number of points, elevation information, and spatial distribution characteristics contained in each voxel, a 3D voxel field that can completely characterize the surface structure of the soil to be measured is constructed. The constructed 3D voxel field of the soil is then input into a pre-trained 3D-UNet. The deep learning model first downsamples and abstracts the features of the 3D voxel data layer by layer through the model's encoding network, extracting deep semantic features that reflect the spatial morphology, structural differences, and distribution patterns of the cracks. Then, the model's decoding network upsamples and restores the dimensions of the abstract features, gradually restoring them to the original 3D spatial scale. It outputs the probability value of each 3D voxel belonging to the crack region, forming a crack probability voxel set. Based on the probability threshold, the initial distribution range of cracks on the soil surface is determined. Simultaneously, the spatial texture features, edge features, and gradient features of the multispectral image after vegetation removal are extracted. Combined with the spatial location information of the 3D voxel field, crack direction inference and missing area completion are performed for point cloud blind spots and crack missing locations caused by vegetation occlusion. Point cloud morphology completion and spatial extension processing are performed for crack breaks and discontinuities to eliminate recognition gaps caused by occlusion and data loss. Finally, a complete, continuous, and occluded initial 3D crack region is obtained.

[0027] Step 105: Use the DGCNN model to perform topological learning on the crack point cloud in the initial three-dimensional region of the crack, construct the spatial relationship between nodes, edges and surfaces, optimize the crack wall morphology by B-spline surface fitting, and generate a three-dimensional crack distribution. In this embodiment, the point cloud within the initial three-dimensional region of the fracture is input into the DGCNN model to construct the local neighborhood relationship of the point cloud; the global topological structure of the fracture point cloud is learned through the DGCNN model to determine the spatial relationship between nodes, edges and faces; the fracture point cloud is grouped according to the topological structure to distinguish the main body, branches and edge positions of the fracture; B-spline surface fitting is performed on the grouped fracture point cloud to smooth the fracture wall morphology, fill in voids and missing areas, remove redundant points and pseudo-structures after fitting, and generate a three-dimensional fracture distribution.

[0028] In this embodiment, B-spline fitting is performed after the DGCNN model completes topological learning, and the exact location and orientation of the crack are already known. The purpose of fitting is to smooth the crack wall morphology and fill in the tiny holes caused by sparse point cloud or noise. The original point cloud usually contains a lot of measurement noise, and direct calculation will result in an uneven crack wall. B-spline fitting approximates the real surface through mathematical functions, and retains the macroscopic geometric features of the crack while eliminating redundant noise.

[0029] Step 106: Calculate and output indices including crack length, width, depth, volume, volume fraction, and spatial dip angle based on the three-dimensional crack distribution, and integrate them to obtain the three-dimensional crack identification results of the soil.

[0030] In this embodiment, a three-dimensional connected domain analysis is performed on the three-dimensional fracture distribution to divide it into independent fracture units and label them with numbers; a three-dimensional skeleton is extracted for each independent fracture unit to obtain the fracture centerline and branch node positions; the cross-sectional width at each position is calculated along the fracture centerline normal, and the fracture depth at each position is calculated along the depth direction; the volume of a single fracture and the total volume fraction of the region are calculated based on the three-dimensional fracture distribution, and the spatial orientation and dip angle of the fracture are calculated; the indicators including fracture length, width, depth, volume, volume fraction, and spatial dip angle are integrated to form the three-dimensional fracture identification result of the soil.

[0031] Please see Figure 2 A schematic diagram of the structure of a three-dimensional soil crack identification and extraction system considering vegetation obstruction provided in this embodiment of the invention. The system includes: The preprocessing module is used to acquire 3D point cloud and multispectral images of soil, and completes the preprocessing by adjusting illumination, denoising, distortion correction, timestamp alignment and coordinate unification. The recognition module is used to input the preprocessed multispectral image into the ME Mask R-CNN model, identify the vegetation area and generate a mask, and remove the vegetation area through the mask to obtain a soil surface image without vegetation interference. The spatial registration module is used to perform spatial registration of soil surface images and 3D point clouds of soil without vegetation disturbance by combining the ICP algorithm and the RANSAC algorithm, so as to obtain the registered point cloud. The completion module is used to construct a three-dimensional voxel field from the registered point cloud and input it into the 3D-UNet model for coarse three-dimensional segmentation of the crack. It combines multispectral spatial features to infer cracks and complete the fracture in the vegetation shading blind area to obtain the initial three-dimensional crack region. The fitting and optimization module is used to perform topological learning on the crack point cloud in the initial three-dimensional region of the crack using the DGCNN model, construct the spatial relationship between nodes, edges and surfaces, optimize the crack wall morphology through B-spline surface fitting, and generate a three-dimensional crack distribution. The calculation module is used to calculate and output indicators including crack length, width, depth, volume, volume fraction, and spatial dip angle based on the three-dimensional crack distribution, and integrate them to obtain the three-dimensional crack identification results of the soil.

[0032] Figure 3 This is a schematic diagram of a soil three-dimensional crack identification and extraction device considering vegetation shading provided in an embodiment of the present invention. This soil three-dimensional crack identification and extraction device 300 considering vegetation shading can vary considerably due to different configurations or performance. It may include one or more central processing units (CPUs) 310 (e.g., one or more processors) and a memory 320, and one or more storage media 330 (e.g., one or more mass storage devices) storing application programs 333 or data 332. The memory 320 and storage media 330 can be temporary or persistent storage. The program stored in the storage media 330 may include one or more modules (not shown in the diagram), each module may include a series of instruction operations on the soil three-dimensional crack identification and extraction device 300 considering vegetation shading. Furthermore, the processor 310 may be configured to communicate with the storage media 330 and execute the series of instruction operations in the storage media 330 on the soil three-dimensional crack identification and extraction device 300 considering vegetation shading to implement the method provided in the above embodiment.

[0033] The soil three-dimensional crack identification and extraction device 300, which takes into account vegetation obstruction, may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input / output interfaces 360, and / or one or more operating systems 331, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 3 The structure of the soil three-dimensional crack identification and extraction device shown, taking into account vegetation obstruction, does not constitute a limitation on the computer device provided by the present invention. It may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0034] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the various steps of the soil three-dimensional crack identification and extraction method considering vegetation shading provided in the above embodiments.

[0035] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the above-described equipment or apparatus / unit can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0036] If the integrated unit is implemented as a software functional 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 solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0037] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended technical solutions and their equivalents.

Claims

1. A method for identifying and extracting three-dimensional soil cracks considering vegetation obstruction, characterized in that, The method includes the following steps: The three-dimensional point cloud and multispectral image of the soil are acquired, and preprocessing is completed through illumination adjustment, noise reduction, distortion correction, timestamp alignment and coordinate unification. The preprocessed multispectral image is input into the ME Mask R-CNN model to identify vegetation areas and generate a mask. The vegetation areas are then removed using the mask to obtain a soil surface image without vegetation interference. The ICP algorithm and RANSAC algorithm are combined to perform spatial registration of soil surface images and 3D point clouds of soil without vegetation interference, and the registered point cloud is obtained. After registration, the point cloud is constructed into a three-dimensional voxel field and input into the 3D-UNet model for coarse three-dimensional segmentation of the crack. Combined with multispectral spatial features, crack inference and fracture completion are performed in the vegetation shading blind area to obtain the initial three-dimensional crack region. The DGCNN model is used to perform topological learning on the fracture point cloud in the initial three-dimensional region of the fracture to construct the spatial relationship between nodes, edges and surfaces. The fracture wall morphology is optimized by B-spline surface fitting to generate a three-dimensional fracture distribution. Based on the calculation of three-dimensional crack distribution, the system outputs indicators including crack length, width, depth, volume, volume fraction, and spatial dip angle, and integrates them to obtain the three-dimensional crack identification results of the soil.

2. The method for identifying and extracting three-dimensional soil fissures considering vegetation obstruction as described in claim 1, characterized in that, The acquisition of 3D point cloud and multispectral images of the soil is preprocessed through illumination adjustment, noise reduction, distortion correction, timestamp alignment, and coordinate unification, including: Outlier filtering and noise smoothing are performed on the 3D point cloud of soil, and salt-and-pepper noise and Gaussian noise removal are performed on the multispectral image. Lens distortion correction is performed on the multispectral image based on the camera calibration parameters, and the processed 3D point cloud of soil is aligned with the processed multispectral image according to the acquisition timestamp. Using the scanning platform's reference coordinate system, the aligned 3D point cloud of soil and the multispectral image are subjected to a unified coordinate transformation to complete the preprocessing operation.

3. The method for identifying and extracting three-dimensional soil fissures considering vegetation obstruction as described in claim 1, characterized in that, The process of inputting the preprocessed multispectral image into the ME Mask R-CNN model to identify vegetation areas and generate a mask, and then removing vegetation areas through the mask to obtain a soil surface image free from vegetation interference, includes: The preprocessed multispectral image is input into the ME Mask R-CNN model, and the backbone network of the ME Mask R-CNN model extracts multi-scale features of the image to generate an image feature map. The ME Mask R-CNN model is used to generate candidate regions through its region proposal network. Potential target regions, including vegetation, are then selected to obtain the candidate regions. Candidate regions are classified and masked for prediction, and vegetation region segmentation mask is output. Based on the vegetation segmentation mask, a mask removal operation is performed on the multispectral image to remove vegetation region pixels, resulting in a soil surface image without vegetation interference.

4. The method for identifying and extracting three-dimensional soil fissures considering vegetation obstruction as described in claim 1, characterized in that, The method employs a combination of ICP and RANSAC algorithms to spatially register a soil surface image without vegetation disturbance and a 3D point cloud of soil, resulting in a registered point cloud, including: First, extract feature points from the soil surface image without vegetation interference, and then extract the three-dimensional feature points from the corresponding three-dimensional point cloud. Using the unified coordinate system as the initial value, an initial correspondence between image feature points and 3D point cloud feature points is established to obtain initial corresponding point pairs; The RANSAC algorithm is used to filter the initial corresponding point pairs, and erroneous matching point pairs are removed. The filtered matching point pairs are then input into the ICP algorithm to iteratively calculate the rotation matrix and translation vector between the point cloud and the image, thereby obtaining the optimal rotation and translation parameters. Perform coordinate transformation on the 3D point cloud according to the optimal rotation and translation parameters, and perform spatial registration to obtain the registered point cloud.

5. The method for identifying and extracting three-dimensional soil fissures considering vegetation obstruction as described in claim 1, characterized in that, The process involves constructing a three-dimensional voxel field from the registered point cloud and inputting it into a 3D-UNet model for coarse three-dimensional segmentation of the cracks. Multispectral spatial features are then used to infer cracks and complete fracture edges in vegetation-occupied blind areas, resulting in an initial three-dimensional crack region, including: The registered point cloud is divided into three-dimensional voxels according to spatial scale to construct a three-dimensional voxel field of soil. The three-dimensional voxel field is input into the 3D-UNet model. The deep features of the voxels are extracted through the encoding network and the three-dimensional spatial scale is recovered through the decoding network. The crack probability voxels are output to determine the preliminary distribution range of the cracks. Spatial features of multispectral images are extracted, cracks are inferred in the locations of vegetation-covered blind spots, and point cloud completion and morphological extension are performed on the crack fracture locations to obtain the initial three-dimensional region of the crack.

6. The method for identifying and extracting three-dimensional soil fissures considering vegetation obstruction as described in claim 1, characterized in that, The method employs a DGCNN model to perform topological learning on the crack point cloud in the initial 3D region of the crack, constructing the spatial relationships between nodes, edges, and surfaces. It then optimizes the crack wall morphology through B-spline surface fitting to generate a 3D crack distribution, including: The point cloud within the initial 3D region of the crack is input into the DGCNN model to construct the local neighborhood relationship of the point cloud. The global topological structure of the fracture point cloud is learned through the DGCNN model to determine the spatial relationships between nodes, edges and faces. The fracture point cloud is grouped according to its topological structure to distinguish the main body, branches and edge locations of the fractures. B-spline surface fitting is performed on the grouped fracture point cloud to smooth the fracture wall morphology, fill voids and missing areas, remove redundant points and pseudo-structures after fitting, and generate a three-dimensional fracture distribution.

7. The method for identifying and extracting three-dimensional soil fissures considering vegetation obstruction as described in claim 1, characterized in that, The method calculates and outputs indices including crack length, width, depth, volume, volume fraction, and spatial dip angle based on three-dimensional crack distribution, and integrates these to obtain the three-dimensional crack identification results of the soil, including: Perform three-dimensional connected domain analysis on the three-dimensional fracture distribution, divide it into independent fracture elements and label them with numbers; Three-dimensional skeleton extraction is performed on each independent fracture element to obtain the fracture centerline and branch node positions; Calculate the cross-sectional width at each location along the normal to the centerline of the fracture, and calculate the fracture depth at each location along the depth direction; Calculate the volume of a single fracture and the total volume fraction of the region based on the three-dimensional distribution of fractures, and calculate the spatial orientation and dip angle of the fractures. By integrating indicators including crack length, width, depth, volume, volume fraction, and spatial dip angle, a three-dimensional crack identification result for soil is formed.

8. A three-dimensional soil crack identification and extraction system considering vegetation shading, characterized in that, The system includes: The preprocessing module is used to acquire 3D point cloud and multispectral images of soil, and completes the preprocessing by adjusting illumination, denoising, distortion correction, timestamp alignment and coordinate unification. The recognition module is used to input the preprocessed multispectral image into the ME Mask R-CNN model, identify the vegetation area and generate a mask, and remove the vegetation area through the mask to obtain a soil surface image without vegetation interference. The spatial registration module is used to perform spatial registration of soil surface images and 3D point clouds of soil without vegetation disturbance by combining the ICP algorithm and the RANSAC algorithm, so as to obtain the registered point cloud. The completion module is used to construct a three-dimensional voxel field from the registered point cloud and input it into the 3D-UNet model for coarse three-dimensional segmentation of the crack. It combines multispectral spatial features to infer cracks and complete the fracture in the vegetation shading blind area to obtain the initial three-dimensional crack region. The fitting and optimization module is used to perform topological learning on the crack point cloud in the initial three-dimensional region of the crack using the DGCNN model, construct the spatial relationship between nodes, edges and surfaces, optimize the crack wall morphology through B-spline surface fitting, and generate a three-dimensional crack distribution. The calculation module is used to calculate and output indicators including crack length, width, depth, volume, volume fraction, and spatial dip angle based on the three-dimensional crack distribution, and integrate them to obtain the three-dimensional crack identification results of the soil.

9. A soil three-dimensional crack identification and extraction device considering vegetation obstruction, characterized in that, The soil three-dimensional crack identification and extraction device considering vegetation shading includes a memory and at least one processor, wherein the memory stores instructions; the at least one processor invokes the instructions in the memory to cause the soil three-dimensional crack identification and extraction device considering vegetation shading to perform the various steps of the soil three-dimensional crack identification and extraction method considering vegetation shading as described in any one of claims 1-7.

10. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement the various steps of the three-dimensional soil crack identification and extraction method considering vegetation shading as described in any one of claims 1-7.

Images