Oil surface crack identification method and system based on three-dimensional laser scanning

Abstract

This invention discloses a method and system for identifying surface cracks in oil-worked surfaces based on three-dimensional laser scanning, relating to the field of non-destructive testing. The method includes: performing three-dimensional laser scanning on the oil-worked surface to acquire surface point cloud data; performing denoising, resampling, and registration to establish a three-dimensional point cloud model of the surface; pre-segmenting cracks based on process feature data to establish a potential crack region model; performing probability detection based on multi-dimensional crack types to establish a crack type probability detection space; performing reliability crack classification on the potential crack region model to obtain a crack identification classification space; performing quantitative evaluation to obtain a crack classification quantization space; and mapping and labeling the point cloud model to establish a surface crack distribution model. This application solves the problems of low accuracy, easy omissions and misjudgments, and low efficiency in existing oil-worked surface crack identification methods, which cannot adapt to the needs of large-area, refined crack identification. It achieves the effect of improving identification accuracy, avoiding omissions and misjudgments, and improving efficiency to meet the needs of large-area, refined crack identification.

Description

Technical Field

[0001] This application relates to the field of nondestructive testing, and in particular to a method and system for identifying cracks on oil-coated surfaces based on three-dimensional laser scanning. Background Technology

[0002] As core decorative and protective components of traditional architecture, the cracks and defects on the surfaces of ancient building plasterwork directly affect the structural stability of the components and the overall cultural relic value of the building. Accurate identification and effective control of cracks on plasterwork surfaces are crucial aspects of the protection and restoration of ancient buildings. Currently, the industry mainly relies on two conventional methods for identifying cracks on plasterwork surfaces: manual visual inspection and conventional two-dimensional image analysis. Manual inspection involves professionals observing the surface with their naked eyes and measuring with handheld tools to examine the morphology and distribution of cracks. Two-dimensional image analysis uses ordinary cameras to capture images of the plasterwork surface and relies on basic image software for preliminary identification and labeling of cracks. Existing technologies have significant shortcomings. Manual inspection is affected by factors such as personnel experience, visual fatigue, and ambient lighting, easily leading to missed or misjudged cracks, and its low efficiency makes it unsuitable for the batch inspection needs of large-area plasterwork components in ancient buildings. Conventional two-dimensional image analysis can only obtain planar information of the plasterwork surface and cannot reconstruct the three-dimensional morphology of the component, making it difficult to accurately identify deep and hidden cracks, and thus failing to meet the actual needs for refined identification of cracks in ancient building plasterwork.

[0003] At present, the identification of surface cracks in ancient building paint has technical problems such as low identification accuracy, easy to miss or misjudge, and low detection efficiency, which cannot meet the needs of large-area, fine and hidden crack identification. Summary of the Invention

[0004] This application provides a method and system for identifying cracks on oil-work surfaces based on 3D laser scanning. It acquires point cloud data of the oil-work surface through 3D laser scanning, and constructs a 3D point cloud model of the oil-work surface after denoising, resampling, and registration processing. Potential crack regions are pre-segmented based on oil-work process characteristics. Multi-type probability detection is performed for various crack types, including hairline cracks, network cracks, through cracks, and hollow cracks, and a probability detection space is constructed. Based on this space, crack reliability is classified using neighborhood constraints and weighted discrimination to obtain an identification classification space. Crack levels are quantitatively evaluated based on the classification space, and the quantification results are mapped and labeled onto the 3D point cloud model, forming a complete crack distribution model of the oil-work surface. These technical means solve the technical problems of low identification accuracy, easy omissions and misjudgments, and low detection efficiency in existing ancient building oil-work surface crack identification methods, which are unable to meet the needs of large-area, refined, and hidden crack identification. This method achieves the technical effects of improving crack identification accuracy, avoiding omissions and misjudgments, increasing detection efficiency, and adapting to the needs of large-area, refined, and hidden crack identification.

[0005] This application provides a method for identifying cracks on oil-work surfaces based on three-dimensional laser scanning, comprising: performing three-dimensional laser scanning on the oil-work surface to acquire point cloud data of the oil-work surface, and performing denoising, resampling, and registration on the point cloud data of the oil-work surface to establish a three-dimensional point cloud model of the oil-work surface; performing oil-work crack pre-segmentation on the three-dimensional point cloud model of the oil-work surface according to oil-work process feature data to establish a potential crack region model; performing multi-type probability detection on the potential crack region model according to multi-dimensional crack types to establish a crack type probability detection space; performing reliability crack classification on the potential crack region model according to the crack type probability detection space to obtain a crack identification classification space; performing quantitative evaluation based on the crack identification classification space to obtain an oil-work crack classification quantization space, and mapping and labeling the three-dimensional point cloud model of the oil-work surface according to the oil-work crack classification quantization space to establish an oil-work surface crack distribution model.

[0006] In a possible implementation, the oilfield surface point cloud data is denoised, resampled, and registered to establish a three-dimensional point cloud model of the oilfield surface. The following processing is performed: adaptive denoising based on local neighborhood statistical distribution characteristics is applied to the oilfield surface point cloud data to obtain denoised oilfield point cloud data; non-uniform resampling is performed on the denoised oilfield point cloud data to obtain updated oilfield point cloud data, and a point cloud feature descriptor is constructed based on the updated oilfield point cloud data; initial alignment is performed on the updated oilfield point cloud data based on the point cloud feature descriptor to obtain an initial registration model of the oilfield point cloud; multi-point registration error correction is performed on the initial registration model of the oilfield point cloud to generate the three-dimensional point cloud model of the oilfield surface.

[0007] In a possible implementation, the three-dimensional point cloud model of the oil-worked surface is pre-segmented for oil-worked cracks based on the oil-worked process feature data to establish a potential crack region model. The following processing is then performed: Pattern distribution features, coating layer boundary features, and component partitioning features are extracted based on the oil-worked process feature data; pattern region identification and masking processing is performed on the three-dimensional point cloud model of the oil-worked surface based on the pattern distribution features to obtain a pattern-masked point cloud model; spatial partitioning constraints are applied to the pattern-masked point cloud model based on the component partitioning features to obtain a component partitioning point cloud model; hierarchical partitioning constraints are applied to the component partitioning point cloud model based on the coating layer boundary features to determine crack-sensitive areas; local geometric analysis of the point cloud is performed on the crack-sensitive areas to determine curvature abrupt change features and height gradient anomaly features; joint judgment is performed based on the curvature abrupt change features and the height gradient anomaly features to extract candidate crack regions; spatial continuity constraints are applied to the candidate crack regions to filter them, generating the potential crack region model.

[0008] In a possible implementation, multi-type probability detection is performed on the potential fracture region model based on the multi-dimensional fracture type to establish a fracture type probability detection space. The following processing is then performed: Based on the multi-dimensional fracture type, a sample set of oil-based hairline fractures, a sample set of oil-based network fractures, a sample set of oil-based through fractures, and a sample set of oil-based hollow fractures are loaded; the potential fracture region model is probabilistically detected based on the oil-based hairline fracture sample set to obtain the hairline fracture probability at each point; the potential fracture region model is probabilistically detected based on the oil-based network fracture sample set to obtain the network fracture probability at each point; the potential fracture region model is probabilistically detected based on the oil-based through fracture sample set to obtain the through fracture probability at each point; the potential fracture region model is probabilistically detected based on the oil-based hollow fracture sample set to obtain the hollow fracture probability at each point; the probabilistics of hairline fractures, network fractures, through fractures, and hollow fractures at each point are then organized to generate the fracture type probability detection space.

[0009] In a possible implementation, the potential fracture region model is probabilistically detected based on the oil-based hairline fracture sample set to obtain the hairline fracture probability at each point, and the following processing is performed: multi-point feature recognition is performed on the potential fracture region model to obtain the nth point feature vector of the nth potential point, where n is a positive integer; feature recognition is performed on the oil-based hairline fracture sample set to obtain each hairline fracture sample vector; similarity evaluation is performed on the nth point feature vector and each hairline fracture sample vector based on a Siamese neural network to obtain the nth hairline fracture similarity set; the maximum value is filtered from the nth hairline fracture similarity set to generate the hairline fracture probability at the nth point.

[0010] In a possible implementation, the potential fracture region model is classified for reliability fractures based on the fracture type probability detection space to obtain a fracture identification classification space. The following processes are then performed: an nth point probability detection sequence is extracted from the fracture type probability detection space; a superiority test is performed on the nth point probability detection sequence based on a predetermined probability value to obtain an nth superior probability detection sequence; if the nth superior probability detection sequence is a single value, the nth dominant fracture type is determined, and the nth dominant fracture type is output as the nth fracture identification classification result; if the nth superior probability detection sequence is not a single value, a weighted discrimination based on spatial neighborhood consistency constraints is performed on the nth superior probability detection sequence to generate the nth fracture identification classification result; and the nth fracture identification classification result is added to the fracture identification classification space.

[0011] In a possible implementation, a quantitative evaluation is performed based on the fracture identification and classification space to obtain an oilfield fracture classification quantization space. The following processing is then performed: fracture features are captured from the potential fracture region model according to the fracture identification and classification space to obtain each fracture feature vector; a fracture level evaluation network is constructed; each fracture feature vector is input into the fracture level evaluation network to obtain each fracture level; and each fracture level is mapped to the fracture identification and classification space to generate the oilfield fracture classification quantization space.

[0012] In a possible implementation, the following processing is performed: the multidimensional fracture types include hairline fractures, network fractures, through fractures, and hollow fractures.

[0013] In a possible implementation, the following process is performed: generating multi-level crack early warning signals based on the oil surface crack distribution model.

[0014] This application also provides a three-dimensional laser scanning-based oil surface crack identification system, comprising: a three-dimensional point cloud model establishment module for oil surface, used to perform three-dimensional laser scanning on the oil surface, acquire oil surface point cloud data, and perform denoising, resampling, and registration on the oil surface point cloud data to establish a three-dimensional point cloud model of the oil surface; an oil surface crack pre-segmentation module, used to pre-segment oil surface cracks on the three-dimensional point cloud model of the oil surface according to oil process feature data to establish a potential crack region model; a multi-type probability detection module, used to perform multi-type probability detection on the potential crack region model according to multi-dimensional crack types to establish a crack type probability detection space; a reliability crack classification module, used to perform reliability crack classification on the potential crack region model according to the crack type probability detection space to obtain a crack identification classification space; and an oil surface crack distribution model establishment module, used to perform quantitative evaluation based on the crack identification classification space to obtain an oil surface crack classification quantization space, and to map and label the three-dimensional point cloud model of the oil surface according to the oil surface crack classification quantization space to establish an oil surface crack distribution model.

[0015] The proposed method and system for identifying cracks on oil-work surfaces based on three-dimensional laser scanning involves: first, performing three-dimensional laser scanning on the oil-work surface to acquire point cloud data; then, denoising, resampling, and registering the point cloud data to establish a three-dimensional point cloud model of the oil-work surface; next, pre-segmenting the oil-work surface cracks on the three-dimensional point cloud model based on oil-work process characteristic data to establish a potential crack region model; then, performing multi-type probability detection on the potential crack region model based on multi-dimensional crack types to establish a crack type probability detection space; then, classifying the potential crack region model based on the crack type probability detection space to obtain a crack identification classification space; finally, performing quantitative evaluation based on the crack identification classification space to obtain an oil-work crack classification quantization space; and finally, mapping and labeling the three-dimensional point cloud model of the oil-work surface based on the oil-work crack classification quantization space to establish an oil-work surface crack distribution model. Through the above process, the method and system proposed in this application achieve the technical effects of improving the accuracy of crack identification, avoiding missed detections and misjudgments, improving detection efficiency, and adapting to the needs of large-area, refined and hidden crack identification. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments of the present invention will be briefly described below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0017] Figure 1 This is a flowchart illustrating the method for identifying surface cracks in oil based on three-dimensional laser scanning, as provided in an embodiment of this application.

[0018] Figure 2 This is a schematic diagram of the structure of the oil surface crack identification system based on three-dimensional laser scanning provided in the embodiments of this application.

[0019] Figure labeling: Module 10 for establishing a 3D point cloud model of the oil surface, Module 20 for pre-segmenting oil cracks, Module 30 for multi-type probability detection, Module 40 for reliability crack classification, and Module 50 for establishing a crack distribution model of the oil surface. Detailed Implementation

[0020] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0021] This application provides a method for identifying surface cracks in oil based on three-dimensional laser scanning, such as... Figure 1 As shown, the method includes: Step S100: Perform a three-dimensional laser scan on the oil surface to obtain point cloud data of the oil surface, and perform noise reduction, resampling and registration on the point cloud data of the oil surface to establish a three-dimensional point cloud model of the oil surface.

[0022] Specifically, a high-precision 3D laser scanner was used on-site to continuously scan the entire surface of the ancient building's painted components. During the scanning process, the scanning spacing and ranging accuracy of the equipment were controlled to completely collect the spatial coordinate data and morphological reflection data of the painted surface, which were then integrated to generate original point cloud data of the painted surface. The original point cloud data of the painted surface was then processed, and a standardized processing flow was executed in sequence, including data denoising and removal of redundant points, resampling to normalize data density, and spatial registration and alignment of multi-view point cloud data. Based on the processed and normalized point cloud data, a 3D point cloud model of the painted surface that can realistically reproduce the three-dimensional morphology of the painted surface was constructed.

[0023] In one possible implementation, the oil surface point cloud data is denoised, resampled, and registered to establish a three-dimensional point cloud model of the oil surface. Step S100 further includes step S110, which performs adaptive denoising processing on the oil surface point cloud data based on the statistical distribution characteristics of local neighborhoods to obtain denoised oil surface point cloud data. Specifically, the original oil surface point cloud data acquired in step S100 is retrieved as input. The surrounding local neighborhood space corresponding to each single point in the point cloud data is defined, and two types of statistical characteristic parameters, namely spatial distance distribution and coordinate deviation distribution, are statistically analyzed for all point clouds in the local neighborhood. An adaptive noise judgment threshold is set according to the statistical distribution characteristics. Abnormal point clouds that deviate from the normal distribution range and belong to environmental interference and scanning noise are directly removed, while valid point cloud data that fits the true shape of the oil surface are retained. The denoised oil surface point cloud data after impurity removal is output.

[0024] Step S120 involves non-uniform resampling of the denoised oil painting point cloud data to obtain updated oil painting point cloud data, and constructing a point cloud feature descriptor based on the updated oil painting point cloud data. Specifically, the denoised oil painting point cloud data obtained in step S110 is used as the basic input. Differentiated non-uniform resampling operations are performed on flat areas and complex texture areas of the oil painting surface. The point cloud sampling density is appropriately reduced in flat areas, while the point cloud sampling density is increased at texture edges and component transition areas to balance the overall data volume and detail preservation effect. After processing, updated oil painting point cloud data is obtained. Three basic morphological features are extracted from each point in the updated oil painting point cloud data: spatial curvature, normal vector orientation, and neighborhood point distribution spacing. Multiple types of feature information are integrated to construct a standardized point cloud feature descriptor, realizing the structured representation of point cloud morphological information.

[0025] Step S130: Perform initial alignment of the updated oilfield point cloud data based on the point cloud feature descriptor to obtain an initial registration model of the oilfield point cloud. Specifically, retrieve the point cloud feature descriptor constructed in step S120 and the updated oilfield point cloud data. Using the point cloud feature descriptor as the matching basis, retrieve the corresponding points with the highest feature similarity among the updated oilfield point cloud data obtained from multiple batches of scanning. Based on the spatial correspondence of the matching points, perform preliminary spatial translation and rotation alignment of multiple sets of point cloud data to eliminate the overall positional deviation caused by multi-view scanning. After alignment and integration, generate a preliminary fused initial registration model of the oilfield point cloud.

[0026] Step S140 involves performing multi-point registration error correction on the initial registration model of the oil painting point cloud to generate a three-dimensional point cloud model of the oil painting surface. Specifically, the initial registration model of the oil painting point cloud obtained in step S130 is used as the correction object. Multiple component reference points and pattern feature points on the model surface are selected as correction reference points. The registration deviation values ​​generated by the superposition of multiple sets of point clouds at each reference point are detected one by one. The local point cloud positions are finely adjusted and corrected according to the deviation values ​​to eliminate the local misalignment and gap errors remaining in the initial alignment stage. After all points are corrected, the data structure is integrated and solidified to generate a three-dimensional point cloud model of the oil painting surface with the required accuracy.

[0027] Step S200: Based on the oiling process characteristic data, pre-segment the oiling surface three-dimensional point cloud model to create oiling cracks and establish a potential crack region model.

[0028] Specifically, the process involves pre-entering the traditional oil painting techniques of ancient buildings, including data on pattern creation specifications, coating layering standards, and component boundary divisions. This process feature data is then linked and matched with the 3D point cloud model of the oil-painted surface generated in step S140. Based on the process features, the target area prone to cracking is defined. The point cloud model is then pre-segmented across the entire region, eliminating conventional areas without crack risk and delineating the spatial range with the potential for crack development. Finally, the defined range is integrated to construct an initial model of potential crack areas.

[0029] In one possible implementation, the three-dimensional point cloud model of the oil-painted surface is pre-segmented for oil-painting cracks based on the oil-painting process feature data to establish a potential crack region model. Step S200 further includes step S210, extracting pattern distribution features, coating layer boundary features, and component partitioning features based on the oil-painting process feature data. Specifically, using standardized archived oil-painting process feature data as the sole data source, the arrangement position, size specifications, and extension direction information of various decorative patterns on the oil-painted surface recorded in the data are sorted out to extract pattern distribution features; the paint film layer thickness, layer boundary position, and layer transition morphology information corresponding to multi-layer oil-painting construction are sorted out to extract coating layer boundary features; and according to the building component classification standards, the boundary range and connection position information of independent components such as columns, beams, and carved panels are separated to extract component partitioning features.

[0030] Step S220: Based on the pattern distribution characteristics, perform pattern region identification and masking processing on the three-dimensional point cloud model of the oil surface to obtain a pattern-masked point cloud model. Then, based on the component partitioning characteristics, perform spatial partitioning constraints on the pattern-masked point cloud model to obtain a component partitioning point cloud model. Specifically, call the pattern distribution characteristics extracted in step S210 to locate the point cloud regions corresponding to all decorative patterns in the three-dimensional point cloud model of the oil surface. Perform data masking and marking operations on these regions to avoid interference from the pattern's concave and convex morphology with subsequent crack identification. After processing, a pattern-masked point cloud model is obtained. Then, introduce the component partitioning characteristics extracted in step S210, and use the component boundary range as the boundary to spatially divide and label the pattern-masked point cloud model. Add partitioning constraint markers to the point cloud range corresponding to each independent component. Perform cross-region identification operations for invalid constraints to generate a component partitioning point cloud model.

[0031] Step S230: Based on the coating layer boundary features, perform layer-level partitioning constraints on the component partition point cloud model to determine the sensitive areas for crack occurrence. Specifically, using the coating layer boundary features extracted in step S210 as the constraint basis, mark the boundary of multiple layers of paint film and the interlayer connection position within the component partition point cloud model obtained in step S220. Combining common knowledge of painting processes, delineate the high-incidence areas of interlayer cracking and surface peeling, add layer-level partitioning constraint marks to these areas, eliminate safe areas with intact and thick paint films, and finally lock in all spatial ranges prone to crack defects, defining them as sensitive areas for crack occurrence.

[0032] Step S240 involves performing local geometric analysis of the point cloud in the crack-sensitive area to determine curvature abrupt change features and height gradient anomaly features. Based on these features, a joint determination is made to extract candidate crack regions. Specifically, focusing on the point cloud data corresponding to the crack-sensitive area determined in step S230, local geometric morphology detection and calculation are performed point by point. The degree of curvature change of the surrounding point cloud at each point is statistically analyzed, and a set of points with a sudden change in curvature is selected, defined as curvature abrupt change features. Simultaneously, the continuous change difference in vertical height at each point is detected, and a set of points with a sudden increase in height difference and abnormal slope trend is selected, defined as height gradient anomaly features. The overlapping coverage of the two types of features is compared simultaneously, and only the spatial region possessing both curvature abrupt change and height gradient anomaly features is retained, defining this region as the preliminary candidate crack region.

[0033] Step S250 involves performing spatial continuity constraint screening on the candidate fracture regions to generate the potential fracture region model. Specifically, the point cloud data corresponding to the candidate fracture regions extracted in step S240 is retrieved, and the spatial continuity and connectivity of the point clouds within each candidate region are checked. Scattered, isolated, and non-continuous single-point abnormal regions and fragmented areas are eliminated, while effective regions with continuous extension trends and conforming to the elongated or sheet-like development pattern of fractures are retained. The boundary contours and data ranges of the effective regions are regularized, and after screening and optimization, the final potential fracture region model is generated.

[0034] Step S300: Perform multi-type probability detection on the potential fracture region model according to the multi-dimensional fracture type to establish a fracture type probability detection space, wherein the multi-dimensional fracture type includes hairline fracture, network fracture, through fracture and hollow fracture.

[0035] Specifically, four common standard crack morphologies in ancient architectural painted components are pre-defined as multi-dimensional crack types: hairline cracks (fine, narrow slits), network cracks (interlaced, densely packed slits), through cracks (penetrating the entire paint film), and hollow cracks (bulging, detached layers). Using the potential crack area model generated in step S250 as the detection carrier, the similarity probability is calculated for each spatial point within the model by matching the morphological features of the four crack types. The probability values ​​of all points corresponding to the four crack types are summarized, and all probability data are integrated according to spatial location to build a visualized and structured crack type probability detection space.

[0036] In one possible implementation, multi-type probability detection is performed on the potential fracture region model based on multi-dimensional fracture types to establish a fracture type probability detection space. Step S300 further includes step S310, loading sample sets of oil-based hairline fractures, oil-based network fractures, oil-based through fractures, and oil-based hollow fractures according to the multi-dimensional fracture types. Specifically, based on previously archived historical image data and measured point cloud samples of ancient building oil-based defects, the samples are classified and grouped according to four types of multi-dimensional fractures: hairline fractures, network fractures, through fractures, and hollow fractures, and independent standardized sample datasets are constructed for each. Before performing probability detection, the four complete sample sets are uniformly imported into the recognition and computing terminal to complete the loading, storage, and feature pre-reading of the sample data, providing a reference benchmark for similarity comparison.

[0037] Step S320: Based on the oil-based hairline crack sample set, perform probability detection on the potential fracture region model to obtain the hairline crack probability at each point. Specifically, call the oil-based hairline crack sample set loaded in step S310, use all typical hairline crack morphological features in the sample set as comparison standards, and match each spatial point in the potential fracture region model one by one. Calculate the degree of fit between the morphological features of each point and the standard hairline crack features, convert the degree of fit into a quantified probability value, and after traversing all points in the model, summarize and organize to obtain the hairline crack probability results corresponding to all points.

[0038] Step S330: Based on the oilfield network fracture sample set, perform probability detection on the potential fracture region model to obtain the network fracture probability at each point. Specifically, following the complete feature recognition and similarity calculation process from steps S320 to S324, use the oilfield network fracture sample set loaded in step S310 as the comparison benchmark, sequentially extract the feature vectors and network fracture sample vectors of each point in the potential fracture region model, use a Siamese neural network for similarity evaluation, select the maximum similarity and convert it into a probability value, and summarize the network fracture probability at each point after traversing all points.

[0039] Step S340: Based on the oil-based through-crack sample set, perform probability detection on the potential fracture region model to obtain the through-crack probability at each point. Specifically, simultaneously reuse the standardized calculation process of hairline crack probability detection, call the oil-based through-crack sample set loaded in step S310, and complete the extraction and comparison of the model point feature vector and the through-crack sample vector. Calculate the similarity using a Siamese neural network, filter peak values, and complete probability conversion. After a full traversal, integrate and output the through-crack probabilities corresponding to all points in the potential fracture region model.

[0040] Step S350: Based on the oilfield void crack sample set, perform probability detection on the potential fracture region model to obtain the void crack probability at each location. Specifically, strictly match the probability detection execution logic of the first three types of fractures, retrieve the oilfield void crack sample set stored in step S310, and perform a full comparison analysis between the location features and the void crack sample features. Using a unified neural network similarity evaluation and numerical conversion rule, generate the void crack probability corresponding to each location one by one, and summarize to form complete location void crack probability data.

[0041] Step S360: Organize the probabilities of hairline cracks, network cracks, through cracks, and hollow cracks at each point to generate the crack type probability detection space. Specifically, based on the spatial point arrangement order of the potential crack region model, the four types of values ​​corresponding to each point—hairline crack probability, network crack probability, through crack probability, and hollow crack probability—are bound and stored according to a fixed sequence rule to form a multi-dimensional probability detection sequence exclusive to each point. Then, based on the three-dimensional spatial location of the points, the probability detection sequences of all points are arranged and integrated across the entire domain to construct a crack type probability detection space that corresponds one-to-one with the point cloud model space and is completely associated with the data.

[0042] In one possible implementation, the potential fracture region model is probabilistically detected based on the oil-based hairline fracture sample set to obtain the hairline fracture probability at each point. Step S320 further includes step S321, which involves multi-point feature recognition of the potential fracture region model to obtain the nth potential point's feature vector, where n is a positive integer. Specifically, all spatial points within the potential fracture region model are traversed, and multi-point feature recognition is performed sequentially on each potential point numbered a positive integer n, extracting four core morphological parameters: surface curvature, fracture width, depression depth, and extension length around the point. These four morphological parameters are integrated and combined according to a fixed sorting rule to form a standardized data sequence uniquely corresponding to the point, which is defined as the nth potential point's feature vector.

[0043] Step S322 involves feature recognition of the oil-based hairline crack sample set to obtain vectors for each hairline crack sample. Specifically, for each historical defect sample included in the oil-based hairline crack sample set, the feature extraction standards of step S321 are uniformly applied to simultaneously extract four types of morphological parameters: surface curvature, crack width, depression depth, and extension length at typical crack locations within the sample. The parameter information is then integrated according to the same sorting rules to generate a standardized data sequence specific to each hairline crack sample. All data sequences are then aggregated to form a complete set of vectors for each hairline crack sample.

[0044] Step S323: Based on the Siamese neural network, the similarity between the feature vector of the nth point and the sample vectors of each hairline fracture is evaluated to obtain the similarity set of the nth hairline fracture. Specifically, the Siamese neural network, which has been pre-trained under supervised supervision and has its parameters fixed, is invoked. This network has two feature extraction branches with identical structures and shared parameters. One branch inputs the feature vector of the nth point alone to complete deep feature encoding compression, while the other branch inputs each hairline fracture sample vector sequentially to complete deep feature encoding compression of the same dimension. The feature data of the same dimension after encoding by the two branches are connected to the cosine similarity calculation module built into the network. The matching similarity value between the current point feature and the feature of a single hairline fracture sample is calculated group by group. All the similarity values ​​obtained by each calculation are collected and summarized to form the similarity set of the nth hairline fracture for the corresponding point.

[0045] Step S324: The maximum value is filtered from the nth hairline fracture similarity set to generate the hairline fracture probability at the nth point. Specifically, the nth hairline fracture similarity set obtained in step S323 is retrieved, all similarity values ​​within the set are traversed, and the peak similarity with the largest value is extracted. This peak similarity is then standardized and converted into a quantized probability value between 0 and 1. This value is the hairline fracture probability at the nth point corresponding to the current numbered point.

[0046] Step S400: Based on the crack type probability detection space, perform reliability crack classification on the potential crack region model to obtain the crack identification classification space.

[0047] Specifically, the crack type probability detection space built in step S360 is used as the data source for classification and judgment. For each spatial point in the potential crack area model, the probability values ​​of the four types of cracks are hierarchically identified and the type is determined according to the preset probability screening rules and spatial constraints. The specific crack type to which each point belongs is clarified or it is determined that there are no effective cracks. The classification results of all points are uniformly collected and organized according to their spatial location, and integrated to generate a crack recognition classification space with full coverage.

[0048] In one possible implementation, the potential fracture region model is classified for reliability fractures based on the fracture type probability detection space to obtain a fracture identification classification space. Step S400 further includes step S410, extracting the probability detection sequence of the nth point based on the fracture type probability detection space. Specifically, all spatial points numbered as positive integers n in the fracture type probability detection space are traversed, and four types of data—hairline fracture probability, network fracture probability, through fracture probability, and hollow fracture probability—pre-bound and stored for that point are retrieved. The four types of probability parameters are collected in a fixed order, and then extracted separately to form an independent and complete probability detection sequence for the nth point, providing basic data for a single point for screening and judgment.

[0049] Step S420: Perform a superiority test on the probability detection sequence of the nth point according to the predetermined probability value to obtain the nth superiority probability detection sequence. Specifically, a fixed pass / fail probability value is preset in the computing terminal as a screening threshold. The probability detection sequence of the nth point extracted in step S410 is retrieved, and the relationship between each probability parameter in the sequence and the predetermined probability value is compared one by one. Valid probability parameters in the sequence whose values ​​are greater than or equal to the predetermined probability value are screened and collected separately, and invalid probability parameters below the threshold value are removed. All the remaining valid parameters are integrated and summarized to form a simplified nth superiority probability detection sequence.

[0050] Step S430: If the nth winning probability detection sequence is a single value, determine the nth dominant fracture type and output the nth dominant fracture type as the nth fracture identification and classification result. Specifically, check the number of probability parameters contained in the nth winning probability detection sequence obtained in step S420. When only one valid probability parameter remains in the sequence, determine that the current sequence is a single value state. Retrieve the fracture category identifier corresponding to the unique probability parameter, define the fracture category as the nth dominant fracture type of the current point, and directly archive the dominant fracture type as the final nth fracture identification and classification result of the current point.

[0051] Step S440: If the nth winning probability detection sequence is not a single value, a weighted discrimination based on spatial neighborhood consistency constraints is performed according to the nth winning probability detection sequence to generate the nth crack identification and classification result. Specifically, when the nth winning probability detection sequence contains two or more valid probability parameters, it is determined that the current sequence is not a single value. Simultaneously, the crack classification results of the neighboring areas around the current point are retrieved, and spatial neighborhood consistency constraint rules are introduced. Priority is given to conforming to the development pattern of continuous cracks in the surrounding area. First, a basic weighting coefficient is added to the corresponding probability parameter based on the crack type with the highest proportion in the neighborhood. Then, the numerical weights are added according to the value of each probability parameter in the sequence. Each valid probability parameter is multiplied by its corresponding weighting coefficient in turn, and all product values ​​are summed. The comprehensive weighted score of each candidate crack type is calculated. The values ​​of all comprehensive weighted scores are compared, and the crack type with the highest score is selected as the current point's judgment result. Finally, the nth crack identification and classification result that conforms to the actual disease distribution is generated. If there are no valid probability parameters remaining in the nth winning probability detection sequence, the crack identification and classification result of the current point is directly marked as empty, and it is determined that there is no crack disease at the point.

[0052] Step S450: Add the nth crack identification and classification result to the crack identification and classification space. Specifically, for each nth crack identification and classification result generated in steps S430 and S440, enter it into the dedicated storage location of the crack identification and classification space according to the three-dimensional spatial position corresponding to the point. This completes the full entry and data binding of all point classification results, achieving the matching and solidification of crack type classification results with the point cloud spatial model, and perfecting the complete data architecture of the crack identification and classification space.

[0053] Step S500: Based on the crack identification and classification space, a quantitative evaluation is performed to obtain the oilfield crack classification quantization space. Then, the three-dimensional point cloud model of the oilfield surface is mapped and labeled according to the oilfield crack classification quantization space to establish an oilfield surface crack distribution model. Based on the oilfield surface crack distribution model, a multi-level crack early warning signal is generated.

[0054] Specifically, based on the fracture identification and classification space refined in step S450, a quantitative assessment of fracture severity is conducted, assigning corresponding hazard level values ​​to fractures of different types and developmental stages. The level values ​​are integrated and bound with the fracture classification results to construct a quantitative classification space for oilfield fractures containing both fracture type and hazard level data. All data within this quantitative space are then mapped and labeled onto the three-dimensional point cloud model of the oilfield surface generated in step S140, displaying the distribution location, type, and severity of fractures, thus creating a visualized model of fracture distribution on the oilfield surface. Based on the hazard levels labeled in the model, three warning levels—mild, moderate, and severe—are established, and corresponding multi-level fracture warning signals are automatically matched and output according to the actual fracture development.

[0055] In one possible implementation, a quantitative evaluation is performed based on the crack identification and classification space to obtain a quantitative crack classification space for oil processing. Step S500 further includes step S510, which involves capturing crack features in the potential crack region model according to the crack identification and classification space to obtain feature vectors for each crack. Specifically, guided by the crack range and type defined by the crack identification and classification space, the entire range of all identified cracks is locked within the potential crack region model. Five types of quantitative morphological parameters are extracted for each crack: extension length, crack width, indentation depth, distribution area, and paint film damage degree. These five types of parameters are integrated and sorted according to fixed rules to generate a feature data sequence specific to each crack. All data sequences are then summarized to obtain complete feature vectors for each crack.

[0056] Step S520: Construct a crack level evaluation network. Specifically, a dedicated multilayer perceptron architecture is built as the crack level evaluation network. The network input end is set with access nodes matching the number of crack feature vector parameters. The intermediate layers are set with multilayer feature operation nodes for parameter weighting comparison and level conversion. The output end is set with classification output nodes corresponding to multiple hazard levels. Oil samples labeled with actual damage levels are used as crack sample data in advance to complete the training, debugging, and finalization of network parameters, ensuring that the network can stably output standard level judgment results based on input features. Specifically, this multilayer perceptron architecture includes one input layer, two hidden layers, and one output layer. The number of nodes in the input layer is set to five, corresponding to five types of feature parameters: crack extension length, crack width, indentation depth, distribution area, and paint film damage degree. Each input node receives the quantified value of one type of feature parameter. The first hidden layer has 16 computation nodes and uses the ReLU activation function for nonlinear feature mapping. The second hidden layer has 8 computation nodes, also using the ReLU activation function. The output layer has 3 nodes, corresponding to the three crack levels of mild, moderate, and severe, and uses the Softmax activation function to output the probability values ​​of each level. During training, sample data with labeled crack levels are selected as the training set, and the training and validation sets are divided in a 7:3 ratio. The error between the predicted level and the actual labeled level is calculated using the cross-entropy loss function. The network weight parameters are iteratively adjusted using the gradient descent method, with a learning rate of 0.001 and 500 iterations. When the error on the validation set tends to stabilize and is below a preset threshold, training is stopped and the network parameters are fixed, completing the construction of the crack level evaluation network.

[0057] Step S530: Input the feature vectors of each fracture into the fracture level evaluation network to obtain the fracture level. Specifically, retrieve all the fracture feature vectors obtained in step S510, and input them one by one into the fracture level evaluation network constructed in step S520 according to a unified input format. The network automatically performs feature parameter comparison and level matching, outputs the specific hazard level label corresponding to each fracture, and after traversing all fracture feature vectors, summarizes and collects all judgment results to form complete fracture level data.

[0058] Step S540: Map each fracture level to the fracture identification and classification space to generate the oilfield fracture classification and quantification space. Specifically, the fracture level data obtained in step S530 are embedded one-to-one into the fracture identification and classification space formed in step S450 according to the spatial location and classification identifier of the fracture. Based on the original fracture type classification data, hazard level quantification labeling information is added to achieve data fusion and integration of fracture type and hazard degree, generating an oilfield fracture classification and quantification space that has both classification attributes and quantification rating attributes.

[0059] This application's embodiments utilize three-dimensional laser scanning to acquire point cloud data of the oil-worked surface. After denoising, resampling, and registration processing, a three-dimensional point cloud model of the oil-worked surface is constructed. Potential crack regions are pre-segmented based on oil-worked process characteristics. Multi-type probability detection is performed for various crack types, including hairline cracks, network cracks, through cracks, and hollow cracks, and a probability detection space is constructed. Based on this space, combined with neighborhood constraints and weighted discrimination, crack reliability classification is performed to obtain an identification classification space. Crack level quantification is performed based on the classification space, and the quantification results are mapped and labeled onto the three-dimensional point cloud model, forming a complete oil-worked surface crack distribution model. These technical methods solve the technical problems of low identification accuracy, easy missed detections and misjudgments, and low detection efficiency in existing ancient building oil-worked surface crack identification, which cannot meet the needs of large-area, refined, and hidden crack identification. This achieves the technical effects of improving crack identification accuracy, avoiding missed detections and misjudgments, increasing detection efficiency, and adapting to the needs of large-area, refined, and hidden crack identification.

[0060] In the above text, refer to Figure 1 A method for identifying surface cracks in oil based on three-dimensional laser scanning according to an embodiment of the present invention is described in detail. Next, reference will be made to... Figure 2 A three-dimensional laser scanning-based surface crack identification system for oil processing is described according to an embodiment of the present invention.

[0061] The oil surface crack identification system based on three-dimensional laser scanning according to embodiments of the present invention addresses the technical problems of low identification accuracy, easy omissions and misjudgments, and low detection efficiency in existing ancient building oil surface crack identification methods, which cannot meet the needs of large-area, high-precision, and hidden crack identification. It achieves the technical effects of improving crack identification accuracy, avoiding omissions and misjudgments, increasing detection efficiency, and adapting to the needs of large-area, high-precision, and hidden crack identification. The oil surface crack identification system based on three-dimensional laser scanning includes: an oil surface three-dimensional point cloud model establishment module 10, an oil crack pre-segmentation module 20, a multi-type probability detection module 30, a reliability crack classification module 40, and an oil surface crack distribution model establishment module 50.

[0062] The three-dimensional point cloud model establishment module 10 for oiling surface is used to perform three-dimensional laser scanning on the oiling surface to obtain point cloud data of the oiling surface, and to perform denoising, resampling and registration on the point cloud data of the oiling surface to establish a three-dimensional point cloud model of the oiling surface; the oiling crack pre-segmentation module 20 is used to pre-segment the oiling surface three-dimensional point cloud model of the oiling surface according to the oiling process feature data to establish a potential crack region model; the multi-type probability detection module 30 is used to perform multi-type probability detection on the potential crack region model according to multi-dimensional crack types to establish a crack type probability detection space; the reliability crack classification module 40 is used to perform reliability crack classification on the potential crack region model according to the crack type probability detection space to obtain a crack identification classification space; the oiling surface crack distribution model establishment module 50 is used to perform quantitative evaluation based on the crack identification classification space to obtain an oiling crack classification quantization space, and to map and label the oiling surface three-dimensional point cloud model of the oiling surface according to the oiling crack classification quantization space to establish an oiling surface crack distribution model.

[0063] The detailed description of the specific configuration of the three-dimensional point cloud model establishment module 10 for the oilfield surface is explained as follows: As mentioned above, the oilfield surface point cloud data is denoised, resampled, and registered to establish a three-dimensional point cloud model for the oilfield surface. The three-dimensional point cloud model establishment module 10 for the oilfield surface may further include: an adaptive denoising processing unit for performing adaptive denoising processing on the oilfield surface point cloud data based on the statistical distribution characteristics of local neighborhoods to obtain denoised oilfield point cloud data; a point cloud feature descriptor construction unit for performing non-uniform resampling on the denoised oilfield point cloud data to obtain updated oilfield point cloud data, and constructing a point cloud feature descriptor based on the updated oilfield point cloud data; an initial alignment unit for performing initial alignment on the updated oilfield point cloud data based on the point cloud feature descriptor to obtain an initial registration model of the oilfield point cloud; and a multi-point registration error correction unit for performing multi-point registration error correction on the initial registration model of the oilfield point cloud to generate the three-dimensional point cloud model for the oilfield surface.

[0064] The detailed description of the specific configuration of the pre-segmentation module 20 for oil-based cracks is explained below: As mentioned above, the pre-segmentation module 20 performs oil-based crack pre-segmentation on the three-dimensional point cloud model of the oil-based surface based on oil-based process feature data to establish a potential crack region model. The pre-segmentation module 20 may further include: a feature extraction unit for extracting pattern distribution features, coating layer boundary features, and component partitioning features based on the oil-based process feature data; and a component partitioning point cloud model acquisition unit for identifying and masking pattern regions on the three-dimensional point cloud model of the oil-based surface based on the pattern distribution features, obtaining a pattern-masked point cloud model, and performing pattern region identification and masking processing based on the component partitioning features. The process involves: 1) Spatial partitioning constraint of the pattern-masking point cloud model to obtain a component partitioned point cloud model; 2) Hierarchical partitioning constraint unit to perform hierarchical partitioning constraint on the component partitioned point cloud model based on the coating layer boundary features to determine the sensitive area for crack occurrence; 3) Candidate crack region extraction unit to perform local geometric analysis of the point cloud in the sensitive area for crack occurrence to determine curvature abrupt change features and height gradient anomaly features, and to extract candidate crack regions based on a joint determination of the curvature abrupt change features and the height gradient anomaly features; and 4) Spatial continuity constraint screening unit to perform spatial continuity constraint screening on the candidate crack regions to generate the potential crack region model.

[0065] The detailed configuration of the multi-type probability detection module 30 is explained below: As mentioned above, multi-type probability detection is performed on the potential fracture region model according to the multi-dimensional fracture types to establish a fracture type probability detection space. The multi-type probability detection module 30 may further include: a sample set loading unit for loading oil-based hairline fracture sample sets, oil-based network fracture sample sets, oil-based through fracture sample sets, and oil-based hollow fracture sample sets according to the multi-dimensional fracture types; and a probability detection unit for performing probability detection on the potential fracture region model according to the oil-based hairline fracture sample sets to obtain the probability of hairline fractures at each point. The probability detection method is as follows: Based on the oil-based network fracture sample set, probability detection is performed on the potential fracture region model to obtain the probability of network fractures at each point; based on the oil-based through fracture sample set, probability detection is performed on the potential fracture region model to obtain the probability of through fractures at each point; based on the oil-based hollow fracture sample set, probability detection is performed on the potential fracture region model to obtain the probability of hollow fractures at each point; the fracture type probability detection space generation unit is used to organize the probabilities of hairline fractures, network fractures, through fractures, and hollow fractures at each point to generate the fracture type probability detection space.

[0066] Specifically, the potential fracture region model is probabilistically detected based on the oil-based hairline fracture sample set to obtain the hairline fracture probability at each point. The probability detection unit may further include: a multi-point feature recognition subunit for performing multi-point feature recognition on the potential fracture region model to obtain the nth point feature vector of the nth potential point, where n is a positive integer; a feature recognition subunit for performing feature recognition on the oil-based hairline fracture sample set to obtain each hairline fracture sample vector; a similarity evaluation subunit for evaluating the similarity between the nth point feature vector and each hairline fracture sample vector based on a Siamese neural network to obtain the nth hairline fracture similarity set; and a maximum value filtering subunit for filtering the maximum value of the nth hairline fracture similarity set to generate the hairline fracture probability at the nth point.

[0067] The detailed configuration of the reliability crack classification module 40 is explained below: As described above, the potential crack region model is classified for reliability cracks according to the crack type probability detection space to obtain a crack identification classification space. The reliability crack classification module 40 may further include: an nth point probability detection sequence extraction unit for extracting the nth point probability detection sequence according to the crack type probability detection space; and a superior detection unit for performing superior detection on the nth point probability detection sequence according to a predetermined probability value to obtain the nth superior probability detection sequence. The nth dominant fracture type determination unit is used to determine the nth dominant fracture type if the nth winning probability detection sequence is a single value, and output the nth dominant fracture type as the nth fracture identification and classification result; the weighted discrimination unit is used to perform weighted discrimination based on spatial neighborhood consistency constraints according to the nth winning probability detection sequence if the nth winning probability detection sequence is not a single value, and generate the nth fracture identification and classification result; the nth fracture identification and classification result adding unit is used to add the nth fracture identification and classification result to the fracture identification and classification space.

[0068] The detailed description of the specific configuration of the oilfield surface crack distribution model establishment module 50 is as follows: As mentioned above, based on the crack identification classification space, a quantitative evaluation is performed to obtain the oilfield crack classification quantization space. The oilfield surface crack distribution model establishment module 50 may further include: a crack feature capture unit for capturing crack features of the potential crack region model according to the crack identification classification space to obtain each crack feature vector; a crack level evaluation network construction unit for constructing a crack level evaluation network; a crack level evaluation unit for inputting each crack feature vector into the crack level evaluation network to obtain each crack level; and an oilfield crack classification quantization space generation unit for mapping each crack level to the crack identification classification space to generate the oilfield crack classification quantization space.

[0069] The multi-type probability detection module 30 may further include: the multi-dimensional crack types include hairline cracks, network cracks, through cracks and hollow cracks.

[0070] The oil surface crack distribution model establishment module 50 may further include: a multi-level crack early warning signal generation unit for generating multi-level crack early warning signals based on the oil surface crack distribution model.

[0071] The oil surface crack identification system based on three-dimensional laser scanning provided in this embodiment of the invention can execute the oil surface crack identification method based on three-dimensional laser scanning provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method.

[0072] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this invention.

[0073] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for identifying surface cracks in oil based on three-dimensional laser scanning, characterized in that, The method includes: A three-dimensional laser scan is performed on the oil surface to obtain point cloud data of the oil surface. The point cloud data of the oil surface is then denoised, resampled, and registered to establish a three-dimensional point cloud model of the oil surface. Based on the oiling process characteristic data, the three-dimensional point cloud model of the oiling surface is pre-segmented to create oiling cracks, and a potential crack region model is established. Based on the multidimensional fracture types, multi-type probability detection is performed on the potential fracture region model to establish a fracture type probability detection space. Based on the crack type probability detection space, the potential crack region model is classified for reliability cracks to obtain a crack identification classification space. Based on the crack identification and classification space, a quantitative evaluation is performed to obtain the oilfield crack classification and quantification space. Then, the three-dimensional point cloud model of the oilfield surface is mapped and labeled according to the oilfield crack classification and quantification space to establish the oilfield surface crack distribution model.

2. The method for identifying surface cracks in oil based on three-dimensional laser scanning as described in claim 1, characterized in that, The oilfield surface point cloud data is denoised, resampled, and registered to establish a three-dimensional point cloud model of the oilfield surface, including: Adaptive denoising processing based on local neighborhood statistical distribution characteristics is performed on the oil surface point cloud data to obtain denoised oil point cloud data; Non-uniform resampling is performed on the denoised oilfield point cloud data to obtain updated oilfield point cloud data, and a point cloud feature descriptor is constructed based on the updated oilfield point cloud data. The updated oilfield point cloud data is initially aligned based on the point cloud feature descriptor to obtain the initial registration model of the oilfield point cloud. Multi-point registration error correction is performed on the initial registration model of the oil surface point cloud to generate the three-dimensional point cloud model of the oil surface.

3. The method for identifying surface cracks in oil based on three-dimensional laser scanning as described in claim 1, characterized in that, Based on the oiling process characteristic data, the three-dimensional point cloud model of the oiling surface is pre-segmented for oiling fractures to establish a potential fracture region model, including: Extract pattern distribution features, coating layer boundary features, and component partitioning features based on the oil painting process feature data; Based on the pattern distribution characteristics, the pattern region is identified and masked in the three-dimensional point cloud model of the oil surface to obtain the pattern masked point cloud model. Based on the component partitioning characteristics, the pattern masked point cloud model is spatially partitioned to obtain the component partitioned point cloud model. Based on the coating layer boundary features, the component partition point cloud model is subjected to hierarchical partition constraints to determine the sensitive area for crack occurrence. Local geometric analysis of point cloud is performed on the sensitive area where the crack occurs to determine the curvature abrupt change feature and the height gradient anomaly feature. Based on the curvature abrupt change feature and the height gradient anomaly feature, a joint judgment is made to extract candidate crack areas. The candidate fracture regions are screened by spatial continuity constraints to generate the potential fracture region model.

4. The method for identifying surface cracks in oil based on three-dimensional laser scanning as described in claim 1, characterized in that, Based on the multidimensional fracture types, multi-type probability detection is performed on the potential fracture region model to establish a fracture type probability detection space, including: Based on the multidimensional fracture types, the oil samples were processed as hairline fractures, network fractures, through fractures, and hollow fractures. Based on the oil-based hairline fracture sample set, the potential fracture region model is subjected to probability detection to obtain the hairline fracture probability at each point. Based on the oil-based network fracture sample set, the potential fracture region model is subjected to probability detection to obtain the network fracture probability at each point. Based on the oil-based through-crack sample set, the potential fracture region model is subjected to probability detection to obtain the through-crack probability at each point. Based on the oil void crack sample set, the potential crack region model is subjected to probability detection to obtain the void crack probability at each point. The probability of hairline cracks, network cracks, through cracks, and hollow cracks at each location are collected to generate the crack type probability detection space.

5. The method for identifying surface cracks in oil based on three-dimensional laser scanning as described in claim 4, characterized in that, Based on the oil-based hairline fracture sample set, probability detection is performed on the potential fracture region model to obtain the hairline fracture probability at each point, including: Multi-point feature identification is performed on the potential fracture region model to obtain the nth potential point feature vector, where n is a positive integer; Feature recognition is performed on the oil as a hairline fracture sample set to obtain the vector of each hairline fracture sample; The similarity between the feature vector of the nth point and the sample vectors of each hair fracture is evaluated based on the Siamese neural network to obtain the similarity set of the nth hair fracture. The maximum value is filtered from the nth hair fracture similarity set to generate the hair fracture probability at the nth point.

6. The method for identifying surface cracks in oil based on three-dimensional laser scanning as described in claim 1, characterized in that, Based on the crack type probability detection space, the potential crack region model is classified for reliability cracks to obtain a crack identification and classification space, including: Extract the probability detection sequence of the nth point based on the probability detection space of the crack type; Based on a predetermined probability value, a superior detection test is performed on the probability detection sequence of the nth point to obtain the nth superior probability detection sequence. If the nth winning probability detection sequence is a single value, the nth dominant fracture type is determined, and the nth dominant fracture type is output as the nth fracture identification and classification result; If the nth winning probability detection sequence is not a single value, a weighted discrimination based on spatial neighborhood consistency constraints is performed on the nth winning probability detection sequence to generate the nth crack identification classification result; The nth crack identification and classification result is added to the crack identification and classification space.

7. The method for identifying surface cracks in oil based on three-dimensional laser scanning as described in claim 1, characterized in that, Based on the fracture identification and classification space, a quantitative evaluation is performed to obtain the oilfield fracture classification quantization space, including: Based on the crack identification classification space, crack features are captured in the potential crack region model to obtain the feature vector of each crack. Construct a fracture level evaluation network; The feature vectors of each fracture are input into the fracture level evaluation network to obtain each fracture level; The fracture levels are mapped to the fracture identification and classification space to generate the oil fracture classification and quantification space.

8. The method for identifying surface cracks in oil based on three-dimensional laser scanning as described in claim 1, characterized in that, The multidimensional fracture types include hairline fractures, network fractures, through fractures, and hollow fractures.

9. The method for identifying surface cracks in oil based on three-dimensional laser scanning as described in claim 1, characterized in that, Based on the aforementioned surface crack distribution model, multi-level crack early warning signals are generated.

10. A surface crack identification system for oil processing based on three-dimensional laser scanning, characterized in that, The system is used to implement the oil surface crack identification method based on three-dimensional laser scanning as described in any one of claims 1-9, the system comprising: The module for establishing a three-dimensional point cloud model of the oil surface is used to perform three-dimensional laser scanning on the oil surface, acquire point cloud data of the oil surface, and perform noise reduction, resampling and registration on the point cloud data of the oil surface to establish a three-dimensional point cloud model of the oil surface. The oiling crack pre-segmentation module is used to pre-segment the oiling cracks on the three-dimensional point cloud model of the oiling surface according to the oiling process feature data, and to establish a potential crack region model. A multi-type probability detection module is used to perform multi-type probability detection on the potential fracture region model according to the multi-dimensional fracture type, and to establish a fracture type probability detection space. The reliability crack classification module is used to classify the potential crack region model for reliability cracks based on the crack type probability detection space, and obtain the crack identification classification space. The oilfield surface crack distribution model establishment module is used to perform quantitative evaluation based on the crack identification and classification space, obtain the oilfield crack classification and quantification space, and map and label the three-dimensional point cloud model of the oilfield surface according to the oilfield crack classification and quantification space to establish the oilfield surface crack distribution model.

Images