An automatic detection method for inconsistency in surface topology relationships

Abstract

An embodiment of the present invention provides a method for automatically detecting inconsistencies in surface topology relationships, wherein the method includes the following steps: obtaining vector map data containing surface cracks, annotating the vector map data to generate a surface crack sample library, constructing a lightweight surface crack target detection model, and training the lightweight surface crack target detection model using the surface crack sample library; extracting vector elements of the data to be detected to generate raster data, inputting the raster data into the lightweight surface crack target detection model, detecting the raster data to generate location information containing surface crack targets; obtaining the pixel coordinates of the surface crack element target in the location information containing the surface crack target, converting the pixel coordinates into geographic coordinate information under geographic coordinates, and visually displaying the surface crack element target to the terminal through the geographic coordinate information. Because this solution can automatically generate visualized surface crack detection results, it provides accurate information support for various application scenarios.

Description

Technical Field

[0001] The present invention relates to the technical field of vector data detection, and in particular to a method for automatically detecting inconsistency in surface topology relationships. Background Art

[0002] Detecting surface topology relationships in vector map data is a crucial step in geospatial data quality control. With the rapid development of Earth observation technology, geospatial data acquisition methods have diversified, including field surveying, remote sensing, map digitization, and internet acquisition. These data are prone to surface topology inconsistencies during production, storage, transmission, and conversion. A typical example of surface topology inconsistency is a sudden inward depression of a surface boundary, forming a sharp inward angle and appearing to be a crack (i.e., a surface crack), which directly affects data integrity and accuracy.

[0003] Facet crack detection aims to ensure the topological consistency of vector map data, specifically, the seamless presentation of polygonal features. This process not only impacts the accuracy of vector map data but also directly impacts the application of foundational geospatial data. Accurately identifying the locations of facet cracks not only improves the spatial and geometric accuracy of the data but also ensures the reliability and practicality of vector map data in geospatial data updates. This is a crucial step in continuously improving the quality of foundational geospatial data.

[0004] Common methods for detecting cracks in vector map data surfaces include direct observation and geometric calculation.

[0005] Direct Observation: This method examines the surface shapes in vector map data to check for obvious inward depressions, sharp corners, and cracks. While simple and intuitive, it relies on manual judgment, resulting in high costs, low efficiency, and a high error rate.

[0006] Geometric calculation method: By calculating parameters such as the angle and length of adjacent edges and comparing the predetermined values ​​with the actual values, the presence of surface cracks is determined. Actual angles and side lengths less than the predetermined values ​​indicate the presence of surface cracks. This method requires pre-setting appropriate inspection parameters and rules. Improper parameter settings can easily lead to missed or incorrect detections. Furthermore, when detecting surface cracks in complex polygonal elements, calculations must be performed for every edge and angle, resulting in lengthy computation times and low efficiency.

[0007] The above analysis shows that the direct observation method is highly dependent on manual interaction and operator experience, resulting in a low degree of automation. The geometric calculation method is complex to set up and lacks adaptability. Both methods suffer from omissions, false detections, and low efficiency. A more efficient surface crack detection method is urgently needed. Summary of the Invention

[0008] In view of this, an embodiment of the present invention provides a method for automatically detecting inconsistencies in surface topology relationships to address the technical problems of low automation, complex parameter settings, and low adaptability in topology consistency detection in the prior art. The method includes:

[0009] Obtain vector map data containing surface cracks, annotate the vector map data to generate a surface crack sample library, build a lightweight surface crack target detection model, and use the surface crack sample library to train the lightweight surface crack target detection model;

[0010] Extracting vector elements of the data to be detected to generate raster data, inputting the raster data into the lightweight surface crack target detection model, and detecting the raster data to generate location information containing surface crack targets;

[0011] The pixel coordinates of the surface crack feature target in the location information containing the surface crack target are obtained, the pixel coordinates are converted into geographic coordinate information under geographic coordinates, and the surface crack feature target is visualized to the terminal through the geographic coordinate information, wherein the pixel coordinates of the surface crack feature target include target pixel coordinate information, corner point coordinate information, width information and height information of the raster data.

[0012] Compared with the prior art, the at least one technical solution adopted in the embodiments of this specification can achieve the following beneficial effects:

[0013] The vector map data to be inspected is converted into raster data and input into the surface crack detection algorithm model to generate surface crack detection results. Finally, the detection results are coordinate-converted and superimposed on the original vector map data to obtain a visual detection result of the surface crack. By automatically generating visual detection results, geospatial data can provide accurate information support in various application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0015] Figure 1 This is a flow chart of a method for automatically detecting inconsistencies in surface topology relationships provided by an embodiment of the present invention;

[0016] Figure 2 This is a flow chart of a method for automatically detecting inconsistencies in surface topology relationships, provided by an embodiment of the present invention;

[0017] Figure 3The vector map data including surface cracks provided by an embodiment of the present invention;

[0018] Figure 4 Schematic diagram of the structure of a lightweight surface crack target detection model provided by an embodiment of the present invention;

[0019] Figure 5 The grid data of the detection result of the element target containing surface cracks provided by the embodiment of the present invention;

[0020] Figure 6 Schematic diagram of pixel coordinates of a surface crack element target provided by an embodiment of the present invention;

[0021] Figure 7 This is a schematic diagram of vector map data of surface crack elements to be detected provided by an embodiment of the present invention;

[0022] Figure 8 It will Figure 7 Schematic diagram of the vector map data surface crack feature target visualization. DETAILED DESCRIPTION

[0023] The embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0024] The following describes the embodiments of the present application through specific examples, and those skilled in the art can easily understand other advantages and effects of the present application from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The present application can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present application. It should be noted that, in the absence of conflict, the features in the following embodiments and embodiments can be combined with each other. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without making creative work are within the scope of protection of this application.

[0025] An embodiment of the present invention proposes a method for automatically detecting inconsistencies in surface topology relationships. In this embodiment, surface crack samples are first collected, labeled, and model trained. If necessary, sample amplification is performed to improve model stability and accuracy. Secondly, vector elements are extracted from the vector map data to be detected and raster data is generated. The raster data is input into the surface crack detection algorithm model to generate surface crack detection results. Finally, the detection results are coordinate transformed and the transformed detection results are superimposed on the original vector map data to obtain a visual detection result of the surface crack.

[0026] This method utilizes artificial intelligence recognition technology. Compared with traditional vector map data surface crack detection methods, it has a high degree of automation, simple operation, strong adaptability, and reduces the missed detection rate. It also improves detection efficiency when performing complex surface inspections.

[0027] In an embodiment of the present invention, a method for automatically detecting inconsistencies in surface topology relationships is provided. Figure 1 and Figure 2 As shown, the method includes:

[0028] Step S101: obtaining vector map data containing surface cracks, annotating the vector map data to generate a surface crack sample library, building a lightweight surface crack target detection model, and training the lightweight surface crack target detection model using the surface crack sample library;

[0029] Step S102: extracting vector elements of the data to be detected to generate raster data, inputting the raster data into a lightweight surface crack target detection model, and detecting the raster data to generate location information containing surface crack targets;

[0030] Step S103: Obtain the pixel coordinates of the surface crack element target in the location information containing the surface crack target, convert the pixel coordinates into geographic coordinate information under geographic coordinates, and visualize the surface crack element target to the terminal through the geographic coordinate information, wherein the pixel coordinates of the surface crack element target include target pixel coordinate information, corner point coordinate information, width information and height information of the raster data.

[0031] In specific implementation, the following steps are performed to generate a surface crack sample library after annotating the vector map data and to build a lightweight surface crack target detection model:

[0032] Obtain the surface crack features in the vector map data, use the sample annotation tool to annotate the surface crack features to generate crack samples behind the annotations; expand the samples based on the crack samples behind the annotations to generate a surface crack sample library.

[0033] Specifically, the sample annotation tool is a circular annotation tool. For surface crack features, the accuracy of the model trained using the circular annotation tool is better than the model trained using the rectangular annotation tool.

[0034] In the specific implementation, in order to improve the richness of the sample library, the following steps are performed to expand the samples based on the annotated fracture samples to generate a surface fracture sample library:

[0035] The marked crack sample is rotated at any angle, and the rotated sample is used as the amplified crack sample and / or the marked crack sample is mirror-flipped, and the flipped sample is used as the amplified crack sample and / or the surface crack portion of the marked crack sample is displaced, and the displaced sample is used as the amplified crack sample and / or the marked crack sample is scaled, and the scaled sample is used as the amplified crack sample.

[0036] Specifically, the sample data is amplified by using methods such as rotation, mirror flipping, displacement, and scaling to enrich the surface crack samples and increase the stability and reliability of the model. Figure 3 shown.

[0037] In specific implementation, in order to achieve fast and accurate detection under limited resources, a lightweight surface crack target detection model is constructed through the following steps:

[0038] The backbone network, neck network and detection head are constructed separately; the backbone network includes the interconnected initial layer and four stage layers, and the stage layers are used to gradually extract features at different levels of the image; the neck network includes the interconnected spatial pyramid pooling fast module and the path aggregation network, and the neck network is used to fuse feature maps of different stages; the detection head is an anchor-free detection head.

[0039] During specific implementation, the first and fourth stages of the construction phase layer are implemented through the following steps:

[0040] Among the four stage layers, the first stage layer includes a dual convolution C2f module and a CBS module; the fourth stage layer includes a C2f module, a CBS module and an SPPF module. The SPPF module is arranged between the backbone network and the neck network and is used to fuse feature information of different scales.

[0041] During specific implementation, the second and third phases of the construction phase layer are implemented through the following steps:

[0042] Among the four-stage layers, the second-stage layer and the third-stage layer both include deformable convolution modules and CBS modules.

[0043] Specifically, CBS stands for Conv BatchNorm Swish (Convolutional Batch Normalization Swish Activation Function), which is a convolutional layer followed by a batch normalization layer and a Swish activation function. SPPF stands for Spatial Pyramid Pooling Fast. C2f stands for Cross Stage Feature Fusion.

[0044] Specifically, surface crack detection uses a target recognition technology. By adding a deformable convolution module to the backbone network of a lightweight surface crack target detection model, model training and surface crack detection are performed. The main reasons are:

[0045] (1) The fixed weights of the convolution kernels result in the same receptive field size when the same convolutional neural network processes different locations of an image. Since objects of different scales or deformations may correspond to different locations in the feature map, the network needs to adaptively adjust the receptive field of the object.

[0046] (2) When sampling, deformable convolution is closer to the size and shape of the object and is more robust, which is impossible with traditional convolution.

[0047] (3) Since the target is small in size and has a non-fixed shape, if traditional convolution is still used, some complex crack targets may not be detected, which will affect the performance of the model.

[0048] like Figure 4 As shown in the figure, in the lightweight surface crack object detection model, the backbone network extracts features from the input image. This feature extraction allows for better understanding and description of the image. The backbone network typically consists of multiple convolutional layers, and different convolutional layer structures can be used to extract features at different levels. By fusing and integrating features from different levels, more comprehensive and accurate image features can be obtained, thereby improving model performance.

[0049] The network structure after adding the deformable convolution module is shown in the figure below: the penultimate and third C2f modules of the backbone network are replaced with deformable convolution modules (C2f_DCN modules). That is, the convolution of the detection layer of the backbone network is replaced with a deformable convolution module to improve the model's attention to the crack target. After adding the deformable convolution module, because the receptive field of the crack target location has adaptively changed, when the prediction box is regressed, the model can better adjust the regression parameters of the prediction box, so that the model increases its attention to the crack target, thereby improving the overall performance of the model. Figure 5 As shown in the figure, the trained model is used for inference to obtain the result of feature target data containing surface cracks.

[0050] Specifically, the anchor-free detection head performs object detection by directly predicting the center point coordinates, width, height, and class probability of the object. This method eliminates the traditional anchor frame step, allowing the model to focus more directly on the location and category of the object.

[0051] An anchor-free detection head typically includes the following key steps: It uses an advanced backbone network and neck network to extract high-dimensional features from images. It directly predicts object bounding boxes instead of relying on anchor boxes. It uses CIOU Loss and DFL (Distribution Focal Loss) as loss functions to optimize model training. It also employs a sample allocation strategy based on Task Alignment Learning (TAL) to improve model training efficiency.

[0052] In specific implementation, the following steps are performed to obtain the pixel coordinates of the surface crack feature target in the location information containing the surface crack target, and convert the pixel coordinates into geographic coordinate information under geographic coordinates:

[0053] Get the pixel coordinates of the detection target area of ​​the raster data of the surface crack feature target (P x ,P y ); Get the latitude and longitude coordinates of the upper left corner (ulx,uly), the upper right corner (urx,ury), the lower left corner (llx,lly) and the lower right corner (lrx,lry) of the raster data of the surface crack feature target; Get the pixel width and height information (pixel_width,pixel_height) of the raster data of the surface crack feature target; Through the target pixel coordinate information (P x ,P y ), upper right corner coordinate information (ulx,uly), upper left corner coordinate information (urx,ury), lower right corner coordinate information (llx,lly), lower right corner coordinate information (lrx,lry), pixel width pixel_width of raster data and pixel height pixel_height of raster data are used to calculate the longitude resolution P w and latitude resolution p h ; According to the upper right corner coordinate information (ulx,uly), longitude resolution p w , latitude resolution p h and the pixel coordinates of the detection target area (P x ,P y ) Calculate the longitude and latitude of the geographic coordinates, where longitude lon=ulx+Pw *P x ; Latitude lat=uly+(-P h )*P y .

[0054] In specific implementation, the following steps are performed to realize the target pixel coordinate information (P x ,P y ), upper right corner coordinate information (ulx,uly), upper left corner coordinate information (urx,ury), lower right corner coordinate information (llx,lly), lower right corner coordinate information (lrx,lry), pixel width pixel_width of raster data and pixel height pixel_height of raster data are used to calculate the longitude resolution P w and latitude resolution p h :

[0055] Longitude resolution Latitude resolution

[0056] Specifically, the latitude and longitude coordinates of the four corner points and the pixel coordinates are as follows Figure 6 As shown. Based on the known longitude and latitude coordinates of the four corner points and the pixel coordinates, the geographic spatial resolution corresponding to the unit pixel is calculated. Geographic spatial resolution includes longitude resolution and latitude resolution. Longitude resolution can also be called horizontal resolution. Latitude resolution can also be called vertical resolution. The corresponding geographic coordinates (longitude and latitude) are calculated by longitude resolution and latitude resolution, that is, the pixel coordinates of the detected target area are converted to the same coordinate system of the vector map data through coordinate conversion.

[0057] In specific implementation, the following steps are taken to obtain the maximum and minimum values ​​of the element coordinates to control the aspect ratio of the generated raster data to remain unchanged: when extracting the vector elements of the data to be detected to generate raster data, the aspect ratio of the vector elements of the data to be detected is the same as the aspect ratio of the generated raster data.

[0058] Specifically, by obtaining the maximum and minimum values ​​of the element coordinates and controlling the aspect ratio of the generated raster data to remain unchanged, the accuracy of the surface crack detection results can be improved.

[0059] Specifically, the pixel coordinates of the detected target area are converted to the same coordinate system as the vector map data through coordinate conversion. By converting the pixel coordinates to the projection coordinate system of the original vector map data, a more accurate visualization can be achieved.

[0060] The obtained target area coordinate information is displayed using visualization software, and the surface crack locations are mapped, allowing users to more intuitively and quickly find the surface crack locations. Surface crack detection ensures the consistency of the topological relationship of the vector map data.

[0061] Figure 7 It is the original vector map data without any processing. Figure 8 For example Figure 7 The vector map data is the vector map data after the surface crack positions are drawn.

[0062] The embodiments of the present invention achieve the following technical effects:

[0063] The automatic detection method for inconsistencies in surface topology relationships in an embodiment of the present invention comprehensively utilizes artificial intelligence methods, breaks through traditional algorithm processing concepts, constructs a surface crack sample library based on deep learning, and obtains a stable algorithm model through training, providing new ideas for artificial intelligence in vector map data quality inspection; the embodiment of the present invention uses artificial intelligence methods to detect surface cracks, cancels the traditional algorithm preset value setting, has a high degree of automation, strong adaptability, and reduces the missed detection rate; the embodiment of the present invention can realize automatic processing of large amounts of data, save cost expenditure, and improve detection efficiency.

[0064] Obviously, those skilled in the art should understand that the various modules or steps of the above-mentioned embodiments of the present invention can be implemented using a general-purpose computing device, they can be concentrated on a single computing device, or distributed across a network composed of multiple computing devices. Alternatively, they can be implemented using program code executable by the computing device, so that they can be stored in a storage device and executed by the computing device. In some cases, the steps shown or described can be performed in a different order than herein, or they can be made into separate integrated circuit modules, or multiple modules or steps can be made into a single integrated circuit module for implementation. Thus, the embodiments of the present invention are not limited to any specific combination of hardware and software.

[0065] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A method for automatically detecting inconsistencies in surface topology relationships, characterized in that: include: Acquire vector map data containing surface cracks, generate a surface crack sample library after annotating the vector map data, build a lightweight surface crack target detection model, and train the lightweight surface crack target detection model using the surface crack sample library; Extracting vector elements of the data to be detected to generate raster data, inputting the raster data into the lightweight surface crack target detection model, and detecting the raster data to generate position information containing the surface crack target; Obtaining pixel coordinates of a surface crack element target in the location information containing the surface crack target, converting the pixel coordinates into geographic coordinate information under geographic coordinates, and visually displaying the surface crack element target on a terminal using the geographic coordinate information, wherein the location information of the surface crack element target includes target pixel coordinate information, corner point coordinate information, and width and height information of raster data; Obtaining pixel coordinates of the surface crack element target in the location information containing the surface crack target, and converting the pixel coordinates into geographic coordinate information under geographic coordinates, including: Get the pixel coordinates of the detection target area of ​​the grid data of the surface crack element target (P x ,P y ); Obtain the latitude and longitude coordinates (ulx, uly) of the upper left corner, the latitude and longitude coordinates (urx, ury) of the upper right corner, the latitude and longitude coordinates (llx, lly) of the lower left corner, and the latitude and longitude coordinates (lrx, lry) of the lower right corner of the raster data of the surface crack feature target; Obtain the pixel width pixel_width and height information pixel_height of the raster data of the surface crack feature target; By the target pixel coordinate information (P x ,P y ), the upper left corner coordinate information (ulx, uly), the upper right corner coordinate information (urx, ury), the lower left corner coordinate information (llx, lly), the lower right corner coordinate information (lrx, lry), the pixel width pixel_width of the raster data and the pixel height pixel_height of the raster data are used to calculate the longitude resolution and latitude resolution ; According to the upper left corner coordinate information (ulx, uly), the longitude resolution , the latitude resolution and the pixel coordinates of the detection target area (P x ,P y ) Calculate the longitude and latitude of the geographic coordinates, where longitude ,latitude .

2. The method for automatically detecting inconsistencies in surface topology relationships according to claim 1, wherein: After annotating the vector map data, a surface crack sample library is generated, and a lightweight surface crack target detection model is constructed, including: Obtaining surface crack elements in the vector map data, and using a sample annotation tool to annotate the surface crack elements to generate crack samples behind the annotations; The samples are expanded based on the annotated crack samples to generate a surface crack sample library.

3. The method for automatically detecting inconsistencies in surface topology relationships according to claim 2, wherein: The sample marking tool is a circular marking tool.

4. The method for automatically detecting inconsistencies in surface topology relationships according to claim 2, wherein: Based on the annotated crack samples, the samples are expanded to generate a surface crack sample library, including: Rotate the annotated crack sample at any angle and use the rotated sample as the amplified crack sample and / or The annotated crack sample is mirror-flipped and the flipped sample is used as the amplified crack sample and / or Displace the surface crack portion of the marked back crack sample, and use the displaced sample as the amplified back crack sample and / or, The annotated back crack sample is scaled, and the scaled sample is used as the amplified back crack sample.

5. The method for automatically detecting inconsistency in surface topology relationships according to claim 1, wherein: Build a lightweight surface crack target detection model, including: Build interconnected backbone network, neck network and detection head respectively; The backbone network includes an initial layer and four stage layers that are interconnected, and the stage layers are used to gradually extract features at different levels of the image; The neck network includes an interconnected spatial pyramid pooling fast module and a path aggregation network, and the neck network is used to fuse feature maps at different stages; The detection head is an anchor-free detection head.

6. The method for automatically detecting inconsistency in surface topology relationships according to claim 5, wherein: Among the four stage layers, the first stage layer includes the C2f module and the CBS module; The fourth stage layer includes a C2f module, a CBS module and an SPPF module. The SPPF module is arranged between the backbone network and the neck network and is used to fuse feature information of different scales.

7. The method for automatically detecting inconsistency in surface topology relationships according to claim 5, wherein: Among the four stage layers, the second stage layer and the third stage layer both include a deformable convolution module and a CBS module.

8. The method for automatically detecting inconsistency in surface topology relationships according to claim 1, wherein: By the target pixel coordinate information (P x ,P y ), the upper left corner coordinate information (ulx, uly), the upper right corner coordinate information (urx, ury), the lower left corner coordinate information (llx, lly), the lower right corner coordinate information (lrx, lry), the pixel width pixel_width of the raster data and the pixel height pixel_height of the raster data are used to calculate the longitude resolution and latitude resolution ,include: Longitude resolution ; Latitude resolution .

9. The method for automatically detecting inconsistency in surface topology relationships according to claim 1, wherein: Also includes: When extracting vector elements of the data to be detected to generate raster data, the aspect ratio of the vector elements of the data to be detected is the same as the aspect ratio of the generated raster data.

Images