A method and system for automatically detecting defects in an electrically adjustable antenna structure

Abstract

The application relates to the technical field of defect detection, and discloses an electrically adjustable antenna structure defect automatic detection method and system, which comprises the following steps: collecting multi-modal data of an electrically adjustable antenna, and generating a three-dimensional model; mapping the three-dimensional model to a two-dimensional space, performing displacement on pixels by analyzing the curvature distribution of the surface of the electrically adjustable antenna, filling in the missing areas in the two-dimensional space, and obtaining a two-dimensional model; obtaining a first defect identification result; extracting a defect endpoint set from the first defect identification result, constructing a defect graph, mapping the defect graph to the three-dimensional model, and obtaining a second defect identification result of the electrically adjustable antenna; the application is suitable for the geometric structure of parabolic surfaces and cylindrical surfaces in the electrically adjustable antenna, reduces image distortion and information loss caused by the curved surface structure of the electrically adjustable antenna, improves the quality of the constructed model, and in combination with thermal image data, can timely find local overheating and loose connectors caused by electrical faults in the electrically adjustable antenna, and improves the accuracy of the defect identification result.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of defect detection, more particularly to an electrically adjustable antenna structure defect automatic detection method and system. BACKGROUND

[0002] At present, as a key component in the communication system, the structural integrity of the electrically adjustable antenna directly affects the stability of signal transmission, the accuracy of beam pointing and the overall performance of the system. With the rapid development of 5G, Internet of Things, satellite communication and other technologies, the use scenarios of electrically adjustable antennas are more complex, and the requirements for their structural quality and reliability are higher. The existing electrically adjustable antenna structure defect detection relies on manual inspection or single sensor, which is time-consuming and labor-intensive and difficult to realize large-scale and high-frequency automatic detection. The data collected by the single sensor cannot fully reflect the structural state of the electrically adjustable antenna, and it is easy to miss or misjudge in the curved and occluded areas, which cannot meet the needs of electrically adjustable antenna structure integrity and defect automatic identification.

[0003] The existing technology has the following problems: using single modal data to detect the integrity of the electrically adjustable antenna structure cannot obtain comprehensive structural information of the electrically adjustable antenna; based on single spatial dimension data for defect detection, it is difficult to identify and detect defects in positions with large curvature and occlusion of the electrically adjustable antenna; based on single data and single model, the identification integrity and accuracy of defects are insufficient, and the accurate defect morphology and type cannot be identified; in order to solve at least one of the above problems, the present application proposes an electrically adjustable antenna structure defect automatic detection method and system. SUMMARY

[0004] In view of the deficiencies of the prior art, the purpose of the present application is to provide an electrically adjustable antenna structure defect automatic detection method and system, which can effectively solve the problems in the background art. The specific technical solution of the present application is as follows:

[0005] An electrically adjustable antenna structure defect automatic detection method, comprising:

[0006] Obtaining multi-modal data of the electrically adjustable antenna by a preset sensor group, and generating a three-dimensional model in combination with the structure of the electrically adjustable antenna;

[0007] Mapping the three-dimensional model to a two-dimensional space according to a preset spatial mapping rule, performing displacement on the pixels by analyzing the curvature distribution of the surface of the electrically adjustable antenna, and filling the missing areas in the two-dimensional space to obtain a two-dimensional model;

[0008] Combining the thermal image data of the electrically adjustable antenna and the two-dimensional model, identifying the defects of the electrically adjustable antenna by a preset defect identification model to obtain a first defect identification result;

[0009] The defect endpoint set is extracted from the first defect identification result. By calculating the connection probability between the defect endpoints, a defect map is constructed. The defect map is then mapped onto a three-dimensional model to obtain the second defect identification result of the electrically tunable antenna.

[0010] Specifically, the process of acquiring multimode data of the electrically tunable antenna through a preset sensor group, and generating a three-dimensional model based on the structure of the electrically tunable antenna, includes:

[0011] Data is collected from the electrically tunable antenna from multiple perspectives using a pre-set sensor group. The collected data is then aligned to obtain multimodal data.

[0012] Based on the structure of the electrically tunable antenna, corresponding structural constraints are generated, and the multi-mode data is fused using the structural constraints to obtain fused data.

[0013] Based on the structure of the electrically tunable antenna, a three-dimensional model of the electrically tunable antenna is constructed using the fused data.

[0014] Specifically, the process of mapping the three-dimensional model to a two-dimensional space according to a preset spatial mapping rule, displacing pixels by analyzing the curvature distribution of the electrically tunable antenna surface, and filling in missing areas in the two-dimensional space to obtain a two-dimensional model includes:

[0015] According to the preset spatial mapping rules, the three-dimensional model is mapped into a two-dimensional space, the two-dimensional coordinates of each point of the three-dimensional model are calculated, and the three-dimensional model is unfolded according to the two-dimensional coordinates to obtain the first unfolded image;

[0016] Boundary recognition is performed on the stitched area in the first unfolded image. Combined with the curvature distribution of the electrically tunable antenna surface, the stitched area is coupled at the boundary through pixel displacement to obtain the second image.

[0017] The location and type of missing regions in the second image are identified by a preset missing region recognition model, and the missing regions are filled in according to the type of missing regions to obtain a two-dimensional model.

[0018] Specifically, the step of performing boundary recognition on the stitched region in the first unfolded image, and combining the curvature distribution of the electrically tunable antenna surface, coupling the stitched region to its boundaries through pixel displacement to obtain the second image includes:

[0019] Based on the curvature distribution of the electrically tunable antenna surface, the curvature gradient is calculated, and a curvature gradient field is generated.

[0020] Based on the curvature gradient field, the boundary of the stitching region in the first unfolded image is identified, and the pixel displacement of the corresponding stitching region is calculated in combination with the curvature. The movement of the stitching region is controlled according to the pixel displacement.

[0021] By analyzing the texture trend in the spliced ​​area after movement, the spliced ​​areas with consistent texture trends are bounded together to obtain the second image.

[0022] Specifically, the step involves identifying the location and type of missing regions in the second image using a preset missing region identification model, and filling in the missing regions according to their type to obtain a two-dimensional model, including:

[0023] The location and type of missing regions in the second image are identified using a preset missing region identification model. The missing regions include structural missing regions, texture missing regions, and boundary missing regions.

[0024] For structurally missing regions, the boundary is fitted by combining the curvature distribution of the electrically tunable antenna surface, and the missing regions are reconstructed to obtain the first filling region.

[0025] For texture-deficient areas, the texture trend on the surface of the electrically tunable antenna is analyzed to predict the texture trend and fill in the deficient areas to obtain a second filling area;

[0026] For regions with missing boundaries, the third filling region is obtained by analyzing the connection probability between boundary endpoints and the boundary trend.

[0027] By combining the first, second, and third filling regions, a two-dimensional model is obtained.

[0028] Specifically, by combining the thermal image data and two-dimensional model of the electrically tunable antenna, and using a preset defect identification model, defects in the electrically tunable antenna are identified to obtain a first defect identification result, including:

[0029] The thermal image data and two-dimensional model of the electrically adjustable antenna are matched and corresponded. The features of the thermal image data and the two-dimensional model are extracted respectively, and the features are fused to obtain the fused features.

[0030] Based on the fusion features, the defects of the electrically adjustable antenna are identified using a preset defect identification model to obtain the first defect identification result.

[0031] Specifically, the thermal image data and two-dimensional model of the electrically adjustable antenna are matched and mapped, and the features of the thermal image data and the two-dimensional model are extracted respectively. These features are then fused to obtain fused features, including:

[0032] According to the structure of the electrically tunable antenna, the thermal image data and the two-dimensional model are matched in position. In the thermal image data and the two-dimensional model after the position is matched, the temperature similarity of the corresponding position points is calculated. Based on the temperature similarity, the thermal image data and the two-dimensional model are registered to obtain the registered thermal image.

[0033] Feature extraction is performed on the registered thermal image and the two-dimensional model respectively to obtain a first feature set and a second feature set;

[0034] The first feature set and the second feature set are weighted and fused to obtain the fused features.

[0035] Specifically, a set of defect endpoints is extracted from the first defect identification result. A defect map is constructed by calculating the connection probability between the defect endpoints. The defect map is then mapped onto a three-dimensional model to obtain the second defect identification result for the electrically tunable antenna, including:

[0036] Defect endpoints are identified from the first defect identification results to obtain a set of defect endpoints;

[0037] By calculating the connection probability between defect endpoints, connection edges are established between defect endpoints whose connection probability is greater than a preset probability threshold to construct a defect graph;

[0038] The defect map is mapped onto a three-dimensional model to obtain the second defect identification result of the electrically tunable antenna.

[0039] Specifically, the step of identifying defect endpoints from the first defect identification result to obtain a set of defect endpoints includes:

[0040] Using a pre-defined endpoint identification model, defect endpoints are identified from the first defect identification results to obtain the first endpoint set;

[0041] By combining the preset endpoint criteria and curvature verification mechanism, the endpoints in the first endpoint set are filtered to obtain the second endpoint set;

[0042] By analyzing the trend consistency of the endpoint curvature gradient, the endpoints in the second endpoint set are filtered to obtain the defective endpoint set.

[0043] An automatic detection system for electrically tunable antenna structural defects, used to implement the aforementioned automatic detection method for electrically tunable antenna structural defects, includes:

[0044] The 3D model building module collects multi-mode data of the electrically tunable antenna through a preset sensor group, and generates a 3D model based on the structure of the electrically tunable antenna.

[0045] The two-dimensional model mapping module maps the three-dimensional model to a two-dimensional space according to a preset spatial mapping rule. It shifts the pixels by analyzing the curvature distribution of the electrically tunable antenna surface and fills in the missing areas in the two-dimensional space to obtain the two-dimensional model.

[0046] The first defect identification module combines the thermal image data and two-dimensional model of the electrically adjustable antenna to identify defects in the electrically adjustable antenna through a preset defect identification model, and obtains the first defect identification result.

[0047] The second defect identification module extracts a set of defect endpoints from the first defect identification result, constructs a defect map by calculating the connection probability between the defect endpoints, and maps the defect map onto a three-dimensional model to obtain the second defect identification result of the electrically tunable antenna.

[0048] The beneficial effects of this application are as follows: Based on multimodal data and the structure of electrically tunable antennas, a three-dimensional model of the electrically tunable antenna is constructed. This model is mapped according to spatial mapping rules, and pixel displacement and missing region filling are performed using curvature distribution to obtain a two-dimensional model. Defects in the electrically tunable antenna are identified and detected by combining thermal imaging data and the two-dimensional model. A defect map is constructed based on the defect endpoints and mapped onto the three-dimensional model to obtain complete defect identification results. By fusing multimodal data and structural data, an accurate three-dimensional model can be constructed. Pixel displacement and missing region filling are performed using the surface curvature data of the electrically tunable antenna, which is suitable for the parabolic and cylindrical geometric structures in electrically tunable antennas. This reduces image distortion and information loss caused by the curved surface structure of the electrically tunable antenna, improving the quality of the two-dimensional model. Identifying structural defects in the electrically tunable antenna by combining thermal imaging data and the two-dimensional model can promptly detect problems such as localized overheating and loose connectors caused by electrical faults, improving the accuracy of defect identification results. By constructing a defect map and mapping it onto the three-dimensional model, complete defect information can be identified, providing a basis for defect repair. Attached Figure Description

[0049] Figure 1 This is a flowchart illustrating the automatic detection method for structural defects in an electrically tunable antenna according to an embodiment of this application.

[0050] Figure 2 This is a schematic diagram of the two-dimensional model construction process in the embodiments of this application;

[0051] Figure 3 This is a schematic diagram of a defect diagram in an embodiment of this application;

[0052] Figure 4 This is a schematic diagram of the defect map mapped to a three-dimensional model in an embodiment of this application. Detailed Implementation

[0053] The present application will be further described in detail below with reference to the accompanying drawings and embodiments. Identical components are indicated by the same reference numerals. It should be noted that the terms "front," "rear," "left," "right," "upper," and "lower" used in the following description refer to directions in the accompanying drawings, and the terms "bottom surface," "top surface," "inner," and "outer" refer to directions toward or away from the geometric center of a specific component, respectively.

[0054] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0055] Hereinafter, the terms "first," "second," and other generic terms are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0056] refer to Figure 1 The image shows a specific implementation of an automatic detection method for structural defects in electrically tunable antennas according to this application, comprising:

[0057] S101. Multimode data of the electrically adjustable antenna is collected through a preset sensor group, and a three-dimensional model is generated by combining the structure of the electrically adjustable antenna.

[0058] S102. Map the three-dimensional model to a two-dimensional space according to the preset spatial mapping rules. Analyze the curvature distribution of the electrically tunable antenna surface to shift the pixels and fill in the missing areas in the two-dimensional space to obtain the two-dimensional model.

[0059] S103. Combining the thermal image data and two-dimensional model of the electrically adjustable antenna, the defects of the electrically adjustable antenna are identified through a preset defect identification model to obtain the first defect identification result.

[0060] S104. Extract the set of defect endpoints from the first defect identification result, calculate the connection probability between defect endpoints, establish connection edges between defect endpoints with connection probabilities greater than a preset probability threshold, construct a defect graph, map the defect graph onto a three-dimensional model, and obtain the second defect identification result of the electrically tunable antenna.

[0061] In the application of electrically tunable antennas, real-time detection of structural defects is required. The technical solution of this application improves detection efficiency and accuracy, and reduces missed and false detections through automated, high-precision defect detection. This embodiment collects multi-mode data of the electrically tunable antenna using a sensor array, fuses the multi-mode data to generate a three-dimensional model, reflecting the structural and physical characteristics of the antenna. The three-dimensional model is mapped to a two-dimensional space, and a two-dimensional model is obtained through pixel displacement and filling in missing areas. Combining thermal imaging data and the two-dimensional model, a defect identification model is used to initially identify the first defect. Defect endpoints are extracted, a defect map is constructed, and mapped to the three-dimensional model to obtain a complete second defect identification result, demonstrating the defect's position in three-dimensional space. By combining multimodal data fusion, two-dimensional model processing, defect identification, and defect map construction, and integrating various characteristic information of electrically tunable antennas, the accuracy of defect identification results can be improved, the probability of missed detection and false detection can be reduced, and a basis for defect analysis and maintenance can be provided, which helps to improve maintenance efficiency and quality. Non-contact, automated, and high-precision detection of structural defects in electrically tunable antennas can be achieved, which can effectively detect faults such as cracks, deformation, and overheating. This has significant practical implications for ensuring the stable operation of communication systems and realizing predictive maintenance.

[0062] In this embodiment, in S101, the sensor group includes, but is not limited to, a laser scanner, an infrared thermal imager, and a high-definition industrial camera. The laser scanner acquires the three-dimensional coordinate information of the electrically tunable antenna surface, the infrared thermal imager collects the temperature distribution data of the electrically tunable antenna, and the high-definition industrial camera captures the texture image of the electrically tunable antenna surface. The acquired multimodal data is filtered and preprocessed to remove noise and interference. The preprocessed multimodal data is then fused, and a three-dimensional model is constructed by combining the shape, size, and other structural characteristics of the electrically tunable antenna. Through multimodal data acquisition, the structural and physical characteristic information of the electrically tunable antenna can be comprehensively obtained, generating an accurate three-dimensional model, which provides an accurate structural basis for defect detection of the electrically tunable antenna.

[0063] Specifically, in S102, the three-dimensional model can fully reflect the structure of the electrically tunable antenna. However, the processing complexity of the three-dimensional model is high when performing defect identification. Mapping the three-dimensional model to two-dimensional space can simplify the data processing process and improve data processing efficiency. By analyzing the curvature distribution and displacing pixels, the deformation caused by the three-dimensional to two-dimensional mapping process can be compensated and corrected. Filling in missing areas can ensure the integrity of the two-dimensional model and obtain a more accurate two-dimensional model, thereby enabling defect analysis of the electrically tunable antenna surface. Converting the three-dimensional model into a two-dimensional model through spatial mapping can reduce the complexity of data processing. Combining pixel displacement and filling in missing areas improves the accuracy and integrity of the two-dimensional model, enabling the two-dimensional model to more realistically reflect the characteristics of the electrically tunable antenna surface and provide accurate data support for defect identification.

[0064] After constructing the two-dimensional model, the defects of the electrically adjustable antenna are identified by a pre-set defect identification model, based on the thermal image data and the two-dimensional model. The first defect identification result is obtained. Defects of the electrically adjustable antenna include, but are not limited to, cracks, deformation, and overheating. These defects can cause changes in the surface temperature distribution and appearance of the electrically adjustable antenna. The defect identification model includes, but is not limited to, a convolutional neural network model. The convolutional neural network model is trained using a large amount of defect data to obtain a pre-trained convolutional neural network model. The two-dimensional model, which incorporates thermal image data, is then input into the pre-trained convolutional neural network model, and the model outputs the first defect identification result. By combining thermal image data and the two-dimensional model for defect identification, and integrating the temperature and appearance information of the electrically adjustable antenna, defects that are difficult to detect with single data can be identified, thus improving the accuracy and comprehensiveness of defect identification.

[0065] Specifically, for S104, the initial defect identification results may contain incomplete defect endpoints or unclear connections between defects. Defect endpoints are extracted from the initial defect identification results. Based on the defect type, location, and morphology, the connection probability between any two defect endpoints is calculated. When the connection probability between two defect endpoints exceeds a preset probability threshold, a connection edge is established between these two endpoints, constructing a defect map. This defect map is then mapped onto a three-dimensional model according to spatial mapping rules, yielding the second defect identification result for the electrically tunable antenna. By constructing a defect map and mapping it to a three-dimensional model, the location, morphology, and connection relationships of defects in three-dimensional space can be comprehensively and intuitively displayed, resulting in complete and accurate defect identification results. Combining the relationships between defects can improve defect analysis and repair efficiency.

[0066] This application constructs a 3D model of an electrically tunable antenna based on multimodal data and the antenna structure. Following spatial mapping rules, pixel displacement and missing region filling are performed using curvature distribution to obtain a 2D model. Thermal imaging data and the 2D model are then combined to identify and detect defects in the electrically tunable antenna. A defect map is constructed based on the defect endpoints and mapped onto the 3D model to obtain complete defect identification results. By fusing multimodal and structural data, an accurate 3D model can be constructed. Pixel displacement and missing region filling using the surface curvature data of the electrically tunable antenna are applicable to parabolic and cylindrical geometric structures in electrically tunable antennas, reducing image distortion and information loss caused by the curved surface structure of the antenna and improving the quality of the 2D model. Identifying structural defects in the electrically tunable antenna using thermal imaging data and the 2D model can promptly detect problems such as localized overheating and loose connectors caused by electrical faults, improving the accuracy of defect identification results. By constructing a defect map and mapping it onto the 3D model, complete defect information can be identified, providing a basis for defect repair.

[0067] Furthermore, multimode data of the electrically tunable antenna is acquired through a pre-set sensor group, and a three-dimensional model is generated based on the structure of the electrically tunable antenna, including:

[0068] S201. Data is collected from the electrically tunable antenna from multiple perspectives using a preset sensor group. The collected data is then aligned to obtain multimodal data.

[0069] S202. Based on the structure of the electrically tunable antenna, generate corresponding structural constraints, and fuse the multi-mode data through the structural constraints to obtain fused data;

[0070] S203. Based on the structure of the electrically adjustable antenna, construct a three-dimensional model of the electrically adjustable antenna by fusing data.

[0071] This embodiment acquires data from multiple perspectives using a pre-set sensor group and performs coordinate alignment to obtain multimodal data, ensuring comprehensive and spatially consistent data. Corresponding structural constraints are generated based on the electrically tunable antenna structure, and the multimodal data is fused using these constraints to obtain fused data, guaranteeing its accuracy. A three-dimensional model of the electrically tunable antenna is constructed based on the fused data, following the antenna structure. The multimodal data acquired through multi-view acquisition comprehensively reflects information from all parts of the electrically tunable antenna. The three-dimensional model constructed based on the fused data and the antenna structure accurately displays the structural features of the electrically tunable antenna, providing an accurate model for defect detection and thus improving the efficiency and accuracy of defect detection.

[0072] In this embodiment S201, acquiring data from multiple perspectives can avoid blind spots from a single perspective and obtain more comprehensive information about the electrically tunable antenna. The sensor group includes, but is not limited to, laser scanners, infrared thermal imagers, and high-definition industrial cameras. Based on the structure of the electrically tunable antenna, multiple perspectives are set, including but not limited to the front, back, and side. Using the coordinate system of the front perspective as the reference coordinate system, the coordinates of the data acquired from other perspectives are transformed into the reference coordinate system for coordinate alignment. Multi-view acquisition can ensure the comprehensiveness of the acquired data and avoid missing structural information of the electrically tunable antenna. Coordinate alignment unifies data from different perspectives and modes into the same coordinate system, enabling multi-mode data to be fused in the same coordinate system.

[0073] Specifically, in S202, the electrically tunable antenna has a regular shape and its component connection relationships have corresponding structural characteristics. Based on the structure of the electrically tunable antenna, corresponding structural constraints are determined. The coordinate-aligned multimodal data is fused using a preset fusion model. The fusion model includes, but is not limited to, a neural network model. The neural network model is trained using a large amount of multimodal data to obtain a pre-trained neural network model. The coordinate-aligned multimodal data is input into the pre-trained neural network model, and the model fuses the multiple modal data according to their corresponding positions to obtain fused data. Combining structural constraints with data fusion improves the accuracy and reliability of the fused data, reduces the impact of noise and abnormal data, and makes the fused data more realistically and accurately reflect the structure of the electrically tunable antenna.

[0074] In S203, based on the structure of the electrically tunable antenna, the fused data is combined according to the corresponding structural positions to construct a three-dimensional spatial structure, resulting in a three-dimensional model. The constructed three-dimensional model is then optimized to remove redundant surfaces and lines. The three-dimensional model constructed based on the electrically tunable antenna structure can completely and accurately reflect the structure and characteristics of the electrically tunable antenna, providing a precise model basis for defect detection.

[0075] Furthermore, the three-dimensional model is mapped to a two-dimensional space according to a preset spatial mapping rule. The pixels are displaced by analyzing the curvature distribution of the electrically tunable antenna surface, and missing regions in the two-dimensional space are filled to obtain the two-dimensional model, including:

[0076] S301. According to the preset spatial mapping rules, the three-dimensional model is mapped to the two-dimensional space, the two-dimensional coordinates of each point of the three-dimensional model are calculated, and the three-dimensional model is unfolded according to the two-dimensional coordinates to obtain the first unfolded image.

[0077] S302. For the splicing region in the first unfolded image, perform boundary recognition, and combine the curvature distribution of the electrically tunable antenna surface to perform boundary coupling of the splicing region through pixel displacement to obtain the second image;

[0078] S303. The location and type of missing regions in the second image are identified by a preset missing region identification model, and the missing regions are filled in according to the type of missing regions to obtain a two-dimensional model.

[0079] like Figure 2As shown, in this embodiment, the coordinates of the 3D model in 2D space are calculated according to a preset spatial mapping rule. The 3D model is mapped to the 2D space, and the first unfolded image is obtained by unfolding according to the 2D coordinates of each point. By identifying the boundary of the splicing region of the first unfolded image and combining the curvature distribution with pixel displacement for boundary coupling, a second image is obtained. The missing regions of the second image are identified using a preset missing region identification model, and filled in according to different missing types to obtain a 2D model. By converting the 3D model into a 2D model, the complexity of data processing is reduced and the processing efficiency is improved. By splicing region boundary coupling and pixel displacement, deformation and errors in the mapping process are reduced. The obtained 2D image can reflect the surface features of the electrically tunable antenna. Filling in the missing regions can avoid missed defects due to missing information and improve the accuracy of defect detection results.

[0080] In this embodiment, in S301, a corresponding spatial mapping rule is set according to the shape and structure of the electrically adjustable antenna. For a regular rectangular electrically adjustable antenna area, an orthographic projection mapping rule is used, and for an electrically adjustable antenna with curvature, a cylindrical unfolding rule is used to unfold the curved part of the three-dimensional model to a two-dimensional plane along the axial direction. Each point on the three-dimensional model is transformed into a two-dimensional space according to the spatial mapping rule, and the two-dimensional coordinates of each point on the three-dimensional model are calculated. All points on the three-dimensional model are arranged in the two-dimensional space according to the two-dimensional coordinates to obtain the first unfolded image. By converting the three-dimensional model into a two-dimensional first unfolded image, the data processing complexity is reduced, and the defect detection efficiency can be improved.

[0081] Specifically, in S302, during the process of mapping the 3D model to 2D space, the splicing area may experience boundary mismatch due to the complexity of the 3D structure. The curvature distribution of the electrically tunable antenna surface reflects the degree of curvature of the electrically tunable antenna surface. When spatial mapping is performed in areas with high curvature, deformation will occur. The boundary of the splicing area is identified, and the pixel value is adjusted by combining the curvature value corresponding to the boundary, so that the two boundaries of the splicing area can be accurately aligned and coupled. By coupling the boundary of the splicing area, the splicing error generated during the 3D to 2D mapping process can be eliminated. Pixel displacement combined with curvature distribution can reduce image deformation caused by the curvature of the electrically tunable antenna surface. The resulting second image can more accurately reflect the surface features of the electrically tunable antenna.

[0082] After obtaining the second image through boundary coupling, the location and type of missing regions in the second image are identified using a pre-set missing region identification model. The missing regions are then filled in according to their type to obtain a two-dimensional model. During the mapping process from three-dimensional to two-dimensional, missing regions may appear in the two-dimensional image due to occlusion or limited sensor acquisition range. The location and type of these missing regions are identified using a pre-set missing region identification model. This model includes, but is not limited to, the U-Net model based on deep learning. The U-Net model is trained using a large amount of electrically adjustable antenna two-dimensional image data containing various missing conditions to obtain a pre-trained U-Net model. The second image is input into the pre-trained U-Net model, and the model outputs the location and corresponding type of the identified missing regions. Appropriate filling methods are used for different types of missing regions to obtain a two-dimensional model. By identifying missing regions and filling them in according to their type, the integrity and accuracy of the two-dimensional model can be improved, avoiding missed defects due to missing regions and improving the accuracy of defect detection results.

[0083] Furthermore, boundary recognition is performed on the stitched region in the first unfolded image. Combined with the curvature distribution of the electrically tunable antenna surface, the stitched region is bounded together by pixel displacement to obtain the second image, including:

[0084] S401. Based on the curvature distribution of the electrically adjustable antenna surface, calculate the curvature gradient and generate the curvature gradient field.

[0085] S402. Based on the curvature gradient field, perform boundary recognition on the stitching region in the first unfolded image, calculate the pixel displacement of the corresponding stitching region in combination with the curvature, and control the movement of the stitching region according to the pixel displacement.

[0086] S403. By analyzing the texture trend in the spliced ​​area after movement, the spliced ​​areas with consistent texture trends are bounded together to obtain the second image.

[0087] In this embodiment, in step S401, the curvature values ​​of each point on the surface of the electrically tunable antenna are calculated using a three-dimensional model of the antenna. For each point on the surface of the electrically tunable antenna, the curvature gradient is obtained by calculating the ratio of the curvature difference between adjacent points to the distance between adjacent points using the finite difference method. Based on the calculated curvature gradient of each point, the direction of curvature increase is represented by an arrow, and the length of the arrow represents the magnitude of the curvature gradient value, thus constructing a curvature gradient field. By constructing the curvature gradient field, the curvature change of the surface of the electrically tunable antenna can be reflected, and the trend of curvature change can be displayed. The speed and direction of curvature change can be accurately shown, providing data support for the boundary coupling of the splicing area.

[0088] Specifically, for S402, in the curvature gradient field, regions with drastic changes in curvature gradient correspond to the boundaries of the stitching region. Based on the defect identification accuracy requirements, a curvature gradient threshold is set, and regions with curvature gradients greater than the threshold are selected to obtain the stitching region. The corresponding region boundaries serve as the boundaries of the stitching region. For the identified stitching region, pixel displacement is calculated based on the curvature values ​​and curvature gradients of each point within the region. Positions with larger curvature and larger curvature gradients indicate more severe bending in three-dimensional space, resulting in greater deformation after mapping to two-dimensional space and requiring a larger displacement. The corresponding pixel displacement is obtained by multiplying the curvature values ​​and curvature gradients. Following the calculated pixel displacement, the stitching region is moved by the corresponding number of pixels along the direction of the curvature gradient arrow. Based on the curvature gradient field, accurate stitching region boundaries can be identified. Combining this with curvature calculations yields accurate displacement, and controlling the movement of the stitching region with the corresponding displacement effectively reduces deformation and makes the stitching region closer to the actual electrically tunable antenna structure.

[0089] Simultaneously, by analyzing the texture trends in the spliced ​​areas after movement, spliced ​​areas with consistent texture trends are bounded together to obtain a second image. Texture trends reflect the surface features of the electrically tunable antenna. Textures in the same area have continuous and consistent trends. For the spliced ​​areas after movement, the texture features of the electrically tunable antenna surface are calculated using the gray-level co-occurrence matrix, including the direction, spacing, and shape of the texture. The texture trends of adjacent spliced ​​areas are compared to identify spliced ​​areas with consistent texture trends. By weighted averaging of pixels at the boundaries, the boundaries of corresponding areas are aligned and merged to complete the corresponding boundary coupling, resulting in the second image. Boundary coupling of the spliced ​​areas through texture trend analysis can improve the coherence of the spliced ​​areas, eliminate splicing errors between spliced ​​areas, and the resulting second image can more accurately reflect the surface features of the electrically tunable antenna, providing accurate image evidence for defect detection.

[0090] Furthermore, the location and type of missing regions in the second image are identified using a preset missing region recognition model. Based on the type of missing region, the missing regions are filled in to obtain a two-dimensional model, including:

[0091] S501. The location and type of missing regions in the second image are identified by a preset missing region identification model. The missing regions include structural missing regions, texture missing regions, and boundary missing regions.

[0092] S502. For structural missing regions, the boundary is fitted by combining the curvature distribution of the electrically adjustable antenna surface, and the missing region is reconstructed to obtain the first filling region.

[0093] S503. For texture-type missing areas, the texture trend is predicted by analyzing the texture trend on the surface of the electrically tunable antenna to fill in the missing areas and obtain the second filling area.

[0094] S504. For regions with missing boundaries, by analyzing the connection probability between boundary endpoints and the boundary trend, a filling boundary is generated to obtain the third filling region.

[0095] S505. By combining the first filling region, the second filling region, and the third filling region, a two-dimensional model is obtained.

[0096] In this embodiment, in S501, the missing region identification model includes, but is not limited to, the U-Net model based on deep learning. The U-Net model is trained by a large amount of two-dimensional image data of electrically adjustable antennas containing various missing conditions to obtain a pre-trained U-Net model. The second image is input into the pre-trained U-Net model, and the model outputs the location and corresponding type of the identified missing region. Through model identification, the location and type of the missing region can be identified quickly and accurately, improving the efficiency and accuracy of the missing region identification results.

[0097] In S502, the structural missing region includes the geometric missing structure of the electrically tunable antenna. The curvature distribution reflects the bending of the surface of the electrically tunable antenna. The curvature data corresponding to the missing region is extracted, and the trend of the boundary of the missing region is analyzed based on the curvature data. The boundary of the missing region is fitted using the least squares method to determine the shape and range of the boundary. Based on the fitted boundary and the corresponding structural features, the boundary of the missing region is reconstructed. The boundary of the missing region is fitted and reconstructed by combining the curvature distribution, so that the filled missing region conforms to the overall structural features of the electrically tunable antenna.

[0098] In S503, the texture of the electrically tunable antenna surface has continuity and regularity. Texture features around the missing area are extracted, texture trends are analyzed based on texture features, and texture spacing is determined. Based on the surrounding texture trends, the texture trend in the missing area is predicted. Using a sample-based texture synthesis algorithm, the missing area is filled in according to the predicted trend, using the surrounding texture as samples, to obtain the second filled area. Combining texture trend with texture filling can make the filled missing area transition naturally with the surrounding texture, maintain the continuity and consistency of the texture, avoid texture breakage or disorder, and improve the accuracy of the two-dimensional model.

[0099] In S504, the location of the missing boundary region and the coordinates of the endpoints of the missing boundary are obtained from the missing boundary region. The connection probability between the two boundary endpoints is obtained by assigning values ​​to the direction, distance and orientation of the endpoints and the direction of the surrounding boundaries and calculating them in a weighted manner. If the connection probability is greater than the preset connection probability threshold, a filling boundary is generated in combination with the corresponding boundary trend to obtain the third filling region. The filling boundary generated in combination with the endpoint connection probability conforms to the overall trend of the boundary, which can improve the continuity and integrity of the missing region boundary and avoid the deviation in the judgment of the electrical tunable antenna structure defect caused by the missing boundary.

[0100] Specifically, in S505, the first filling region, the second filling region, and the third filling region are integrated into the second image to replace the original missing region. The integrated two-dimensional model fills in the missing region, which can completely and accurately reflect the surface features of the electrically tunable antenna, providing high-quality image data for defect detection of the electrically tunable antenna, thereby improving the accuracy of defect detection results.

[0101] Furthermore, combining the thermal image data and two-dimensional model of the electrically tunable antenna, defects in the electrically tunable antenna are identified using a pre-defined defect identification model, yielding the first defect identification result, including:

[0102] S601. Match the thermal image data and the two-dimensional model of the electrically adjustable antenna, extract the features of the thermal image data and the two-dimensional model respectively, and perform feature fusion to obtain the fused features;

[0103] S602. Based on the fusion characteristics, the defects of the electrically adjustable antenna are identified through a preset defect identification model to obtain the first defect identification result.

[0104] In this embodiment, thermal image data and two-dimensional model of electrically tunable antenna are matched and corresponded. The thermal image data and two-dimensional model reflect the temperature information and structural information of electrically tunable antenna, respectively. Features are extracted from the thermal image data and two-dimensional model and fused to obtain fused features. The fused features are input into a preset defect recognition model to identify the location, type and other information of defects, and obtain the first defect recognition result. Through feature fusion, temperature information and structural information can be integrated to more comprehensively reflect defect information. The model can quickly and accurately identify defect information.

[0105] In this embodiment, in S601, the thermal image data reflects the temperature distribution of the electrically adjustable antenna, and the two-dimensional model reflects the structural and textural features of the electrically adjustable antenna. The thermal image data and the two-dimensional model have a spatial correspondence. The thermal image data and the two-dimensional model are registered, and features are extracted based on the registered thermal image data and the two-dimensional model respectively. The extracted features are then fused. The fused features integrate temperature and structural information, which can more comprehensively reflect the state of the electrically adjustable antenna.

[0106] Specifically, in S602, the defect identification model includes, but is not limited to, a random forest model. The random forest model is trained using a large number of electrically adjustable antenna samples containing various defects. Each sample includes fused features and a corresponding defect label to obtain a pre-trained random forest model. The fused features are input into the pre-trained random forest model, and the model determines whether there are defects in each region of the electrically adjustable antenna and determines the type and location of the defects to obtain the first defect identification result. Through the preset defect identification model, the fused features can be analyzed quickly and automatically to accurately identify the location and type of defects, thereby improving the efficiency and accuracy of defect identification.

[0107] Furthermore, the thermal image data and the two-dimensional model of the electrically tunable antenna are matched and mapped, and the features of the thermal image data and the two-dimensional model are extracted respectively. These features are then fused to obtain the fused features, including:

[0108] S701. According to the structure of the electrically adjustable antenna, the thermal image data and the two-dimensional model are matched in position. In the thermal image data and the two-dimensional model after the position is matched, the temperature similarity of the corresponding position points is calculated. The thermal image data and the two-dimensional model are registered according to the temperature similarity to obtain the registered thermal image.

[0109] S702. Extract features from the registered thermal image and the two-dimensional model respectively to obtain the first feature set and the second feature set;

[0110] S703. The first feature set and the second feature set are weighted and fused to obtain the fused features.

[0111] In this embodiment, the structure of the electrically tunable antenna in S701 is fixed, and the positional relationship of its components is determined. Based on the structural characteristics of the electrically tunable antenna, and combined with landmark structures such as edges, mounting holes, and feed points, the positions of these landmark structures are identified in the thermal image data and two-dimensional model, and these positions are mapped. The thermal image data and two-dimensional model are then mapped to their corresponding positions to obtain the mapped thermal image data and two-dimensional model. In the mapped thermal image data and two-dimensional model, multiple corresponding position points are selected. For each corresponding position point, the normal temperature range of the corresponding position point is determined by statistically analyzing the thermal image data of the electrically tunable antenna. The matching degree between the actual temperature of a point in the thermal image data and the normal temperature range of the corresponding position in the two-dimensional model is calculated to obtain the temperature similarity of each position point. The position of the thermal image data is adjusted according to the temperature similarity. For areas with low temperature similarity, the translation and rotation angles of the thermal image data are adjusted to improve the temperature similarity of more corresponding position points. This adjustment process is repeated until the overall temperature similarity reaches a preset similarity threshold to obtain a registered thermal image. Through position correspondence and temperature similarity registration, the thermal image data can be accurately matched with the two-dimensional model, avoiding errors in feature extraction and fusion caused by position deviation, which would affect the defect identification results.

[0112] Specifically, in S702, the registration thermal image includes the temperature distribution characteristics of the electrically tunable antenna, and the two-dimensional model includes structural and texture features. Statistical features, regional features, and gradient features are extracted from the registration thermal image to represent temperature distribution features, while edge features, texture features, and shape features are extracted from the two-dimensional model to represent structural and texture features. The extracted features are then integrated to obtain the first feature set and the second feature set. Extracting features separately can fully explore the feature information of the registration thermal image and the two-dimensional model, providing a complete feature data foundation for feature fusion.

[0113] After extracting the features, the first feature set and the second feature set are weighted and fused to obtain the fused features. The first feature set and the second feature set have different importance in defect identification. The weights of each feature are set according to the importance of different features. The corresponding features in the first feature set and the second feature set are multiplied by their respective weights and then summed to obtain the fused features. By combining temperature distribution features and structural texture features through feature weighting fusion, comprehensive and accurate feature information can be provided, thereby improving the accuracy of defect identification.

[0114] Furthermore, a set of defect endpoints is extracted from the first defect identification result. By calculating the connection probability between defect endpoints, connection edges are established between defect endpoints with a connection probability greater than a preset probability threshold to construct a defect graph. The defect graph is then mapped onto a three-dimensional model to obtain the second defect identification result of the electrically tunable antenna, including:

[0115] S801. Identify the defect endpoints from the first defect identification results to obtain a set of defect endpoints;

[0116] S802. By calculating the connection probability between defect endpoints, establish connection edges between defect endpoints whose connection probability is greater than a preset probability threshold, and construct a defect graph.

[0117] S803. Map the defect map onto the three-dimensional model to obtain the second defect identification result of the electrically adjustable antenna.

[0118] This embodiment extracts defect endpoints from the first defect identification result, constructs a defect endpoint set, calculates the connection probability between endpoints, establishes connection edges for endpoints with probabilities greater than a threshold, constructs a defect graph by combining the correlation between defects, and maps the defect graph onto a 3D model to obtain the second defect identification result. By integrating the scattered defect endpoints into a defect graph through connection edges, defects can be identified by combining the correlation between defects, improving the accuracy of defect identification results, reducing false positives and false negatives, and improving the overall efficiency and accuracy of defect detection.

[0119] In this embodiment, in S801, the first defect identification result determines the defects existing on the electrically tunable antenna. The defect endpoints are the key feature points of the defects, including the two ends of the crack, the edge turning point of the deformation area, etc. By extracting the set of defect endpoints, we can focus on the key feature points of the defects, providing a data foundation for subsequent calculation of the connection probability between endpoints and construction of the defect map.

[0120] like Figure 3 As shown, by calculating the connection probability between defect endpoints, connecting edges are established between defect endpoints with a connection probability greater than a preset probability threshold to construct a defect graph. There will be connections between defect endpoints. When the endpoints of two cracks are very close and have the same direction, they belong to different parts of the same crack. For each pair of endpoints in the defect endpoint set, the connection probability between endpoints is calculated by assigning values ​​to each factor, considering the distance between endpoints, the defect type, and the defect direction, and then weighted and summing them. Based on the defect sample data and the defect detection accuracy requirements, a probability threshold is set. All endpoint pairs are traversed, and for endpoint pairs with a connection probability greater than the preset threshold, a connecting edge is established between them. All defect endpoints are treated as nodes, and connecting edges represent the relationships between nodes, thus constructing a defect graph. By constructing a defect graph, the connection relationships between defects can be displayed. Combining the correlations between defects for defect identification can improve the accuracy of defect identification results.

[0121] like Figure 4 As shown, the defect map is mapped onto a 3D model to obtain the second defect identification result of the electrically tunable antenna. The defect map is constructed on a 2D model, while the 3D model can reflect the actual spatial structure of the electrically tunable antenna. According to the spatial mapping rules, the corresponding positions of each node and connecting edge in the defect map in the 3D model are determined, and each endpoint in the defect map is mapped to the 3D model. After spatial mapping, the position, type, and connection relationship of the defect are mapped onto the 3D model to obtain the second defect identification result of the electrically tunable antenna. By mapping the defect map onto the 3D model, the accurate defect position and shape can be obtained, improving the accuracy of the defect identification result.

[0122] Furthermore, defect endpoints are identified from the first defect identification results, resulting in a set of defect endpoints, including:

[0123] S901. Using a preset endpoint identification model, identify defect endpoints from the first defect identification results to obtain a first endpoint set;

[0124] S902. Combining the preset endpoint criteria and curvature verification mechanism, the endpoints in the first endpoint set are filtered to obtain the second endpoint set;

[0125] S903. By analyzing the trend consistency of the endpoint curvature gradient, the endpoints in the second endpoint set are filtered to obtain the defect endpoint set.

[0126] In this embodiment, in S901, the endpoint recognition model includes, but is not limited to, a convolutional neural network model. The convolutional neural network model is trained using a large number of sample images containing various defective endpoints. The positions of the endpoints in these samples have been labeled, resulting in a pre-trained convolutional neural network model. The first defect recognition result is input into the pre-trained convolutional neural network model, and the model performs feature extraction and analysis on the input data to identify the defective endpoints, thus obtaining the first endpoint set. The endpoint recognition model can quickly and automatically identify defective endpoints, improving the efficiency and accuracy of endpoint recognition.

[0127] Specifically, in S902, endpoint criteria are formulated based on the endpoint characteristics of different defect types. For crack endpoints, the criteria include the gray value change rate of the line at the endpoint being greater than a preset threshold and the contrast between the endpoint and the surrounding non-defect area being greater than a certain value. For regional defect endpoints, the criteria include the endpoint being located on the boundary of the region. The curvature value at each endpoint in the first endpoint set is calculated, and each endpoint in the first endpoint set is compared with the preset endpoint criteria. Combined with the curvature verification results, endpoints that simultaneously meet the endpoint criteria and pass the curvature verification are selected to obtain the second endpoint set. Through the endpoint criteria and curvature verification mechanism, defect endpoints can be screened out, improving the accuracy of endpoint identification results.

[0128] Specifically, in S903, the trend consistency of the endpoint curvature gradient indicates that the endpoints of the same defect have similar curvature gradient change trends. For each endpoint in the second endpoint set, the curvature gradient vector is calculated. For endpoints of the same defect type, the cosine similarity of the curvature gradient vector is calculated. Based on the calculated cosine similarity, the trend consistency of the endpoint curvature gradient is analyzed, and the defect endpoints that belong to the same defect type and have a cosine similarity greater than a preset cosine similarity threshold are selected to obtain the defect endpoint set. By analyzing the trend consistency of the endpoint curvature gradient, endpoints that do not conform to the defect characteristics can be removed to obtain a more accurate defect endpoint set.

[0129] An automatic detection system for electrically tunable antenna structural defects, used to implement an automatic detection method for electrically tunable antenna structural defects, includes:

[0130] The 3D model building module collects multi-mode data of the electrically tunable antenna through a preset sensor group, and generates a 3D model based on the structure of the electrically tunable antenna.

[0131] The two-dimensional model mapping module maps the three-dimensional model to two-dimensional space according to the preset spatial mapping rules. It displaces pixels by analyzing the curvature distribution of the electrically tunable antenna surface and fills in the missing areas in the two-dimensional space to obtain the two-dimensional model.

[0132] The first defect identification module combines the thermal image data and two-dimensional model of the electrically adjustable antenna to identify defects in the electrically adjustable antenna through a preset defect identification model, and obtains the first defect identification result.

[0133] The second defect identification module extracts a set of defect endpoints from the first defect identification result, calculates the connection probability between defect endpoints, establishes connection edges between defect endpoints with a connection probability greater than a preset probability threshold, constructs a defect graph, maps the defect graph onto a three-dimensional model, and obtains the second defect identification result of the electrically tunable antenna.

[0134] In this embodiment, the 3D model construction module comprehensively collects multi-mode data of the electrically tunable antenna through a preset sensor group. It then processes and fuses this data in conjunction with the antenna's own structure to generate a 3D model that accurately reflects the antenna's structure and physical characteristics. Collecting multi-mode data ensures comprehensive information, and the generated 3D model provides a precise structural foundation for defect detection, reducing missed detections or misjudgments due to insufficient information. The 2D model mapping module maps the 3D model to 2D space according to preset rules. By analyzing the surface curvature distribution of the electrically tunable antenna, it adjusts pixel displacement and fills in missing areas in the 2D space, obtaining a complete and accurate 2D model. This transforms complex 3D data into a more easily processed 2D form, reducing the complexity of data processing and defect identification. By correcting deformations and filling in missing areas, the accuracy of the 2D model is ensured, reflecting the surface features of the electrically tunable antenna and providing high-quality 2D image data for defect identification.

[0135] Specifically, the first defect identification module matches and fuses the thermal image data of the electrically adjustable antenna with a two-dimensional model. Using a pre-defined defect identification model, it analyzes the fused features to identify defects in the electrically adjustable antenna and obtains the first defect identification result. Combining the temperature information reflected in the thermal image data with the structural texture information embodied in the two-dimensional model improves the accuracy and efficiency of the defect identification process. The second defect identification module extracts a set of defect endpoints from the first defect identification result, calculates the connection probability between endpoints, constructs a defect map, and maps the defect map onto a three-dimensional model to obtain the second defect identification result. By constructing a defect map and combining independent defect information with the correlation information between defects, mapping it to a three-dimensional model reveals the location, shape, and connection relationship of defects in three-dimensional space, providing accurate data support for defect identification and improving the accuracy of defect detection results.

[0136] The above description is merely a preferred embodiment of this application. The scope of protection of this application is not limited to the above embodiments. All technical solutions falling within the scope of this application's concept are within the scope of protection of this application. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of this application should also be considered within the scope of protection of this application.

Claims

1. An automatic detection method for structural defects in electrically tunable antennas, characterized in that, include: Multimode data of the electrically tunable antenna is collected by a pre-set sensor group, and a three-dimensional model is generated by combining the structure of the electrically tunable antenna. According to the preset spatial mapping rules, the three-dimensional model is mapped into a two-dimensional space, the two-dimensional coordinates of each point of the three-dimensional model are calculated, and the three-dimensional model is unfolded according to the two-dimensional coordinates to obtain the first unfolded image; Based on the curvature distribution of the electrically tunable antenna surface, the curvature gradient is calculated, and a curvature gradient field is generated. Based on the curvature gradient field, boundary recognition is performed on the stitching region in the first unfolded image, and the pixel displacement of the corresponding stitching region is calculated in combination with the curvature. The movement of the stitching region is controlled according to the pixel displacement. By analyzing the texture trend in the spliced ​​area after the movement, the spliced ​​areas with consistent texture trends are bounded together to obtain the second image; The location and type of missing regions in the second image are identified by a preset missing region identification model, and the missing regions are filled in according to the type of missing regions to obtain a two-dimensional model; By combining the thermal image data and two-dimensional model of the electrically adjustable antenna, the defects of the electrically adjustable antenna are identified through a preset defect identification model, and the first defect identification result is obtained. The defect endpoint set is extracted from the first defect identification result. By calculating the connection probability between the defect endpoints, a defect map is constructed. The defect map is then mapped onto a three-dimensional model to obtain the second defect identification result of the electrically tunable antenna.

2. The automatic detection method for structural defects in an electrically tunable antenna according to claim 1, characterized in that, The process involves acquiring multimode data of the electrically tunable antenna through a preset sensor group, and generating a three-dimensional model based on the structure of the electrically tunable antenna, including: Data is collected from the electrically tunable antenna from multiple perspectives using a pre-set sensor group. The collected data is then aligned to obtain multimodal data. Based on the structure of the electrically tunable antenna, corresponding structural constraints are generated, and the multi-mode data is fused using the structural constraints to obtain fused data. Based on the structure of the electrically tunable antenna, a three-dimensional model of the electrically tunable antenna is constructed using the fused data.

3. The automatic detection method for structural defects in an electrically tunable antenna according to claim 1, characterized in that, The step involves identifying the location and type of missing regions in the second image using a preset missing region identification model, and filling in the missing regions according to their type to obtain a two-dimensional model, including: The location and type of missing regions in the second image are identified using a preset missing region identification model. The missing regions include structural missing regions, texture missing regions, and boundary missing regions. For structurally missing regions, the boundary is fitted by combining the curvature distribution of the electrically tunable antenna surface, and the missing regions are reconstructed to obtain the first filling region. For texture-deficient areas, the texture trend on the surface of the electrically tunable antenna is analyzed to predict the texture trend and fill in the deficient areas to obtain a second filling area; For regions with missing boundaries, the third filling region is obtained by analyzing the connection probability between boundary endpoints and the boundary trend. By combining the first, second, and third filling regions, a two-dimensional model is obtained.

4. The automatic detection method for structural defects in an electrically tunable antenna according to claim 1, characterized in that, The thermal image data and two-dimensional model of the electrically tunable antenna are combined, and defects of the electrically tunable antenna are identified through a preset defect identification model to obtain a first defect identification result, including: The thermal image data and two-dimensional model of the electrically adjustable antenna are matched and corresponded. The features of the thermal image data and the two-dimensional model are extracted respectively, and the features are fused to obtain the fused features. Based on the fusion features, the defects of the electrically adjustable antenna are identified using a preset defect identification model to obtain the first defect identification result.

5. The automatic detection method for structural defects of electrically tunable antennas according to claim 4, characterized in that, The process involves matching the thermal image data of the electrically adjustable antenna with the two-dimensional model, extracting features from both the thermal image data and the two-dimensional model, and then fusing these features to obtain fused features, including: According to the structure of the electrically tunable antenna, the thermal image data and the two-dimensional model are matched in position. In the thermal image data and the two-dimensional model after the position is matched, the temperature similarity of the corresponding position points is calculated. Based on the temperature similarity, the thermal image data and the two-dimensional model are registered to obtain the registered thermal image. Feature extraction is performed on the registered thermal image and the two-dimensional model respectively to obtain a first feature set and a second feature set; The first feature set and the second feature set are weighted and fused to obtain the fused features.

6. The automatic detection method for structural defects in an electrically tunable antenna according to claim 1, characterized in that, The first defect identification result is used to extract a set of defect endpoints. By calculating the connection probability between the defect endpoints, a defect map is constructed. The defect map is then mapped onto a three-dimensional model to obtain the second defect identification result of the electrically tunable antenna, including: Defect endpoints are identified from the first defect identification results to obtain a set of defect endpoints; By calculating the connection probability between defect endpoints, connection edges are established between defect endpoints whose connection probability is greater than a preset probability threshold to construct a defect graph; The defect map is mapped onto a three-dimensional model to obtain the second defect identification result of the electrically tunable antenna.

7. The automatic detection method for structural defects of an electrically tunable antenna according to claim 6, characterized in that, The step of identifying defect endpoints from the first defect identification result to obtain a set of defect endpoints includes: Using a pre-defined endpoint identification model, defect endpoints are identified from the first defect identification results to obtain the first endpoint set; By combining the preset endpoint criteria and curvature verification mechanism, the endpoints in the first endpoint set are filtered to obtain the second endpoint set; By analyzing the trend consistency of the endpoint curvature gradient, the endpoints in the second endpoint set are filtered to obtain the defective endpoint set.

8. An automatic detection system for structural defects in electrically tunable antennas, characterized in that, An automatic detection method for structural defects in electrically tunable antennas as described in any one of claims 1 to 7, comprising: The 3D model building module collects multi-mode data of the electrically tunable antenna through a preset sensor group, and generates a 3D model based on the structure of the electrically tunable antenna. The two-dimensional model mapping module maps the three-dimensional model to a two-dimensional space according to a preset spatial mapping rule. It shifts the pixels by analyzing the curvature distribution of the electrically tunable antenna surface and fills in the missing areas in the two-dimensional space to obtain the two-dimensional model. The first defect identification module combines the thermal image data and two-dimensional model of the electrically adjustable antenna to identify defects in the electrically adjustable antenna through a preset defect identification model, and obtains the first defect identification result. The second defect identification module extracts a set of defect endpoints from the first defect identification result, constructs a defect map by calculating the connection probability between the defect endpoints, and maps the defect map onto a three-dimensional model to obtain the second defect identification result of the electrically tunable antenna.

Images