Concrete apparent quality defect detection method and device based on image recognition

Abstract

The invention relates to the technical field of concrete crack recognition, and discloses a concrete apparent quality defect detection method and device based on image recognition, and the method comprises the following steps: obtaining an original image of a concrete surface, and carrying out geometric correction and registration to obtain an orthoimage; performing multi-scale invariant feature extraction on the orthoimage to obtain defect feature data; segmenting the defect candidate region based on the defect feature data to obtain a defect candidate region mask; performing defect topological structure reconstruction on the defect candidate region mask and the defect feature data to obtain a defect topological network; based on the defect topology network, extracting a set and physical parameters of defects, and classifying the set and the physical parameters to obtain defect parameter data; and integrating the orthoimage, the defect topology network and the defect parameter data to generate a defect detection result report. According to the method, the potential defect area can be accurately positioned, the extraction precision of defect features can be improved, and the accuracy of defect classification is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of concrete crack identification, and in particular to a method and device for detecting concrete surface quality defects based on image recognition. Background Art

[0002] External defects in reinforced concrete are extremely common in engineering, and many reinforced concrete structures fail due to these defects. These defects can negatively impact concrete structures by reducing their strength, decreasing their durability, and increasing the risk of penetration.

[0003] Among traditional methods for detecting surface defects in concrete, manual inspection is the most basic. While this method relies on close-up visual inspection by inspectors, while intuitive, it suffers from low detection efficiency, significant safety risks associated with working at height, and is susceptible to operator experience and subjective judgment. With the advancement of computer vision and artificial intelligence technologies, a growing number of research and applications are exploring the use of image processing techniques for automated detection of surface defects in concrete. By capturing images of the concrete surface and utilizing image analysis techniques, various surface defects can be automatically identified and located.

[0004] However, image-based detection methods also face technical challenges. Concrete structures have complex surface textures and may be subject to a variety of interfering factors, such as rebar, formwork joints, stains, and coatings, creating a complex background. Furthermore, concrete surface defects vary greatly in form, size, color, and other characteristics, and multiple defects may coexist. Traditional image recognition methods have low accuracy for identifying multiple types of defects in complex backgrounds, and are prone to missed or false detections. Summary of the Invention

[0005] In view of this, the present application provides a method and device for detecting concrete surface quality defects based on image recognition to solve the problem of low recognition accuracy of multiple types of defects in complex backgrounds.

[0006] In a first aspect, a method for detecting apparent quality defects of concrete based on image recognition is provided, which is applied to a device for detecting apparent quality defects of concrete. The method comprises: Obtain the original image of the concrete surface, perform geometric correction and registration, and obtain the orthophoto; Perform multi-scale invariant feature extraction on orthophotos to obtain defect feature data; Based on the defect feature data, the defect candidate area is segmented to obtain the defect candidate area mask; Reconstruct the defect topology structure based on the defect candidate area mask and defect feature data to obtain the defect topology network; Based on the defect topology network, the defect set and physical parameters are extracted, the set and physical parameters are classified, and the defect parameter data is obtained; Integrate orthophotos, defect topology networks, and defect parameter data to generate defect detection result reports.

[0007] Through the above technical solutions, image distortion is eliminated, a standardized data foundation is provided for subsequent analysis, detection robustness is enhanced, preliminary defect positioning is achieved, the spatial correlation and development trend of defects are revealed, defect characteristics are quantified, and data support is provided for evaluation. Finally, multi-source data is integrated to generate a report, achieving a systematic presentation of detection results and improving the comprehensiveness and accuracy of defect detection.

[0008] Optionally, performing multi-scale invariant feature extraction on the orthophoto to obtain defect feature data includes: Perform multi-directional and multi-scale decomposition on the orthophoto to obtain a multi-directional and multi-scale coefficient matrix; Based on the multi-directional and multi-scale coefficient matrix, the scale-invariant eigenvector is calculated to generate the defect-invariant feature map; Combined with the preset concrete surface recognition results, material-guided feature compensation is performed on the defect-invariant feature map to obtain a compensated feature map; The view-invariant feature transformation based on Riemannian manifold learning is applied to the compensation feature map to obtain the view-invariant feature representation. The view-invariant feature representation is combined with the semantic information of concrete surface quality to perform context-aware feature enhancement and generate defect feature data.

[0009] The above technical solutions provide a basis for comprehensive analysis of image features, ensure the stability of features at different scales, enhance the ability to identify defects in specific materials, eliminate the impact of viewing angles and scene differences, improve the semantic expression ability of features, and make defect feature data more targeted and reliable.

[0010] Optionally, calculating the scale-invariant feature vector and generating the defect-invariant feature map includes: Calculate the coefficient differences between adjacent scales for the multi-directional and multi-scale coefficient matrix, construct the scale normalization factor and calculate the extreme points in the scale space to generate a multi-scale difference extreme point set; Based on the multi-scale differential extreme point set and the multi-directional multi-scale coefficient matrix, the defect direction consistency features are extracted and the direction consistency feature map is generated; Combining the directional consistency feature map and the multi-directional multi-scale coefficient matrix, we conduct projection profile analysis and calculate the curvature and linearity of the defect to generate a slenderness ratio continuity feature map. Determine the target sub-region based on the direction consistency feature map and the aspect ratio continuity feature map, calculate the feature statistics of the target sub-region, and generate a scale-invariant feature vector; A defect invariant feature map is generated based on the scale-invariant feature vector and the multi-directional multi-scale coefficient matrix.

[0011] Through the above technical solutions, multi-scale differential analysis is realized to highlight the defect characteristics; the directional consistency characteristics and the slenderness ratio continuity characteristics are extracted to characterize the directionality and geometric shape of the defects respectively, ensuring the stability and uniqueness of the characteristics, and providing accurate and reliable feature expression for subsequent defect analysis.

[0012] Optionally, the defect candidate area is segmented to obtain a defect candidate area mask including: Construct a multi-scale feature pyramid based on defect feature data; Based on the multi-scale feature pyramid, the suppression defect noise is identified and the suppression defect interference features are generated; According to the interference features of the defect suppression class, the defect and the background structure are distinguished to obtain the defect area. The defect area is segmented through the self-attention dense prediction network to obtain the defect candidate area mask.

[0013] Through the above technical solutions, feature information of different scales is integrated to improve the recognition ability of defects of different sizes, reduce the impact of irrelevant information on segmentation, achieve accurate segmentation through the self-attention intensive prediction network, generate high-quality defect candidate area masks, and provide an accurate basis for subsequent topology reconstruction.

[0014] Optionally, the defect candidate region mask and defect feature data are reconstructed into a defect topology structure to obtain a defect topology network including: According to the defect candidate area mask, the defect center line is extracted to obtain the defect skeleton map; Analyze defect skeleton diagrams, identify and classify defect breakpoints, and generate defect breakpoint data; Obtain surface stress field distribution based on the preset concrete surface 3D model and defect skeleton diagram, predict defect extension direction, and obtain stress field direction data; Using defect breakpoint data, stress field direction data and defect feature data, the optimal defect connection path is generated to form a connection defect skeleton; The defect topology network is obtained based on the connected defect skeleton.

[0015] Through the above technical solutions, the defect geometry is simplified, the discontinuity of the defect is clarified, the extension direction of the defect is predicted, the influence of mechanical factors on the development of the defect is considered, the spatial structure and development trend of the defect are fully presented, and a basis is provided for evaluating the degree of defect hazard.

[0016] Optionally, the surface stress field distribution is obtained based on the preset concrete surface 3D model and defect skeleton diagram, and the defect extension direction is predicted. The obtained stress field direction data includes: Based on the preset concrete surface recognition results, mechanical parameters are assigned to different areas of the concrete surface 3D model to generate a mechanical parameter model; Based on the concrete surface 3D model and mechanical parameter model, a structured mesh is constructed to generate a finite element model; Convert the defect skeleton diagram into geometric relationships in the finite element model, construct and solve the equilibrium equation based on the geometric relationships, and obtain surface stress field data; Based on the surface stress field data and the defect skeleton diagram, the stress intensity factor at the defect tip is calculated to generate stress intensity factor data; The stress intensity factor data and surface stress field data are used to predict the defect expansion direction and obtain the stress field direction data.

[0017] Through the above technical solutions, defect information is converted into mechanical problems, surface stress distribution is quantified, the possibility of defect expansion is evaluated, defect development prediction based on mechanical principles is realized, and the scientificity and accuracy of defect risk assessment are improved.

[0018] Optionally, the method further includes: Based on the preset light sensor data and scene reflectivity, the exposure parameters are calculated to obtain the original image and identify the target light area in the original image to generate a light area segmentation mask; Based on the illumination region segmentation mask and the original image, an adaptive illumination response function model is used to perform illumination compensation on different regions to obtain an illumination-balanced image. Based on the illumination equalization image, contrast enhancement is performed to generate preprocessed image data.

[0019] Through the above technical solutions, the original image acquisition is optimized, targeted illumination compensation is performed, and illumination balance is achieved, providing clear and accurate image data for subsequent feature extraction and analysis, reducing the interference of illumination factors on the detection results.

[0020] In a second aspect of the present application, a concrete surface quality defect detection system based on image recognition is provided, which includes an image acquisition module, an image feature extraction module, an image segmentation module, a defect reconstruction module, a defect parameter acquisition module, and a detection report generation module, wherein: An image acquisition module is configured to obtain an original image of the concrete surface, perform geometric correction and registration, and obtain an orthophoto; An image feature extraction module configured to perform multi-scale invariant feature extraction on the orthophoto to obtain defect feature data; An image segmentation module configured to segment the defect candidate area based on the defect feature data to obtain a defect candidate area mask; A defect reconstruction module is configured to reconstruct a defect topology structure based on a defect candidate region mask and defect feature data to obtain a defect topology network; A defect parameter acquisition module is configured to extract a set of defects and physical parameters based on a defect topology network, classify the set and physical parameters, and obtain defect parameter data; The inspection report generation module is configured to integrate orthophotos, defect topology networks, and defect parameter data to generate a defect inspection result report.

[0021] In the third aspect of the present application, an electronic device is provided, including a processor, a memory, a user interface and a network interface, the memory is used to store instructions, the user interface and the network interface are both used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device performs any of the methods described above.

[0022] In a fourth aspect of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores instructions. When the instructions are executed, any one of the methods described above is executed.

[0023] In summary, one or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages: The present invention obtains orthophotos through geometric correction and registration, eliminates image distortion, and provides a standardized data basis for subsequent analysis; multi-scale invariant feature extraction can capture defect characteristics at different resolutions and enhance detection robustness; defect candidate area segmentation generates masks to achieve preliminary defect positioning; topological structure reconstruction forms a defect topological network to reveal the spatial correlation and development trend of defects; geometric and physical parameters are extracted and classified, and defect characteristics are quantified to provide data support for evaluation; finally, multi-source data are integrated to generate a report, realize the systematic presentation of detection results, and improve the comprehensiveness and accuracy of defect detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is an exemplary system architecture diagram of a method for detecting apparent quality defects of concrete based on image recognition or a system for detecting apparent quality defects of concrete based on image recognition according to the present application; Figure 2 This is a flow chart of a method for detecting surface quality defects of concrete based on image recognition in an embodiment of the present application; Figure 3 yes Figure 2 A schematic flow chart of a specific implementation of step S202; Figure 4 yes Figure 2 A schematic flow chart of a specific implementation of step S203; Figure 5 Book Figure 2 A schematic flow chart of a specific implementation of step S204; Figure 6 This is another flow chart of a method for detecting surface quality defects of concrete based on image recognition in an embodiment of the present application; Figure 7 This is a module diagram of a concrete surface quality defect detection system based on image recognition in an embodiment of the present application; Figure 8 It is a structural diagram of an electronic device disclosed in an application embodiment.

[0025] Explanation of the accompanying drawings: 100, system architecture; 101, first terminal device; 102, second terminal device; 103, third terminal device; 104, network; 105, server; 701, image acquisition module; 702, image feature extraction module; 703, image segmentation module; 704, defect reconstruction module; 705, defect parameter acquisition module; 706, detection report generation module; 707, image illumination compensation module; 801, processor; 802, communication bus; 803, user interface; 804, network interface; 805, memory. DETAILED DESCRIPTION

[0026] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this application, not all of the embodiments.

[0027] In the description of the embodiments of this application, words such as "for example" or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "for example" or "for instance" in the embodiments of this application should not be construed as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "for example" or "for instance" is intended to present the relevant concepts in a concrete manner.

[0028] In the description of the embodiments of the present application, the term "multiple" means two or more. For example, multiple systems refer to two or more systems, and multiple screen terminals refer to two or more screen terminals. In addition, the terms "first" and "second" are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized.

[0029] This embodiment discloses a method and device for detecting concrete surface quality defects based on image recognition. Figure 1 A schematic diagram of an exemplary system architecture of an embodiment of a method for detecting apparent concrete quality defects based on image recognition or a system for detecting apparent concrete quality defects based on image recognition to which the present application may be applied is shown.

[0030] like Figure 1 As shown, system architecture 100 may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. Network 104 is a medium for providing communication links between terminal devices 101, 102, 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links or fiber optic cables.

[0031] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, and 103, such as model training applications, video recognition applications, web browser applications, social platform software, etc.

[0032] Terminal devices 101, 102, and 103 can be hardware or software. When terminal devices 101, 102, and 103 are hardware, they can be various electronic devices with display screens, including but not limited to smartphones, tablet computers, e-book readers, MP3 (Moving Picture Experts Group Audio Layer III, Moving Picture Experts Group Audio Layer 3) players, MP4 (Moving Picture Experts Group Audio Layer IV, Moving Picture Experts Group Audio Layer 4) players, laptop computers, and desktop computers, etc. When terminal devices 101, 102, and 103 are software, they can be installed in the electronic devices listed above. They can be implemented as multiple software or software modules (for example, multiple software or software modules used to provide distributed services), or they can be implemented as a single software or software module. No specific limitation is made here.

[0033] When terminals 101, 102, and 103 are hardware, they may also be equipped with a video capture device. The video capture device may be any device capable of capturing video, such as a camera, a sensor, and the like. Users can use the video capture device on terminals 101, 102, and 103 to capture video.

[0034] The server 105 may be a server that provides various services, such as a background server that processes data displayed on the terminal devices 101, 102, and 103. The background server may analyze and process the received data, and may feed back the processing results (e.g., recognition results) to the terminal device.

[0035] It should be noted that a server can be either hardware or software. When a server is hardware, it can be implemented as a distributed server cluster consisting of multiple servers, or as a single server. When a server is software, it can be implemented as multiple software programs or software modules (for example, multiple software programs or software modules used to provide distributed services), or as a single software program or software module. This is not specifically limited here.

[0036] It should be understood that Figure 1 The number of terminal devices, networks, and servers in the above description is merely illustrative. Any number of terminal devices, networks, and servers may be used as needed. In particular, if target data does not need to be acquired remotely, the above system architecture may not include a network, but may instead include only terminal devices or servers.

[0037] Figure 2 This is a flow chart of a method for detecting surface quality defects of concrete based on image recognition disclosed in an embodiment of the present application. Figure 2As shown, this embodiment includes: Step S201: Acquire the original image of the concrete surface, perform geometric correction and registration, and obtain an orthophoto.

[0038] For example, a composite sensor consisting of a camera and an infrared thermal imager mounted on a drone is used to inspect and photograph the concrete surface of the box girder. The drone flies along a preset route and collects distorted images of the bridge deck, web, and bottom plate. The original images are imported into the inspection platform to eliminate the perspective distortion caused by the drone's tilted shooting. The images are corrected for radial and tangential distortion to generate a preliminarily corrected image sequence. The theoretical coordinates of the box girder surface are extracted as a reference plane, and adjacent images are aligned to eliminate mismatched point pairs. The aligned image sequence is projected into a unified geographic coordinate system to generate an orthorectified mapping relationship. Each image is resampled and stitched together to generate a complete orthophoto of the box girder concrete surface.

[0039] Step S202 : performing multi-scale invariant feature extraction on the orthophoto to obtain defect feature data.

[0040] For example, the generated orthophoto is fed into a Gabor filter bank to generate directional-scale subband images. Local Binary Patterns (LBP) are calculated for each subband image to extract texture periodicity features. HOG (Histogram of Oriented Gradient) features are extracted in crack regions to capture the directional consistency of gradients at crack edges. Local contrast features are calculated for each subband to quantify the brightness difference between the defect and the background. The LBP, HOG, and contrast features at different scales are concatenated into a high-dimensional feature vector. Principal Components Analysis (PCA) is then used to reduce the dimensionality to 256, yielding defect feature data.

[0041] Step S203 : Segment the defect candidate area based on the defect feature data to obtain a defect candidate area mask.

[0042] For example, defect feature data is fed into a Feature Pyramid Network (FPN) to generate feature maps at multiple scales. Statistical analysis is used to determine the feature distribution of non-defective areas, identify and suppress pseudo-defect noise such as rebar shadows and formwork seams, and generate a noise mask. Structure-texture decomposition is performed on the orthophoto to separate structural information (such as crack orientation and hole outlines) from texture information (concrete aggregate distribution) in the defective area, enhancing the defect edges. The enhanced multi-scale features are fed into an improved U-Net network. The encoder captures long-range dependencies (such as crack continuity within a 1-meter range). The decoder upsamples and fuses skip connection features (for example, concatenating 512×512 pixel shallow texture features with 64×64 pixel deep semantic features) to generate a defect candidate region mask (white represents defects, black represents background).

[0043] Step S204 , reconstructing the defect topology structure based on the defect candidate region mask and the defect feature data to obtain a defect topology network.

[0044] For example, the defect candidate region mask is skeletonized to generate a skeleton graph, preserving the geometric orientation of the original crack. Curvature analysis is performed on the skeleton graph to generate a breakpoint data file, annotating the coordinates and curvature value of each breakpoint (reflecting the severity of the crack's deflection). The bridge design BIM (Building Information Modeling) is imported, and the three-dimensional geometric parameters of the inspected section are extracted. A finite element model is constructed and a structured mesh is created. The crack skeleton graph is converted into a geometric crack in the finite element model. An equivalent bending load is applied to obtain the stress intensity factor at the crack tip, and the crack propagation direction is predicted. Using breakpoints as nodes and the predicted crack propagation direction as edge weights, the optimal connection path (e.g., connecting adjacent crack endpoints with a spacing of less than 50 cm) is searched within the local area, with breakpoints with consistent propagation directions being prioritized (weight coefficient 0.8). A defect topology network containing multiple paths is generated, with each edge annotated with a connection probability and predicted propagation rate. Node attributes include crack width and characteristic type (load crack / shrinkage crack).

[0045] Step S205 , based on the defect topology network, extracting defect sets and physical parameters, classifying the sets and physical parameters, and obtaining defect parameter data.

[0046] For example, the defect topology network (including crack skeleton diagrams, connection paths, node attributes, etc.) was imported into the bridge inspection data analysis platform and spatially matched with the BIM 3D model. For example, the crack network node numbered WB-07 on the web was mapped to the area with mileage pile numbers K1+230 to K1+245 and a height of 1.2 to 1.8 meters in the BIM. This ensured that the location and type of each defect node (e.g., structural cracks, shrinkage cracks) corresponded to the connection relationship in the network.

[0047] Defect set parameters include: defect number, bridge component (such as web, base plate, pier), spatial coordinates (X, Y, Z), node degree (number of connected edges) in the defect network, and clustering coefficient (degree of defect aggregation). For example, a node degree of 4 at an intersection in the base plate crack network indicates that the node is connected to four cracks and is a highly correlated defect node.

[0048] Physical parameters include: geometric parameters (such as crack length, crack width, crack depth, crack curvature), mechanical parameters (such as stress intensity factor, surrounding concrete strength), and development parameters (such as historical expansion rate and current expansion direction).

[0049] An automated script traverses the defect topology network and extracts parameters for each node and connecting edge. For example, for the 12 nodes in the crack network of web WB-07, the length and width of each crack and the curvature of the connecting paths between nodes are automatically extracted.

[0050] A three-tier classification system is used: by defect type: structural defects (such as load cracks and cracks near prestressed ducts); non-structural defects (such as shrinkage cracks and surface cracks). Severity classification: Level I (high risk): crack width ≥ 0.3mm and stress intensity factor > 70% of the material's fracture toughness; Level II (medium risk): crack width 0.1-0.3mm with bifurcations or intersections; Level III (low risk): surface cracks or fine cracks < 0.1mm in width. Classification by development trend: stable; developing. For example, a 0.4mm wide crack in the base plate with a stress intensity factor of 85% of the material's fracture toughness is classified as a "structural defect - Level I - developing"; while a 0.05mm wide crack on the web surface is classified as a "non-structural defect - Level III - stable."

[0051] The classified parameters are organized into a structured table. Each defect contains a unique number, classification label, geometric parameters, mechanical parameters, development parameters and spatial location link (click to view the corresponding BIM area).

[0052] Step S206 , integrating the orthophoto, defect topology network, and defect parameter data to generate a defect detection result report.

[0053] For example, orthophotos, defect topology networks, and defect parameter data are input into a software platform integrating a geographic information system and BIM, achieving precise spatial alignment of images, networks, and parameters. A defect location layer is overlaid on the orthophoto, with different symbols used to indicate defect types: red lines indicate cracks, yellow patches indicate spalling, and blue circles indicate honeycombs. Clicking on a marked point triggers a pop-up window displaying defect parameters, including the defect number, location coordinates, and a high-definition partial image. A parameter table is generated for all defects, categorized by "location-type-parameter-level-recommendation." By integrating multi-source data, an inspection report is generated that integrates visual presentation, parameter analysis, and decision-making recommendations.

[0054] Figure 3 yes Figure 2 A specific implementation flow diagram of step S202 is shown in FIG. Figure 3 As shown, this embodiment includes: Step S301 : performing multi-directional and multi-scale decomposition on the orthophoto to obtain a multi-directional and multi-scale coefficient matrix.

[0055] For example, oblique images acquired by drones are processed using Pix4Dmapper to generate orthophotos, eliminating and correcting perspective distortion. The orthophotos are grayscaled and converted into single-channel images to reduce interference from the color channel and highlight the grayscale differences between the cracks and the concrete matrix. Multiple decomposition scales are set to cover the common crack widths found in bridges. Multiple Gabor filters (4 scales × 6 directions) are designed and convolved with the orthophotos to generate multiple direction-scale subband coefficient matrices. Local extrema are calculated for each subband matrix, and the grayscale gradient amplitude of the crack pixels is extracted. The coefficient matrices for the same direction but different scales are superimposed to generate a multi-directional and multi-scale coefficient matrix.

[0056] Step S302: Calculate the scale-invariant feature vector and generate a defect-invariant feature map.

[0057] For example, the difference between adjacent scale subband matrices is calculated based on the multi-directional, multi-scale coefficient matrix to identify local extreme points. Within a 16×16 pixel neighborhood around each extreme point, the gradient histograms in eight directions are calculated to form a 128-dimensional SIFT feature vector. The feature vector is normalized to eliminate the effects of illumination variations. The dominant direction of the gradient within the neighborhood is calculated, and the feature vector is rotated to that dominant direction to ensure that the feature is insensitive to rotation (for example, the weld features remain consistent at different shooting angles). The 128-dimensional vectors of all feature points are mapped to a lower-dimensional space (e.g., 32 dimensions) to obtain a defect-invariant feature map.

[0058] Step S303 : combining the preset concrete surface recognition result, performing material-guided feature compensation on the defect-invariant feature map to obtain a compensated feature map.

[0059] For example, a bridge BIM model was imported to extract the 3D coordinates of the prestressed ducts, generating a ROI (Region of Interest) mask for the buffer zone surrounding the ducts. Concrete rebound testing and coring analysis were used to determine material parameters surrounding the ducts, including aggregate particle size and cement paste content. SIFT feature extraction was performed on orthophotos of the duct ROI region. It was found that the feature vectors of fine cracks along the ducts exhibited a weak response in the gradient direction histogram (due to low edge gradients caused by fine aggregate). The feature point density was 30% lower than that of normal areas, making them susceptible to misclassification as non-defects. A material weight matrix W was applied to the defect feature vectors to enhance the feature dimensions associated with the ducts. The weight of the gradient direction dimension (0° along the duct axis) was increased from 1.0 to 1.8 to suppress noise features in the perpendicular direction (90°). The weight of the grayscale contrast dimension was reduced from 1.0 to 0.5 to reduce grayscale interference caused by material differences. This yielded a compensated feature map.

[0060] Step S304 , applying a view-invariant feature transformation based on Riemannian manifold learning to the compensated feature map to obtain a view-invariant feature representation, combining the view-invariant feature representation with the concrete surface quality semantic information to perform context-aware feature enhancement and generate defect feature data.

[0061] For example, the 12-dimensional feature vector of each pixel is constructed as a 3×3 covariance matrix, forming a point on a Riemann manifold. For the same concrete area photographed from different perspectives, the features are projected into the tangent space using exponential and logarithmic mapping, eliminating the manifold differences caused by perspective changes and obtaining a perspective-invariant feature representation. The bridge BIM is imported to obtain the component type and structural functional area corresponding to each pixel, and a semantic mask is generated. An attention mechanism is used to fuse the perspective-invariant feature representation with the semantic mask: for the web area (a high-stress sensitive area), the directional gradient feature weight is enhanced; for defects in the repair layer near the support (a high-incidence area of ​​historical defects), the scale energy ratio feature weight is increased. This generates defect feature data that includes spatial location, material properties, and structural semantics.

[0062] Optionally, calculating the scale-invariant feature vector and generating the defect-invariant feature map further includes: The coefficient differences between adjacent scales of the multi-directional and multi-scale coefficient matrix are calculated, the scale normalization factor is constructed, and the extreme points in the scale space are calculated to generate a multi-scale differential extreme point set; based on the multi-scale differential extreme point set and the multi-directional and multi-scale coefficient matrix, the defect directional consistency features are extracted to generate a directional consistency feature map; combining the directional consistency feature map and the multi-directional and multi-scale coefficient matrix, a projection profile analysis is performed and the curvature and linear measurement of the defect are calculated to generate a slenderness ratio continuity feature map; based on the directional consistency feature map and the slenderness ratio continuity feature map, the target sub-region is determined, the characteristic statistics of the target sub-region are calculated, and a scale-invariant feature vector is generated; based on the scale-invariant feature vector and the multi-directional and multi-scale coefficient matrix, a defect invariant feature map is generated.

[0063] For example, the main direction and directional concentration parameters are extracted, and a directional consistency scoring function is constructed to quantify the degree of directional consistency within a local area. A non-maximum suppression algorithm is applied to retain candidate points with high directional consistency, resulting in a directional consistency feature map. Combining the directional consistency feature map with the multi-directional and multi-scale coefficient matrix, the slenderness ratio and continuity characteristics of the defect are calculated: projected profile analysis is performed along the main direction and perpendicular direction of the defect to calculate the width-to-length ratio. An adaptive threshold segmentation algorithm is used to extract the skeleton line and boundary of the defect. A curve fitting method is applied to calculate the local curvature and linearity of the defect. A feature score combining slenderness ratio and continuity is constructed to generate a slenderness ratio continuity feature map.

[0064] Scale-invariant feature vectors are constructed for the key points in the directional consistency feature map and the aspect ratio continuity feature map: with the key point as the center, local feature regions are extracted, and the size of the region is proportional to the feature scale of the key point. The region is divided into multiple sub-regions, and the feature statistics of each sub-region (such as gradient direction histogram) are calculated. The vectors are normalized to improve their robustness to illumination changes. Principal component analysis is applied to reduce the dimensionality, retaining the most discriminative feature dimensions, and generating scale-invariant feature vectors.

[0065] Verify and optimize the scale-invariant feature vector: Test the scale invariance of features on artificially synthesized multi-scale defect samples, calculate the feature matching stability index, analyze the variation pattern of features at different scales, establish a scale variation compensation model, adjust the feature extraction parameters through iterative optimization to maximize the scale-invariant performance, and apply the optimized feature vector to the multi-directional and multi-scale coefficient matrix to generate a defect-invariant feature map.

[0066] Figure 4 yes Figure 2 A specific implementation flow diagram of step S203 is shown in FIG. Figure 4 As shown, this embodiment includes: Step S401: construct a multi-scale feature pyramid based on defect feature data.

[0067] For example, based on defect feature data, such as edges and corners, these features are used as the bottom-level data of the feature pyramid. Pooling and convolution operations are used to generate feature maps of different scales. The bottom-level feature map is subjected to maximum pooling to obtain a low-resolution feature map of the previous level, while the semantic information of the feature is maintained through the convolution operation. For example, in the image processing of the main tower concrete surface, although the resolution of the second-level feature map is reduced, it still retains key information such as the direction of the cracks and the outline of the holes, and the pooling operation suppresses local noise interference; the pooling and convolution operations are then repeated on the second-level feature map to construct the third, fourth, and fifth-level feature maps in turn. The feature maps at each level are integrated to form a complete multi-scale feature pyramid.

[0068] Step S402 : identifying suppression-type defect noise based on a multi-scale feature pyramid, and generating suppression-type defect interference features.

[0069] For example, the feature maps at each level of the multi-scale feature pyramid are checked one by one to study the distribution patterns of defect features at different scales. The feature maps at each level of the multi-scale feature pyramid are used as model input, and the model outputs the probability value of each pixel belonging to noise. For the identified noise area, its feature information in the multi-scale feature pyramid is extracted, including the pixel value distribution, gradient features, texture features, etc. of the feature maps at each level, to generate a suppression defect interference feature vector. These feature vectors are fused to form a multi-dimensional suppression defect interference feature vector, which is used to suppress and eliminate noise in the subsequent defect segmentation and identification process.

[0070] In step S403, the defect and the background structure are distinguished according to the interference feature of the defect suppression class to obtain the defect area, and the defect area is segmented through the self-attention dense prediction network to obtain a defect candidate area mask.

[0071] For example, the interference feature vector of the suppression defect is fused with the original features of the multi-scale feature pyramid for analysis. Each pixel is weighted using the grayscale variance, texture roughness and other information in the interference feature vector. A self-attention intensive prediction network based on the U-Net architecture is built. The network is trained using the marked defect images from the historical bridge inspection as training data. The image of the defect area obtained by preliminary distinction is input into the trained self-attention intensive prediction network, and the network outputs the probability value of each pixel belonging to the defect. A threshold is set to binarize the probability map, and pixels with probability values ​​greater than the threshold are marked as defect areas, and pixels with probability values ​​less than the threshold are marked as background areas. The isolated noise points in the binarized image are removed, and small holes in the defect area are filled to make the outline of the defect area smoother and more complete, and finally a defect candidate area mask is generated.

[0072] Figure 5 yes Figure 2 A specific implementation flow diagram of step S204 is shown in FIG. Figure 5 As shown, this embodiment includes: Step S501 : extract the defect centerline based on the defect candidate area mask to obtain a defect skeleton map.

[0073] For example, a defect candidate region mask is input into an image processing system. The system iteratively removes pixels at the crack edge, gradually thinning the crack width to a single pixel and connecting the crack breakpoints. After thinning and breakpoint processing, an image containing only single-pixel lines is obtained, known as the defect skeleton map.

[0074] Step S502 , analyzing the defect skeleton graph, identifying and classifying defect breakpoints, and generating defect breakpoint data.

[0075] For example, the lines in the defect skeleton map are traversed pixel by pixel, and breakpoints are identified based on the connectivity of the pixels. Lines are further refined to highlight breakpoint features and ensure accurate breakpoint identification. Breakpoints are categorized as endpoints, bifurcations, and intersections based on their location in the skeleton map and the direction of surrounding lines. A data record is generated for each breakpoint, containing information such as its location coordinates, category, and the direction of adjacent cracks. The data for all breakpoints is aggregated to form a complete defect breakpoint dataset.

[0076] Step S503: acquiring the surface stress field distribution according to the preset concrete surface three-dimensional model and defect skeleton diagram, predicting the defect extension direction, and obtaining stress field direction data.

[0077] For example, a three-dimensional model of the concrete surface of the bridge is imported into the finite element analysis software. The coordinate information of the defect skeleton diagram is accurately matched with the three-dimensional model so that the crack position is accurately mapped to the model surface. The three-dimensional model is solved using the finite element analysis software to calculate the stress field distribution on the concrete surface. Combining the defect skeleton diagram and the stress field distribution, the stress direction at the crack endpoints and bifurcation points is analyzed. The predicted extension direction of each defect breakpoint is recorded to generate stress field direction data. The data contains information such as the position coordinates of the breakpoint, the predicted extension direction angle, and the stress value at the corresponding position. The stress field direction data of all breakpoints are aggregated to form a complete data set.

[0078] Step S504 , using the defect breakpoint data, stress field direction data and defect feature data, an optimal defect connection path is generated to form a connection defect skeleton.

[0079] For example, defect breakpoint data, stress field direction data, and defect feature data are imported into a geographic information system (GIS) platform. Weights are assigned to each data point, and potential connecting edges are constructed between adjacent nodes based on the stress field direction data and the probability of defect extension, using the defect breakpoints as nodes. The attributes of each edge include information such as length, directional difference, and defect feature similarity. An optimal connection path is searched within the graph model composed of nodes and edges, with preference given to paths that pass through high-weight nodes (such as intersections) and have high overall edge attribute scores. The optimal connection path is visualized within the GIS platform, forming a connection defect skeleton.

[0080] Step S505 , obtaining a defect topology network based on the connection defect skeleton.

[0081] For example, nodes in the defect skeleton are supplemented with information such as defect type, severity rating, and historical development data; edges in the defect skeleton are added with connection reliability scores and impact coefficients. Using nodes as the foundation and edges as the connecting links, a defect topology network is constructed using professional network analysis software. Based on the connections between nodes, the network structure is automatically laid out, presenting clear logical relationships. Node types and severity are distinguished by different colors, sizes, and shapes, such as large red circular nodes indicating high-risk structural cracks. Line thickness and color depth indicate edge connection reliability.

[0082] Optionally, obtaining the surface stress field distribution based on the preset concrete surface three-dimensional model and defect skeleton diagram, predicting the defect extension direction, and obtaining stress field direction data also includes: Based on the preset concrete surface recognition results, mechanical parameters are assigned to different areas of the concrete surface three-dimensional model to generate a mechanical parameter model; based on the concrete surface three-dimensional model and the mechanical parameter model, a structured mesh is constructed to generate a finite element model; the defect skeleton diagram is converted into geometric relationships in the finite element model, and the equilibrium equations are constructed and solved based on the geometric relationships to obtain surface stress field data; based on the surface stress field data and the defect skeleton diagram, the stress intensity factor at the defect tip is calculated to generate stress intensity factor data; using the stress intensity factor data and the surface stress field data, the expansion direction of the defect is predicted to obtain stress field direction data.

[0083] For example, the three-dimensional mesh model is converted into a structured mesh suitable for finite element analysis, the mesh is simplified and optimized, the balance between geometric details and computational efficiency is maintained, a mapping relationship between node displacement and unit stress is established, an overall stiffness matrix is ​​constructed, load conditions and boundary constraints are defined, the effects of the building's own weight, wind load, and temperature changes are simulated, and the model parameters are integrated to generate a simplified finite element model.

[0084] The defect skeleton is converted into a geometric discontinuity in the finite element model. The extended finite element method is used to deal with the stress singularity introduced by the defect. The equilibrium equation is solved iteratively, and the nodal displacement field and unit stress field are calculated. An adaptive mesh refinement algorithm is applied to improve the calculation accuracy in high stress gradient areas such as the defect tip. The complete surface stress distribution is obtained through post-processing analysis, and surface stress field data is generated.

[0085] Identify the tip position in the defect skeleton, establish a local polar coordinate system at the defect tip, extract the stress distribution near the tip, calculate the stress intensity factor, evaluate the critical conditions for defect extension based on the defect extension criterion under the mixed mode, calculate the energy release rate of each defect tip, and generate stress intensity factor data.

[0086] Multiple theoretical methods such as the maximum circumferential stress criterion, the maximum energy release rate criterion, and the minimum strain energy density criterion are applied to calculate the candidate expansion directions. Combined with the fracture characteristics of building materials, a weight model is constructed to integrate the prediction results of multiple criteria. The Monte Carlo simulation method is used to consider the uncertainty of material parameters and generate a probability distribution of the expansion direction. The predicted expansion path vector field is generated along the defect skeleton to obtain the stress field direction data.

[0087] Figure 6 This is another flow chart of a method for detecting surface quality defects of concrete based on image recognition in an embodiment of the present application, such as Figure 6 As shown, this embodiment includes: Step S601 : Calculate exposure parameters based on preset illumination sensor data and scene reflectivity, obtain an original image, identify a target illumination area in the original image, and generate an illumination area segmentation mask.

[0088] For example, a light sensor collects ambient lighting information, including light intensity, spectral distribution, and direction, in the concrete surface quality inspection area. Based on the light intensity and scene reflectivity, a dynamic exposure adjustment algorithm calculates exposure parameters to generate the original image. The original multi-exposure image set is analyzed, and a gradient consistency detection algorithm is used to identify areas of uneven lighting within the image, generating a segmentation mask for the illuminated area.

[0089] Step S602 : Based on the illumination region segmentation mask and the original image, an adaptive illumination response function model is used to perform illumination compensation processing on different regions to obtain an illumination-balanced image.

[0090] For example, the illumination problem of different areas is confirmed by combining the illumination area segmentation mask with the original image. These areas have low brightness, blurred details, and it is difficult to distinguish defects such as concrete cracks; the overly bright spots in the reflective area lead to the loss of texture information and may cover up defects such as rust and scratches. According to the characteristics of different illumination areas, an adaptive illumination response function is constructed. For the shadow area, the logarithmic transformation function is used to enhance the brightness, and for the reflective area, the gamma correction function is used to reduce the brightness. Traverse each pixel point of the original image, determine the area to which the pixel belongs based on the illumination area segmentation mask, and call the corresponding illumination response function for processing. Recombine the pixel values ​​after illumination compensation to generate an illumination-balanced image.

[0091] Step S603: performing contrast enhancement based on the illumination-equalized image to generate pre-processed image data.

[0092] For example, the illumination-balanced image is divided into multiple 10×10 pixel regions, and the pixel grayscale distribution within each region is analyzed. For cracked areas, the grayscale range is automatically stretched to increase the difference between the crack pixel grayscale value and the concrete background grayscale value. For honeycombed areas, the grayscale contrast between the defect edge and the normal concrete surface is enhanced to highlight the defect outline. After contrast enhancement, the image is normalized to uniformly map pixel values ​​to a standard grayscale range of 0–255. The resulting preprocessed image data can be directly used in subsequent defect identification and analysis processes.

[0093] Figure 7 This is a module diagram of a concrete surface quality defect detection system based on image recognition in an embodiment of the present application. Figure 7 As shown, the system includes: an image acquisition module 701; an image feature extraction module 702; an image segmentation module 703; a defect reconstruction module 704; a defect parameter acquisition module 705; a detection report generation module 706; and an image illumination compensation module 707, wherein: The image acquisition module 701 is configured to obtain an original image of the concrete surface, perform geometric correction and registration, and obtain an orthophoto; An image feature extraction module 702 is configured to perform multi-scale invariant feature extraction on the orthophoto to obtain defect feature data; The image segmentation module 703 is configured to segment the defect candidate area based on the defect feature data to obtain a defect candidate area mask; The defect reconstruction module 704 is configured to reconstruct the defect topology structure based on the defect candidate region mask and the defect feature data to obtain a defect topology network; The defect parameter acquisition module 705 is configured to extract the defect set and physical parameters based on the defect topology network, classify the set and physical parameters, and obtain defect parameter data; The inspection report generation module 706 is configured to integrate the orthophoto, the defect topology network and the defect parameter data to generate a defect inspection result report; Optionally, the image feature extraction module 702 is further configured to: Perform multi-directional and multi-scale decomposition on the orthophoto to obtain a multi-directional and multi-scale coefficient matrix; Based on the multi-directional and multi-scale coefficient matrix, the scale-invariant eigenvector is calculated to generate the defect-invariant feature map; Combined with the preset concrete surface recognition results, material-guided feature compensation is performed on the defect-invariant feature map to obtain a compensated feature map; The view-invariant feature transformation based on Riemannian manifold learning is applied to the compensation feature map to obtain the view-invariant feature representation. The view-invariant feature representation is combined with the semantic information of concrete surface quality to perform context-aware feature enhancement and generate defect feature data.

[0094] Optionally, the image feature extraction module 702 is further configured to: Calculate the coefficient differences between adjacent scales for the multi-directional and multi-scale coefficient matrix, construct the scale normalization factor and calculate the extreme points in the scale space to generate a multi-scale difference extreme point set; Based on the multi-scale differential extreme point set and the multi-directional multi-scale coefficient matrix, the defect direction consistency features are extracted and the direction consistency feature map is generated; Combining the directional consistency feature map and the multi-directional multi-scale coefficient matrix, we conduct projection profile analysis and calculate the curvature and linearity of the defect to generate a slenderness ratio continuity feature map. Determine the target sub-region based on the direction consistency feature map and the aspect ratio continuity feature map, calculate the feature statistics of the target sub-region, and generate a scale-invariant feature vector; A defect invariant feature map is generated based on the scale-invariant feature vector and the multi-directional multi-scale coefficient matrix.

[0095] Optionally, the image segmentation module 703 is further configured to: Construct a multi-scale feature pyramid based on defect feature data; Based on the multi-scale feature pyramid, the suppression defect noise is identified and the suppression defect interference features are generated; According to the interference features of the defect suppression class, the defect and the background structure are distinguished to obtain the defect area. The defect area is segmented through the self-attention dense prediction network to obtain the defect candidate area mask.

[0096] Optionally, the defect reconstruction module 704 is further configured to: According to the defect candidate area mask, the defect center line is extracted to obtain the defect skeleton map; Analyze defect skeleton diagrams, identify and classify defect breakpoints, and generate defect breakpoint data; Obtain surface stress field distribution based on the preset concrete surface 3D model and defect skeleton diagram, predict defect extension direction, and obtain stress field direction data; Using defect breakpoint data, stress field direction data and defect feature data, the optimal defect connection path is generated to form a connection defect skeleton; The defect topology network is obtained based on the connected defect skeleton.

[0097] Optionally, the defect reconstruction module 704 is further configured to: Based on the preset concrete surface recognition results, mechanical parameters are assigned to different areas of the concrete surface 3D model to generate a mechanical parameter model; Based on the concrete surface 3D model and mechanical parameter model, a structured mesh is constructed to generate a finite element model; Convert the defect skeleton diagram into geometric relationships in the finite element model, construct and solve the equilibrium equation based on the geometric relationships, and obtain surface stress field data; Based on the surface stress field data and the defect skeleton diagram, the stress intensity factor at the defect tip is calculated to generate stress intensity factor data; The stress intensity factor data and surface stress field data are used to predict the defect expansion direction and obtain the stress field direction data.

[0098] Optionally, the system further includes an image illumination compensation module 707 configured to: Based on the preset light sensor data and scene reflectivity, the exposure parameters are calculated to obtain the original image and identify the target light area in the original image to generate a light area segmentation mask; Based on the illumination region segmentation mask and the original image, an adaptive illumination response function model is used to perform illumination compensation on different regions to obtain an illumination-balanced image. Based on the illumination equalization image, contrast enhancement is performed to generate preprocessed image data.

[0099] It should be noted that the above embodiments provide systems that implement their functions using only the division of the above functional modules as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the system and method embodiments provided in the above embodiments are based on the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0100] This embodiment also discloses an electronic device, referring to Figure 8 The electronic device may include: at least one processor 801 , at least one communication bus 802 , a user interface 803 , a network interface 804 , and at least one memory 805 .

[0101] The communication bus 802 is used to implement the connection and communication between these components.

[0102] The user interface 803 may include a display screen (Display) and a camera (Camera). Optionally, the user interface 803 may also include a standard wired interface and a wireless interface.

[0103] The network interface 804 may optionally include a standard wired interface or a wireless interface (such as a WI-FI interface).

[0104] The processor 801 may include one or more processing cores. Using various interfaces and circuits, the processor 801 connects to various components within the server. It executes instructions, programs, code sets, or instruction sets stored in the memory 805, as well as accesses data stored in the memory 805, to perform various server functions and process data. Optionally, the processor 801 may be implemented using at least one of the following hardware forms: a digital signal processing (DSP), a field-programmable gate array (FPGA), or a programmable logic array (PLA). The processor 801 may integrate one or a combination of a central processing unit (CPU), a graphics processing unit (GPU), and a modem. The CPU primarily processes the operating system, user interface, and application programs; the GPU is responsible for rendering and drawing content displayed on the display screen; and the modem handles wireless communications. It is understood that the modem may also be implemented independently of the processor 801 and implemented on a separate chip.

[0105] Among them, the memory 805 may include a random access memory (RAM) or a read-only memory (Read-Only Memory). Optionally, the memory 805 includes a non-transitory computer-readable storage medium. The memory 805 can be used to store instructions, programs, codes, code sets or instruction sets. The memory 805 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as a touch function, a sound playback function, an image playback function, etc.), instructions for implementing the above-mentioned various method embodiments, etc.; the data storage area may store data involved in the above-mentioned various method embodiments, etc. The memory 805 may also be optionally at least one storage device located away from the aforementioned processor 801. As Figure 8 As shown, the memory 805 as a computer storage medium may include an operating system, a network communication module, a user interface module, and an application program for a method for detecting surface quality defects of concrete based on image recognition.

[0106] exist Figure 8 In the electronic device shown, the user interface 803 is mainly used to provide an input interface for the user and obtain data input by the user; and the processor 801 can be used to call an application stored in the memory 805 for a method for detecting concrete surface quality defects based on image recognition. When executed by one or more processors 801, the electronic device executes one or more methods as described in the above embodiments.

[0107] It should be noted that for the aforementioned method embodiments, for simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should be aware that this application is not limited by the order of the actions described, because according to this application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required for this application.

[0108] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0109] In the several embodiments provided in this application, it should be understood that the disclosed devices can be implemented in other ways. For example, the device embodiments described above are merely schematic, such as the division of units, which is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some service interface, and the indirect coupling or communication connection of devices or units can be electrical or other forms.

[0110] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0111] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0112] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of this application, or the portion that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory 805 and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the various embodiments of the method of this application. The aforementioned memory 805 includes various media that can store program code, such as a USB flash drive, a mobile hard drive, a magnetic disk, or an optical disk.

[0113] The above is only an exemplary embodiment of the present disclosure and cannot be used to limit the scope of the present disclosure. That is, any equivalent changes and modifications made according to the teachings of the present disclosure are still within the scope of the present disclosure. After considering the disclosure of the specification, those skilled in the art will easily think of other embodiments of the present disclosure. This application is intended to cover any variations, uses or adaptive changes of the present disclosure, which follow the general principles of the present disclosure and include common knowledge or customary technical means in the technical field that are not recorded in the present disclosure. The description and examples are to be regarded as exemplary only, and the scope and spirit of the present disclosure are defined by the claims.

Claims

1. A method for detecting surface quality defects of concrete based on image recognition, characterized in that: Applied to a concrete surface quality defect detection platform, the method includes: Obtain the original image of the concrete surface, perform geometric correction and registration, and obtain the orthophoto; Performing multi-scale invariant feature extraction on the orthophoto to obtain defect feature data; Segmenting the defect candidate area based on the defect feature data to obtain a defect candidate area mask; Reconstructing a defect topology structure on the defect candidate region mask and the defect feature data to obtain a defect topology network; Extracting a set of defects and physical parameters based on the defect topology network, classifying the set and physical parameters, and obtaining defect parameter data; The orthophoto, the defect topology network and the defect parameter data are integrated to generate a defect detection result report.

2. The method for detecting apparent quality defects of concrete based on image recognition according to claim 1, characterized in that: The performing multi-scale invariant feature extraction on the orthophoto to obtain defect feature data includes: Performing multi-directional and multi-scale decomposition on the orthophoto to obtain a multi-directional and multi-scale coefficient matrix; Calculating scale-invariant eigenvectors based on the multi-directional and multi-scale coefficient matrix to generate a defect-invariant feature map; Combining the preset concrete surface recognition result, performing material-guided feature compensation on the defect-invariant feature map to obtain a compensated feature map; A perspective-invariant feature transformation based on Riemannian manifold learning is applied to the compensation feature map to obtain a perspective-invariant feature representation, and the perspective-invariant feature representation is combined with concrete surface quality semantic information to perform context-aware feature enhancement to generate the defect feature data.

3. The method for detecting apparent quality defects of concrete based on image recognition according to claim 2, characterized in that: Calculating the scale-invariant feature vector and generating the defect-invariant feature map includes: Calculating coefficient differences between adjacent scales for the multi-directional multi-scale coefficient matrix, constructing a scale normalization factor and calculating extreme points in the scale space to generate a multi-scale differential extreme point set; Extracting defect directional consistency features based on the multi-scale differential extreme value point set and the multi-directional multi-scale coefficient matrix, and generating a directional consistency feature map; Combining the directional consistency feature map and the multi-directional multi-scale coefficient matrix, performing a projection profile analysis and calculating the curvature and linearity of the defect to generate a slenderness ratio continuity feature map; determining a target subregion according to the direction consistency feature map and the aspect ratio continuity feature map, calculating feature statistics of the target subregion, and generating a scale-invariant feature vector; The defect invariant feature map is generated according to the scale invariant feature vector and the multi-directional multi-scale coefficient matrix.

4. The method for detecting apparent quality defects of concrete based on image recognition according to claim 1, characterized in that: The segmenting of the defect candidate area to obtain the defect candidate area mask includes: Constructing a multi-scale feature pyramid based on the defect feature data; Based on the multi-scale feature pyramid, identifying suppression defect noise and generating suppression defect interference features; The defect and the background structure are distinguished according to the suppression defect interference feature to obtain the defect area, and the defect area is segmented by the self-attention dense prediction network to obtain the defect candidate area mask.

5. The method for detecting apparent quality defects of concrete based on image recognition according to claim 1, characterized in that: The step of reconstructing the defect topology structure based on the defect candidate region mask and the defect feature data to obtain a defect topology network includes: Extracting the defect centerline according to the defect candidate area mask to obtain a defect skeleton map; Analyzing the defect skeleton diagram, identifying and classifying defect breakpoints, and generating defect breakpoint data; Obtain surface stress field distribution based on the preset concrete surface 3D model and defect skeleton diagram, predict defect extension direction, and obtain stress field direction data; generating an optimal defect connection path using the defect breakpoint data, the stress field direction data, and the defect feature data to form a connection defect skeleton; The defect topology network is obtained based on the connection defect skeleton.

6. The method for detecting apparent quality defects of concrete based on image recognition according to claim 5, characterized in that: The method of obtaining the surface stress field distribution based on the preset concrete surface three-dimensional model and defect skeleton diagram, predicting the defect extension direction, and obtaining stress field direction data includes: Based on the preset concrete surface recognition result, assigning mechanical parameters to different areas of the concrete surface three-dimensional model to generate a mechanical parameter model; constructing a structured grid based on the three-dimensional concrete surface model and the mechanical parameter model to generate a finite element model; Converting the defect skeleton diagram into a geometric relationship in the finite element model, constructing and solving equilibrium equations according to the geometric relationship, and obtaining surface stress field data; Calculating the stress intensity factor of the defect tip based on the surface stress field data and the defect skeleton diagram to generate stress intensity factor data; The stress intensity factor data and the surface stress field data are used to predict the expansion direction of the defect and obtain the stress field direction data.

7. The method for detecting apparent quality defects of concrete based on image recognition according to claim 1, characterized in that: The method further comprises: Calculating exposure parameters based on preset light sensor data and scene reflectivity to obtain the original image and identifying the target light area in the original image to generate a light area segmentation mask; Based on the illumination region segmentation mask and the original image, an adaptive illumination response function model is used to perform illumination compensation processing on different regions to obtain an illumination-balanced image; Based on the illumination-balanced image, contrast enhancement is performed to generate pre-processed image data.

8. A concrete surface quality defect detection system based on image recognition, characterized in that: It includes image acquisition module, image feature extraction module, image segmentation module, defect reconstruction module, defect parameter acquisition module and inspection report generation module, among which: The image acquisition module is configured to obtain an original image of the concrete surface, perform geometric correction and registration, and obtain an orthophoto; The image feature extraction module is configured to perform multi-scale invariant feature extraction on the orthophoto to obtain defect feature data; The image segmentation module is configured to segment the defect candidate area based on the defect feature data to obtain a defect candidate area mask; The defect reconstruction module is configured to reconstruct the defect topology structure based on the defect candidate area mask and the defect feature data to obtain a defect topology network; The defect parameter acquisition module is configured to extract a set of defects and physical parameters based on the defect topology network, classify the set and physical parameters, and obtain defect parameter data; The detection report generation module is configured to integrate the orthophoto, the defect topology network and the defect parameter data to generate a defect detection result report.

9. An electronic device, characterized in that: The electronic device comprises a processor, a memory, a user interface and a network interface, wherein the memory is used to store instructions, the user interface and the network interface are both used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device executes the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores instructions, and when the instructions are executed, the method according to any one of claims 1 to 7 is executed.

Images