Fabricated retaining wall defect identification method and system based on image identification

Abstract

The invention relates to the field of earth wall defect recognition, and discloses an assembled retaining wall defect recognition method and system based on image recognition, and the method comprises the steps: obtaining multi-view image data of an assembled retaining wall, and constructing a defect recognition image data set in combination with a boundary detail enhancement mechanism and a region illumination compensation strategy; performing edge guide feature extraction on the defect identification image data set, and constructing an image deep feature model based on a boundary context fusion network; judging whether the response intensity change of the image deep feature model in the crack region reaches a preset threshold value or not through an edge response enhancement mechanism, and if yes, marking that the crack region has potential defects; utilizing a multi-scale morphological structure analysis method to carry out contrast perception optimization on the corrected crack area; and based on a defect identification result, combining a component positioning mechanism and component historical operation and maintenance data to perform severity grading evaluation on the defect. The method has the advantage of improving the sensitivity to the marginal area.

Description

Technical Field

[0001] The present invention relates to the field of earth retaining wall defect identification, and specifically to a method and system for identifying defects of prefabricated retaining walls based on image recognition. Background Technique

[0002] In the existing technology for identifying defects of prefabricated retaining walls, image recognition algorithms mostly focus on identifying significant defects such as large-area cracks and obvious deformations, but there are obvious deficiencies in dealing with tiny displacement cracks caused by construction errors or material aging. When the defect is at the edge of the retaining wall joint, in an area with small light contrast or complex background texture, the existing algorithms are prone to misjudge these tiny cracks as texture noise, resulting in missed detections. This problem is particularly crucial in early structural anomaly detection because tiny cracks are often precursors to the development of structural diseases. If they cannot be accurately identified in a timely manner, it will pose potential hazards to subsequent operation and maintenance and safety assessment. Therefore, it is necessary to design a method and system for identifying defects of prefabricated retaining walls based on image recognition that can improve the sensitivity to edge regions. Summary of the Invention

[0003] Aiming at the deficiencies of the existing technology, the present invention provides a method and system for identifying defects of prefabricated retaining walls based on image recognition, which has the advantage of improving the sensitivity to edge regions and solves the problems in the above background technique.

[0004] To achieve the above purpose of improving the sensitivity to edge regions, the present invention provides the following technical solution: A method for identifying defects of prefabricated retaining walls based on image recognition, including the following steps: Obtain multi-view image data of the prefabricated retaining wall, and combine the boundary detail enhancement mechanism and the regional light compensation strategy to construct a defect recognition image data set; Extract edge-guided features from the defect recognition image data set, construct an image deep feature model based on the boundary context fusion network, and introduce an asymmetric feature aggregation structure to improve the model's perception ability of tiny edge defects; Based on the edge response enhancement mechanism, judge whether the change in the response intensity of the image deep feature model in the crack region reaches a preset threshold. If so, mark that there are potential defects in the edge region, and combine the contrast adjustment mechanism between the crack and the background to dynamically correct the preliminary marking result; According to the corrected edge defect region, apply the multi-scale morphological structure analysis method to perform contrast perception optimization processing on the corrected result, extract the geometric parameters of the crack region, and dynamically generate the defect identification result; Based on the defect identification result, combine the component positioning mechanism and the component historical operation and maintenance data to classify and evaluate the severity of the defect, and generate a defect analysis report.

[0005] Preferably, the process of constructing a defect recognition image dataset is as follows: Collect multi-view image samples under different times, perspectives, and lighting conditions; Conduct a brightness distribution analysis on the multi-view image samples. For the problem of uneven lighting, dynamically evaluate the local brightness features of the multi-view images, and adaptively determine the regional lighting compensation window; On the basis of lighting compensation, adopt a boundary detail enhancement mechanism to enhance the cracks and edge details in the multi-view images, and at the same time fuse multiple data enhancement methods; Uniformly standardize the preprocessed and enhanced multi-view images to construct a standardized defect recognition image dataset.

[0006] Preferably, the process of constructing an image deep feature model based on a boundary context fusion network is as follows: Input the defect recognition image dataset into the boundary context fusion grid; The boundary context fusion network uses multi-scale feature channels to perform layer-by-layer feature fusion on the crack edges and surrounding structures, generating a fusion feature map containing the defect context relationship; Introduce an asymmetric feature aggregation structure to perform differential feature learning on the crack targets and background regions in the fusion feature map; In the training stage of the boundary context fusion network, based on the manually annotated crack edge regions in the defect images; After training is completed, output an image deep feature model with weak defect perception ability.

[0007] Preferably, the process of determining whether the response intensity change of the image deep feature model in the crack region reaches a preset threshold is as follows: Input the standardized defect recognition image into the trained image deep feature model to obtain a deep feature map containing edge structure information; Conduct edge response analysis based on the deep feature map, combine the edge gradient change and boundary context features, and extract the crack regions where cracks exist; Within the crack region, calculate the local response intensity of the crack region; Set a preset threshold for weak crack response, and compare the local response intensity of the crack region with the threshold; If the local response intensity of the crack region meets the requirements, it is determined that the crack region responds; If the local response intensity of the crack region does not meet the requirements, it is determined that the crack region does not respond.

[0008] Preferably, the process of dynamically correcting the preliminary annotation results in combination with the contrast adjustment mechanism between cracks and the background is as follows: Identify the crack area whose response intensity meets the requirement in the output of the deep feature model of the image, and generate a local image block with the crack area as the center; In the local image block, the local contrast value between the crack area and the adjacent background area is calculated; According to the changing trend of local contrast with the boundary position, the boundary clarity of the preliminary marked area is evaluated; If the contrast value is lower than the set threshold, indicating that the boundary is blurred, the boundary trend is dynamically determined based on the pixel gradient direction information and neighborhood grayscale statistical characteristics in the local image block; Based on the boundary trend, edge extension is performed on the preliminary marked boundary to correct the defect contour; The image gradient consistency judgment algorithm is introduced in the dynamic correction process to evaluate the consistency between the correction boundary and the main direction of the crack.

[0009] Preferably, the process of optimizing contrast perception of the corrected crack area using the multi-scale morphological structure analysis method is as follows: Based on the dynamically corrected crack area, multiple size structural elements are selected, and morphological dilation and erosion operations are performed on the crack area at each scale to extract the crack contour; A local contrast perception mechanism is introduced to evaluate the local contrast of the crack area after dilation and corrosion at each scale, calculate the brightness difference distribution between the crack outline and the background area, and evaluate the degree of improvement of the local contrast. The crack contours obtained at each scale are weighted fused to remove low-contrast redundant boundary information, and a unified crack contour map is obtained after fusion optimization. Based on the unified crack contour map, the geometric parameters of the crack area are extracted.

[0010] Preferably, the process of dynamically generating defect identification results is: According to the geometric parameters of the optimized extracted crack area, the defect identification rules and boundary constraints are set to determine the defect type and characteristic boundary; Map the two-dimensional coordinates of the crack area boundary to the image coordinate system in the original image acquisition process, establish the position index relationship of the defect area on the original image, and generate a defect identification label containing defect location information, type and geometric description; According to the attribute characteristics in the defect identification label, combined with the defect label type, structural complexity and crack density, the defect severity assessment index is calculated and a preliminary severity level is assigned; The defect identification labels are mapped to generate a defect identification result layer, and the defect identification result layer is overlaid on the original image through an image overlay mechanism to generate a defect identification result.

[0011] Preferably, the process of generating a defect analysis report is: Map the defect recognition results to the spatial components of the precast retaining wall through the component positioning mechanism, and obtain the numbers and spatial position coordinates of the corresponding structural components; Combined with the historical operation and maintenance database, retrieve whether there are similar defects and development trends in the same components during past inspection cycles; Based on the change trend of defect geometric parameters and historical evolution records, comprehensively judge the current severity of the defects; Generate a defect analysis report according to the defect level, component function importance and operation and maintenance cycle requirements.

[0012] A defect recognition system for precast retaining walls based on image recognition, comprising: Image acquisition module: Obtain multi-angle image data of the precast retaining wall, and perform light compensation and boundary enhancement processing; Image modeling module: Build an image deep feature model based on the boundary context fusion network to realize defect area recognition; Defect detection module: Realize preliminary defect annotation and dynamic correction through the edge response enhancement and contrast analysis mechanism; Structure analysis module: Combine multi-scale morphological analysis methods to extract defect geometric features and generate identification results; Analysis output module: Comprehensively evaluate the defects based on historical data and structural information, and output a defect analysis report.

[0013] Compared with the prior art, the present invention provides a defect recognition method and system for precast retaining walls based on image recognition, having the following beneficial effects: 1. By collecting image samples under different times, perspectives and lighting conditions, and combining the regional light compensation and boundary detail enhancement mechanism, the brightness uniformity and edge sharpness of defect areas such as cracks in the image are significantly improved, thereby constructing a higher-quality and representative defect recognition image dataset, laying a stable foundation for subsequent model training and feature extraction.

[0014] 2. By introducing the edge guidance mechanism and the boundary context fusion network, it is possible to realize the multi-scale fusion extraction of the edge details of the defect area and the surrounding structure, and at the same time use the asymmetric feature aggregation structure to enhance the response sensitivity of the model to the fine-grained defect edges, improving the model's recognition ability for micro-cracks and fuzzy defects.

[0015] 3. Effectively identify potential crack areas with prominent feature responses in the image through the edge response enhancement mechanism, and the contrast adjustment mechanism further corrects the initial annotation results with blurred edges, effectively improving the accuracy and integrity of the crack boundary, reducing false detections and missed detections, and improving the credibility of the annotation results.

[0016] 4. By processing with multi-scale morphological structures, the saliency of the crack contour boundary can be enhanced. Combining with the local contrast perception mechanism, background interference and redundant information can be effectively removed, the accurate extraction quality of crack geometric features can be improved, providing accurate support for defect location, identification and classification, and finally realizing the visual identification of defects.

[0017] 5. By constructing the positioning relationship between the defect recognition results and specific retaining wall components and integrating historical operation and maintenance records, quantitative evaluation and trend judgment of current defects can be realized, facilitating the dynamic control of the structural health status, assisting subsequent maintenance decisions and life cycle management, and improving the scientific nature and efficiency of the operation and maintenance management of prefabricated retaining walls. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a schematic diagram of the method of the present invention; Figure 2 It is a schematic diagram of the structure of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0020] Embodiment 1: Please refer to Figure 1 As shown, a method for identifying defects in prefabricated retaining walls based on image recognition according to an embodiment of the present invention includes the following steps: S1: Obtain multi-view image data of the prefabricated retaining wall, and combine the boundary detail enhancement mechanism and the regional light compensation strategy to construct a defect recognition image data set.

[0021] The process of constructing the defect recognition image data set in S1 is as follows: Collect multi-view image samples at different times, viewpoints and lighting conditions; select representative sample paragraphs of prefabricated retaining walls and set multiple shooting points; collect images during early morning, noon, evening and other time periods to obtain image samples under different natural light intensities; adjust the shooting angles, including front views, side views, and oblique views, to construct multi-view coverage; use an image acquisition device with fixed resolution and exposure parameters to ensure unified sampling quality; uniformly number the original collected image samples and store them in the original image database; Perform brightness distribution analysis on multi-view image samples. For the problem of uneven illumination, dynamically evaluate the local brightness characteristics of multi-view images and adaptively determine the regional illumination compensation window; perform local brightness histogram analysis on each captured image and calculate the gray-scale distribution offset of the image; use a sliding window to evaluate the brightness mean and standard deviation in each area of the image to determine whether there is uneven illumination; construct a regional illumination compensation model based on gray-scale equalization and local contrast enhancement; adaptively select the compensation window size and enhancement parameters according to the analysis results to achieve regional-level illumination balance; output the image after illumination compensation while preserving the original detailed structure of the image. On the basis of illumination compensation, adopt a boundary detail enhancement mechanism to enhance cracks and edge details in multi-view images, and at the same time fuse multiple data enhancement methods; apply a guided filter or gradient enhancement algorithm to the compensated image to highlight thin cracks and edge contours in the image; use techniques such as Laplace edge enhancement and structure tensor analysis to improve the boundary clarity; for the problem of insufficient data sets, introduce data enhancement strategies such as random rotation, scaling, mirroring, and noise perturbation; control the enhancement intensity and quantity ratio to keep the original semantics of the image unchanged; incorporate the enhanced images into a unified processing flow to construct a rich image sample set. Uniformly normalize the preprocessed and enhanced multi-view images to construct a standardized defect recognition image data set.

[0022] S2: Extract edge-guided features from the defect recognition image data set, construct an image deep feature model based on the boundary context fusion network, and introduce an asymmetric feature aggregation structure to improve the model's perception ability of tiny edge defects.

[0023] The process of constructing an image deep feature model based on the boundary context fusion network in S2 is as follows: Input the defect recognition image data set into the boundary context fusion grid; input the previously constructed standardized defect recognition image data set into the neural network training framework batch by batch; the input image format is unified to a fixed resolution and unified channels; set a multi-scale input layer at the front end of the network to perform scale normalization and preliminary feature extraction on the image; establish a data iterator to load the image and its corresponding crack edge annotation mask map batch by batch during the network training stage. The boundary context fusion network utilizes multi-scale feature channels to perform layer-by-layer feature fusion on the crack edge and surrounding structures, generating a fusion feature map containing the context relationship of defects; the main structure of the network includes multiple convolutional encoder and decoder channels, and a pyramid-style multi-scale structure is adopted to obtain features at different semantic levels; the edge detail features extracted at each scale are passed into the high-level decoding network through lateral connection and upsampling methods; the network fuses the local edge features and global semantic information layer by layer; multi-scale information is integrated through skip connections, attention mechanisms or dynamic weighting mechanisms to generate a fusion feature map with context awareness ability; An asymmetric feature aggregation structure is introduced to perform differential feature learning on the crack targets and background regions in the fusion feature map; an asymmetric convolution module is added in the feature aggregation stage to enhance the boundary response difference between cracks and the background; a region contrast loss or a feature difference enhancement module is used to perform contrast learning on the crack region and adjacent background regions in the fusion feature map; effectively suppress background interference features and highlight the local change features and edge responses of cracks; In the training stage of the boundary context fusion network, based on the manually annotated crack edge regions in the defect images; the crack edge regions with manually annotated masks are called from the dataset as training labels; a joint loss function is constructed, including edge supervision loss and feature consistency loss; the network training uses the backpropagation algorithm to optimize the model parameters and update the feature extraction and fusion weights; combined with the validation set to evaluate the accuracy and generalization ability of the model's response to crack regions during the training process; After training is completed, an image deep feature model with weak defect perception ability is output.

[0024] S3: Based on the edge response enhancement mechanism, determine whether the change in the response intensity of the image deep feature model in the crack region reaches a preset threshold. If so, label that there are potential defects in the edge region, and combine the contrast adjustment mechanism between cracks and the background to dynamically correct the preliminary annotation results.

[0025] The process of determining whether the change in the response intensity of the image deep feature model in the crack region reaches a preset threshold in the above S3 is as follows: Input the standardized defect recognition image into the trained image deep feature model to obtain a deep feature map containing edge structure information; Based on the deep feature map, perform edge response analysis, and combine the edge gradient change and boundary context features to extract the crack regions where cracks exist; Within the crack region, calculate the local response intensity of the crack region; Set a preset threshold for weak crack response, and compare the local response intensity of the crack region with the threshold; If the feature response intensity of the crack region is greater than or equal to the threshold, it is determined that the crack region responds; If the response intensity of the crack area feature is greater than or equal to the threshold, it is determined that the crack area does not respond.

[0026] The process of dynamically correcting the preliminary annotation result by combining the contrast adjustment mechanism of the crack and the background in S3 is as follows: Identify the crack area with the response intensity meeting the requirements in the output of the image deep feature model, and generate a local image patch centered on the crack area; Within the local image patch, calculate the local contrast value between the crack area and the adjacent background area; According to the change trend of the local contrast with the boundary position, evaluate the boundary clarity of the preliminary annotation area; compare the local contrast value of each image patch with the preset crack significance contrast threshold, identify the low-contrast pair area, count the brightness and texture continuity near the annotation boundary, and judge whether there are fuzzy or error areas at the boundary; if the contrast change trend is gentle, it is considered that the current boundary clarity is insufficient and there is a risk of mislabeling; output the boundary blur risk score of each area as the basis for whether to enter the dynamic correction process; If the contrast value is lower than the set threshold, indicating that the boundary is blurred, then based on the pixel gradient direction information and neighborhood gray-scale statistical features in the local image patch, dynamically judge the boundary trend; construct a neighborhood sliding window centered on the current annotation boundary, extract the pixel information inside and outside the boundary; judge the expansion trend of the crack according to the gradient direction; if the gradient direction is consistent with the extension direction of the original annotation boundary and the response value of the external neighborhood increases significantly, then perform an edge expansion operation; if the gradient outside the boundary disappears or the proportion of background features increases, then implement boundary contraction to eliminate the mislabeled area; Based on the boundary trend, perform edge expansion on the preliminary annotation boundary to correct the defect contour; In the dynamic correction process, introduce an image gradient consistency discrimination algorithm to evaluate the consistency between the corrected boundary and the main crack direction; perform gradient direction vector field analysis on the corrected boundary area, and calculate the angle between the main crack direction and the boundary extension direction; use the gradient consistency factor as an evaluation index, and the formula is: In the formula, is the main crack gradient direction, is the corrected boundary gradient direction; If the consistency factor is lower than the set threshold, then cancel the current correction operation and keep the original annotation boundary; for all areas that have completed correction and passed the consistency verification, output the final high-precision crack annotation mask image.

[0027] It can be understood that the role of judging whether the change in the response intensity of the image deep feature model in the crack area reaches the preset threshold is: Function 1: By quantitatively analyzing the edge response intensity output by the deep image feature model, real cracks can be effectively distinguished from pseudo-cracks caused by background noise, texture interference, etc. Setting a threshold as the judgment criterion helps to eliminate low-response areas, reduce misjudgment, and improve the accuracy and stability of defect recognition. Function 2: The response intensity in the crack area can reflect its structural integrity and boundary saliency, providing judgment conditions for subsequent processing such as dynamic boundary correction and crack contour optimization. The differences in crack characteristics corresponding to different response intensities can be used as an important reference basis for calculating the severity of cracks, supporting the classification of defect levels and visual identification.

[0028] The technical solution of this embodiment is as follows: The potential defect location in the crack area is judged by the edge response intensity output by the deep image feature model. When the response intensity reaches the set threshold, this area is marked as a suspected crack edge area. Subsequently, a contrast adjustment mechanism between the crack and the background is introduced to dynamically correct the preliminary marking result. According to the local contrast change trend and pixel gradient information, the position and contour of the crack boundary are optimized, enhancing the accuracy of defect marking and the clarity of the boundary. Through the dual judgment and dynamic correction of edge response and local contrast, this embodiment can effectively enhance the model's recognition ability for weak or blurred crack areas, significantly improve the boundary accuracy and stability of the defect recognition result, reduce the false detection rate caused by illumination and texture interference, and provide a more reliable regional basis for subsequent extraction of crack morphology parameters and severity assessment, overall improving the practicality and robustness of image recognition in the defect detection scenario of precast retaining walls.

[0029] Example 2: As Figure 1 shown, a method for identifying defects in precast retaining walls based on image recognition further includes the following steps: S4: According to the corrected edge defect area, apply the multi-scale morphological structure analysis method to perform contrast perception optimization processing on the correction result, extract the geometric parameters of the crack area, and dynamically generate the defect identification result.

[0030] The process of using the multi-scale morphological structure analysis method to perform contrast perception optimization on the corrected crack area in S4 is as follows: Based on the dynamically corrected crack area, multiple size structural elements are selected, and morphological dilation and erosion operations are respectively performed on the crack area at each scale to extract the crack contour. The image of the dynamically corrected crack marked area is cropped to extract the image block of the region of interest. Structural elements of multiple scale sizes are set, and morphological operations are respectively performed on the image: Dilation: Enhance the crack boundary contour and connect the broken areas; Erosion: Remove noise pixels and compress non-structural interferences; A local contrast perception mechanism is introduced to evaluate the local contrast of the crack area after dilation and corrosion at each scale, calculate the brightness difference distribution between the crack outline and the background area, and evaluate the degree of improvement of the local contrast. The crack contours obtained at each scale are weighted fused to remove low-contrast redundant boundary information, and a unified crack contour map is obtained after fusion optimization; multiple structural maps retained in the optimal scale set are unified and integrated using pixel-level or contour-level fusion strategies: Pixel-level fusion: weighted sum or maximum value superposition of all structure images; Contour-level fusion: aggregate connected regions of edge contours extracted at multiple scales, retaining parts with consistent directions and continuous structures; An edge consistency filtering algorithm is introduced to remove redundant edges and repeated responses caused by multi-scale synthesis; a crack contour map with high contrast, clear boundaries and complete structure is output as the final optimized annotation result; Based on the unified crack contour map, the geometric parameters of the crack area, including length, width, direction and distribution density, are extracted to provide parameter basis for defect identification results; the connected domain analysis is performed on the optimized crack contour map to extract the boundary point set of each crack area; The following geometric characteristic parameters are calculated respectively: Length: The longest path is calculated using the principal axis fitting or skeleton extraction algorithm; Width: The maximum and minimum spacing is measured in the direction perpendicular to the main axis of the crack; Direction: Obtain the crack strike angle through principal component analysis or least squares fitting; Distribution density: the number or total length of cracks per unit area, reflecting the overall defect distribution; The above geometric feature parameters are encoded into structured data for defect severity assessment, image statistical analysis or subsequent report generation; the final output includes complete defect identification results in the form of optimized crack contours and geometric feature tables.

[0031] The process of dynamically generating the defect identification result in S4 is as follows: According to the optimized and extracted crack geometric parameters, defect recognition rules and boundary constraints are set to determine the defect type and feature boundary; the geometric parameter data output in the crack profile optimization stage is read, including: length, width, direction, density, etc.; according to the predefined defect recognition rule library, whether the crack meets the standard of a specific defect type is determined; boundary constraints are introduced to identify and eliminate pseudo-defect responses at the edge or overlapping area of the image; a defect type label is assigned to each suspected crack area based on the comprehensive parameter characteristics and spatial distribution, and its boundary coordinates are calibrated; Map the two-dimensional coordinates of the crack area boundary to the image coordinate system during the original image acquisition process, establish the position index relationship of the defect area on the original image, and generate defect recognition labels containing defect position information, type, and geometric description; for each defect area, extract its absolute pixel coordinates in the image; structure and store attribute information such as defect type, geometric parameters, direction angle, and confidence level to form a multi-field identification tuple; according to the image size and acquisition meta-information, support mapping the defect labels to real-space coordinates; summarize all defect labels to form an image-level defect annotation set According to the attribute characteristics in the defect recognition label, combined with the label type, structural complexity, and crack density of the defect, calculate the defect severity evaluation index and assign a preliminary severity level; define the defect severity evaluation index, considering the following factors: Defect type (structural > surface type); Geometric size (the longer and wider, the higher the risk); Crack density (total crack length / number per unit area); Direction consistency and extension trend (e.g., cracks perpendicular to the vertical load direction are more dangerous); Establish a defect level discrimination model, such as using interval rules or machine learning classifiers, to classify the defects into levels: Level1: Slight (no intervention required) Level2: Medium (re-inspection recommended) Level3: Severe (immediate disposal required) Map the defect recognition label to generate a defect identification result layer, and cover the defect identification result layer on the original image through an image overlay mechanism to generate a defect identification result; create a layer canvas with the same size as the original image, and draw all defect boundary information in sequence; mark information such as type, level, number, and length beside each defect area, and use color coding or icons to enhance recognition; for example, red for severe, orange for medium, and green for slight; perform transparency overlay on the layer and the original image to generate an output image containing in-situ defect annotations.

[0032] S5: Based on the defect identification result, combined with the component positioning mechanism and the component historical operation and maintenance data, conduct a severity grading assessment of the defects and generate a defect analysis report.

[0033] The process of generating the defect analysis report in the above S5 is as follows: Map the defect recognition results to the spatial components of the precast retaining wall through the component positioning mechanism to obtain the numbers and spatial position coordinates of the corresponding structural components; map the image coordinate system to the spatial coordinate system of the retaining wall based on the position information bound during image acquisition; read the spatial position parameters in the defect recognition results and match them to the BIM model or structural component database of the retaining wall; determine whether the defect boundary falls within the geometric range of a certain structural component to determine the defect attribution component number; extract the spatial positioning information of the component and bind it together with the defect annotation for storage. Combine with the historical operation and maintenance database to retrieve whether there are similar defects and development trends in the same component during past detection cycles; use the component number as the main index to enter the historical operation and maintenance database or the defect monitoring file system to retrieve the previous detection records of the component; judge whether there are defects in the same or adjacent areas in the historical records through spatial position comparison and crack morphology matching; compare indicators such as the deviation of the crack center position, direction consistency, and image similarity; if there are historical records, extract the length, width, grade, etc. parameters of the defect in multiple detection cycles to form time series evolution data; mark the evolution trend of the defect and save it as an intermediate result of structured analysis. Based on the change trend of defect geometric parameters and historical evolution records, comprehensively determine the severity of the current defect. Generate a defect analysis report according to the defect grade, component function importance, and operation and maintenance cycle requirements; summarize the basic information of the defect, crack parameters, historical evolution trends, etc.; extract priority information from the operation and maintenance database corresponding to the component function importance; combine the structural health assessment specifications and actual operation and maintenance strategies to propose recommended measures; automatically generate an analysis report with pictures and texts.

[0034] Example 3: Please refer to Figure 2 As shown, a precast retaining wall defect recognition system based on image recognition includes: Image acquisition module: Obtain multi-angle image data of the precast retaining wall and perform light compensation and boundary enhancement processing. Image modeling module: Build an image deep feature model based on the boundary context fusion network to realize defect area recognition. Defect detection module: Through the edge response enhancement and contrast analysis mechanism, realize the initial annotation and dynamic correction of defects. Structural analysis module: Combine multi-scale morphological analysis methods to extract defect geometric features and generate identification results. Analysis and output module: Integrate historical data and structural information to grade and evaluate defects and output a defect analysis report.

[0035] It should be noted that in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, article or device.

[0036] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An assembly retaining wall defect recognition method based on image recognition, characterized in that , including the following steps: Obtain multi - perspective image data of the prefabricated retaining wall, and combine the boundary detail enhancement mechanism and the regional light compensation strategy to construct a defect recognition image dataset; Extract edge - guided features from the defect recognition image dataset, construct an image deep - feature model based on the boundary context fusion network, and introduce an asymmetric feature aggregation structure to improve the perception ability of the image deep - feature model for tiny edge defects; Through the edge response enhancement mechanism, judge whether the change in the response intensity of the image deep - feature model in the crack area reaches a preset threshold. If so, mark that there are potential defects in the crack area, and combine the contrast adjustment mechanism between the crack and the background to dynamically correct the marking result of the crack area; Use the multi - scale morphological structure analysis method to optimize the contrast perception of the corrected crack area, extract the geometric parameters of the crack area, and dynamically generate the defect identification result; Based on the defect identification result, combine the component positioning mechanism and the component historical operation and maintenance data to evaluate the severity level of the defect and generate a defect analysis report.

2. The method for identifying defects of an assembled retaining wall based on image recognition according to claim 1, wherein The process of constructing the defect recognition image dataset is as follows: Collect multi - perspective image samples under different times, perspectives and lighting conditions; Conduct brightness distribution analysis on the multi - perspective image samples. Aiming at the problem of uneven lighting, dynamically evaluate the local brightness characteristics of the multi - perspective images and adaptively determine the regional light compensation window; On the basis of light compensation, use the boundary detail enhancement mechanism to enhance the cracks and edge details in the multi - perspective images, and at the same time fuse a variety of data enhancement methods; Uniformly normalize the pre - processed and enhanced multi - perspective images to construct a standardized defect recognition image dataset.

3. The method for identifying defects of a prefabricated retaining wall based on image recognition according to claim 2, wherein The process of constructing an image deep - feature model based on the boundary context fusion network is as follows: Input the defect recognition image dataset into the boundary context fusion grid; The boundary context fusion network uses multi - scale feature channels to perform layer - by - layer feature fusion on the crack edges and surrounding structures to generate a fusion feature map containing defect context relationships; Introduce an asymmetric feature aggregation structure to perform differential feature learning on the crack targets and background regions in the fusion feature map; In the training stage of the boundary context fusion network, based on the manually marked crack edge regions in the defect images; After training, output an image deep - feature model with weak defect perception ability.

4. The method for identifying defects of prefabricated retaining walls based on image recognition according to claim 3, characterized in that, The process of judging whether the change in the response intensity of the image deep - feature model in the crack area reaches a preset threshold is as follows: Input the standardized defect recognition image into the trained image deep - feature model to obtain a deep - feature map containing edge structure information; Conduct edge response analysis based on the deep - feature map, and combine the edge gradient change and the boundary context features to extract the crack area where there are cracks; Within the crack area, calculate the local response intensity of the crack area; Set a preset threshold for weak crack response, and compare the local response intensity of the crack area with the threshold; If the local response intensity of the crack area meets the requirements, it is determined that the crack area responds; If the local response intensity of the crack area does not meet the requirements, it is determined that the crack area does not respond.

5. A method for identifying defects of prefabricated retaining walls based on image recognition according to claim 4, characterized in that, The process of dynamically correcting the preliminary marking result by combining the contrast adjustment mechanism between the crack and the background is as follows: Identify the crack area with the response intensity meeting the requirements in the output of the image deep feature model, and generate a local image patch centered on the crack area; Calculate the local contrast value between the crack area and the adjacent background area within the local image patch; Evaluate the boundary clarity of the preliminary annotation area according to the change trend of the local contrast with the boundary position; If the contrast value is lower than the set threshold, indicating that the boundary is blurred, then dynamically judge the boundary trend based on the pixel gradient direction information and neighborhood gray-scale statistical features in the local image patch; Based on the boundary trend, perform edge expansion on the preliminary annotation boundary to correct the defect contour; Introduce an image gradient consistency discrimination algorithm during the dynamic correction process to evaluate the consistency between the corrected boundary and the main crack direction; 6. The method for identifying defects of an assembled retaining wall based on image recognition according to claim 5, wherein, The process of performing contrast perception optimization on the corrected crack area using the multi-scale morphological structure analysis method is as follows: Based on the dynamically corrected crack area, select multiple size structural elements, and perform morphological dilation and erosion operations on the crack area at each scale to extract the crack contour; Introduce a local contrast perception mechanism. At each scale, evaluate the local contrast of the crack area after dilation and erosion processing, calculate the brightness difference distribution between the crack contour and the background area, and evaluate the improvement degree of the local contrast; Perform weighted fusion on the crack contours obtained at each scale, eliminate the redundant boundary information with low contrast, and obtain a unified crack contour map after fusion optimization; Extract the geometric parameters of the crack area based on the unified crack contour map; 7. A method for identifying defects of prefabricated retaining walls based on image recognition according to claim 6, characterized in that, The process of dynamically generating the defect identification result is as follows: According to the geometric parameters of the optimized extracted crack area, set the defect identification rules and boundary constraint conditions, and determine the defect type and characteristic boundary; Map the two-dimensional coordinates of the crack area boundary to the image coordinate system during the original image acquisition process, establish the position index relationship of the defect area on the original image, and generate a defect identification label containing the defect position information, type, and geometric description; According to the attribute characteristics in the defect identification label, combined with the label type, structural complexity, and crack density of the defect, calculate the defect severity evaluation index and assign a preliminary severity level; Map the defect identification label to generate a defect identification result layer, and cover the defect identification result layer on the original image through the image overlay mechanism to generate the defect identification result; 8. A method for identifying defects of prefabricated retaining walls based on image recognition according to claim 7, characterized in that, The process of generating the defect analysis report is as follows: Map the defect identification result to the spatial components of the precast retaining wall through the component positioning mechanism to obtain the numbers and spatial position coordinates of the corresponding structural components; Combined with the historical operation and maintenance database, retrieve whether there are defects and development trends in the same components during past detection cycles; Based on the change trend of the defect geometric parameters and the historical evolution record, comprehensively determine the current defect severity; Generate a defect analysis report according to the defect level, component function importance, and operation and maintenance cycle requirements; 9. An assembled retaining wall defect recognition system based on image recognition, which is applied to the method described in any one of claims 1-8, and is characterized in that Including: Image acquisition module: Obtain multi-angle image data of the precast retaining wall, and perform light compensation and boundary enhancement processing; Image modeling module: Build an image deep feature model based on the boundary context fusion network to realize defect area identification; Defect Detection Module: Through the edge response enhancement and contrast analysis mechanism, realize the preliminary annotation and dynamic correction of defects; Structure Analysis Module: Combine the multi-scale morphological analysis method to extract the geometric features of defects and generate identification results; Analysis Output Module: Integrate historical data and structural information to conduct hierarchical evaluation of defects and output defect analysis reports.

Images