Precast beam concrete pouring quality real-time monitoring method and system based on AI vision

Abstract

The application discloses an AI vision-based precast beam concrete pouring quality real-time monitoring method and system, relates to the technical field of concrete pouring monitoring, can realize accurate separation of a steel bar region image set and a concrete region image set by performing semantic segmentation on a pouring intermittent period image set, can accurately identify the position of the steel bar and the distribution of the concrete, and thus can ensure the accuracy and reliability of a quality monitoring result; by combining a railway precast beam steel bar three-dimensional model, spatial alignment of the steel bar region image and the railway precast beam steel bar three-dimensional model can accurately identify the submerged position of the steel bar and the height distribution of the concrete, and thus can better reflect the actual pouring condition of the concrete and ensure that the construction quality meets the design requirements.

Description

Technical Field

[0001] This invention relates to the field of concrete pouring monitoring technology, specifically to a method and system for real-time monitoring of the concrete pouring quality of precast beams based on AI vision. Background Technology

[0002] As a key component of railway structures, the quality of concrete pouring for precast beams directly affects the safety and durability of the overall project. With the development of railway engineering construction technology, the quality requirements for railway precast beams are constantly increasing, and the monitoring of precast beam concrete pouring quality has gradually become a focus of industry attention.

[0003] Traditional techniques for monitoring the quality of precast beam concrete pouring mainly rely on manual inspection and sampling. Manual inspection requires personnel to examine the concrete surface piece by piece to determine for defects such as cracks and honeycombing, resulting in low efficiency and a high risk of missed or incorrect inspections. Sampling involves selecting specific areas for physical or chemical analysis to infer the overall quality. However, due to the limited sample size, this method cannot comprehensively reflect the quality of the entire precast beam. Furthermore, it is difficult to monitor key real-time parameters during concrete pouring, hindering dynamic adjustments to the pouring process and failing to meet the high precision and efficiency requirements of modern railway engineering for quality control.

[0004] With the development of artificial intelligence technology, visual perception and intelligent analysis have begun to be introduced into the field of industrial quality monitoring, driving the innovation of construction quality monitoring methods. However, current monitoring technologies lack the ability to dynamically perceive the concrete pouring process, cannot simultaneously capture the distribution of reinforcing bars and surface quality characteristics, and lack the ability to quantitatively analyze quality defects, resulting in a lag in anomaly location and repair decisions. Summary of the Invention

[0005] To address the shortcomings of current monitoring technologies in dynamically sensing the concrete pouring process, their inability to simultaneously capture the distribution of reinforcing bars and surface quality characteristics, and their lack of quantitative analysis capabilities for quality defects, which leads to delays in anomaly location and repair decisions, this invention aims to provide a real-time monitoring method and system for the concrete pouring quality of precast beams based on AI vision. This system can eliminate monitoring interference caused by reinforcing bars obstructing the view and integrate 3D model modal data to achieve comprehensive and dynamic real-time monitoring of the concrete pouring quality of precast railway beams based on AI vision.

[0006] The above-mentioned technical objective of the present invention is achieved through the following technical solution:

[0007] This solution provides a method for real-time monitoring of the concrete pouring quality of precast beams based on AI vision. The method includes:

[0008] Acquire image sets during the intermittent periods of concrete pouring for precast railway beams;

[0009] Semantic segmentation is performed on the image set of the pouring interval period to obtain the image set of the reinforcing steel area and the image set of the concrete area;

[0010] By combining the three-dimensional model of the precast railway beam reinforcement and the image set of the reinforcement area, the concrete height distribution data at each reinforcement submerged in concrete was obtained.

[0011] The preprocessed image set of the concrete area is input into the anomaly recognition model to identify the anomaly analysis information of the concrete surface. The anomaly analysis information of the concrete surface is used to characterize the type and size of the anomaly on the concrete surface.

[0012] By combining the concrete height distribution data and the concrete surface anomaly analysis information, comprehensive quality monitoring parameters are calculated, and concrete quality monitoring results are generated based on the comprehensive quality monitoring parameters.

[0013] A further optimized approach involves acquiring image sets of the intermittent periods during the pouring of concrete for precast railway beams; including the following methods:

[0014] The original image data of the railway precast beam concrete pouring site is acquired by a visual matrix sensing device. The original image data is then processed by image preprocessing technology to remove noise and equalize brightness, resulting in a standardized image.

[0015] Based on the pixel grayscale distribution characteristics in the standardized image, the boundary contour between the concrete area and the pouring boundary area in the standardized image is identified, and a pouring boundary image is generated.

[0016] If the difference between the irrigation boundary images within the preset intermittent judgment period is less than the preset boundary contour change threshold, the standardized image is set as the irrigation intermittent period image in the irrigation intermittent period image set, and the irrigation intermittent period image set is constructed based on the irrigation intermittent period image.

[0017] A further optimized solution involves semantic segmentation of the image set from the pouring intervals to obtain image sets of the reinforcing steel area and concrete area; including the following methods:

[0018] The images of each pouring interval period in the set of pouring interval period images are input into the semantic segmentation model to obtain the semantic segmentation result information of the steel reinforcement and the semantic segmentation result information of the concrete in each of the pouring interval period images;

[0019] Based on the images of the intermittent pouring period and the semantic segmentation results of the reinforcing bars, images of each reinforcing bar region in the image set of the reinforcing bar region are generated, and the image set of the reinforcing bar region is constructed based on the images of the reinforcing bar region.

[0020] Based on the images of the pouring intervals and the semantic segmentation results of the concrete, images of each concrete region in the concrete region image set are generated, and the concrete region image set is constructed based on the concrete region images.

[0021] A further optimization scheme is that the semantic segmentation model is improved by adding a high-frequency enhancement module after the convolutional module with the lowest downsampling factor in the traditional U-shaped neural network model;

[0022] The high-frequency enhancement module includes an attention-enhanced high-frequency branch and a basic high-frequency extraction branch, the expressions of which are:

[0023] ;

[0024] ;

[0025] ;

[0026] In the formula, and These are the attention output feature maps of the attention-enhanced high-frequency branch and the basic output feature maps of the basic high-frequency extraction branch, respectively. To extract feature maps at high frequencies, The input feature maps for the attention-enhanced high-frequency branch and the basic high-frequency extraction branch are... , , and For activation layer, The kernel size is Convolutional layers, For splicing layers, and These are the global average pooling layer and the global max pooling layer, respectively. and These are the inverse discrete cosine transform and the discrete cosine transform, respectively. It is a high-pass filter; For function composition; This is an element-wise multiplication operation.

[0027] A further optimized solution is that the method for generating the image of the reinforcing steel area includes:

[0028] Based on the images from the pouring intervals and the semantic segmentation results of the reinforcing bars, an initial semantic segmentation image of the reinforcing bars is generated;

[0029] The initial semantic segmentation image of the reinforcing bars is contrast-enhanced to generate the image of the reinforcing bar region.

[0030] A further optimized solution is that the method for generating the concrete area image includes:

[0031] An initial concrete semantic segmentation image is generated based on the images from the pouring interval period and the concrete semantic segmentation results.

[0032] Histogram equalization is performed on the initial concrete semantic segmentation image to generate the concrete region image.

[0033] A further optimization scheme involves combining a 3D model of the precast railway beam reinforcement with an image set of the reinforcement area to obtain the concrete height distribution data at each reinforcement submerged in concrete; including the following methods:

[0034] Load the preset three-dimensional model of the railway precast beam reinforcement, which includes the spatial position parameters of each reinforcement bar;

[0035] Spatially align the images of each reinforcing bar region in the image set with the three-dimensional model of the reinforcing bar of the precast railway beam, and register the positional relationship of each reinforcing bar in the images of the reinforcing bar regions and the three-dimensional model of the reinforcing bar of the precast railway beam;

[0036] Based on the image recognition of each of the steel reinforcement regions, the spatial position parameters of the submerged positions of each steel reinforcement in the three-dimensional model of the railway precast beam steel reinforcement are identified, and the concrete height distribution data is set based on the spatial position parameters of the submerged positions of each steel reinforcement.

[0037] A further optimized scheme is as follows: the comprehensive quality monitoring parameters include the comprehensive quality monitoring parameters of each sampling point corresponding to each of the steel bars submerged in concrete during each pouring interval; including the method:

[0038] Combining the concrete height distribution data and the concrete surface anomaly analysis information, comprehensive quality monitoring parameters are calculated, and concrete quality monitoring results are generated based on these parameters; including the following methods:

[0039] By combining the concrete height distribution data and the concrete surface anomaly analysis information, comprehensive quality monitoring parameters are calculated, and concrete quality monitoring results are generated based on the comprehensive quality monitoring parameters.

[0040] Based on the concrete surface anomaly analysis information of the image set of the pouring interval period in the sampling grid constructed by connecting the sampling points in each of the pouring interval periods, a grid surface anomaly score of each sampling grid is generated, and the concrete surface anomaly score of the sampling point is obtained by summing the grid surface anomaly scores of the sampling grids in the neighborhood of the sampling point.

[0041] The comprehensive quality monitoring parameters of the sampling point are calculated by weighting the concrete height distribution score and the concrete surface anomaly score.

[0042] Based on the comprehensive quality monitoring parameters of the sampling points in the key functional areas and the weights of the key functional areas, the concrete quality monitoring result information is generated.

[0043] A further optimization scheme is proposed, in which the expressions for the concrete height distribution score and the concrete surface anomaly score are:

[0044] ;

[0045] ;

[0046] In the formula, and The first The sequence number of the intermittent pouring period mentioned in the paragraph is The concrete height distribution score and the concrete surface anomaly score at the sampling points. The plane conformity coefficient. For the first The sequence number of the intermittent pouring period mentioned in the paragraph is The total number of the sampling points, For the first The sequence number of the intermittent pouring period mentioned in the paragraph is The spatial location parameters of the sampling points, For the first The fitted plane obtained from the sampling points during the pouring interval period described in the section. For the first The fitted plane obtained from the sampling points during the intermittent period of the pouring process is located at the sequence number. The spatial position fitting parameters at the sampling points, For the first The sequence number of the intermittent pouring period mentioned in the paragraph is The total number of sampling grids in the neighborhood of the sampling point. For the first The sequence number of the intermittent pouring period mentioned in the paragraph is The first in the neighborhood of the sampling point The total number of anomaly types in the sampling grid. For the first The weighting coefficients for the aforementioned anomaly types, For the first The sequence number of the intermittent pouring period mentioned in the paragraph is The first in the neighborhood of the sampling point The sampling grid in the Grid The size of the anomaly of the aforementioned anomaly type For the first The maximum permissible size for each of the aforementioned anomaly types.

[0047] A further optimized approach involves preprocessing the concrete area image set and then inputting it into an anomaly recognition model to obtain concrete surface anomaly analysis information; this includes the following methods:

[0048] Construct a set of discrete sub-images of concrete regions for each concrete region image based on the discrete sub-images of concrete regions segmented by the reinforcing bars in the concrete region image set.

[0049] The discrete sub-images of the concrete region in the discrete sub-image set of the concrete region are subjected to input preprocessing for the anomaly recognition model to obtain standard discrete sub-images of the concrete region.

[0050] The standard discrete sub-images of the concrete areas are input into the anomaly recognition model to obtain concrete surface anomaly category information and concrete surface anomaly size information for each concrete area image; wherein, the concrete surface anomaly category information includes bubble anomaly category information, crack anomaly category information, and texture anomaly category information, and the concrete surface anomaly size information includes bubble anomaly size information corresponding to the bubble anomaly category information, crack anomaly size information corresponding to the crack anomaly category information, and texture anomaly size information corresponding to the texture anomaly category information;

[0051] The concrete surface anomaly category information and the concrete surface anomaly size information of each concrete area image are summarized to generate the concrete surface anomaly analysis information.

[0052] This solution provides an AI vision-based real-time monitoring system for the concrete pouring quality of precast beams, used to implement the aforementioned AI vision-based real-time monitoring method for the concrete pouring quality of precast beams. The system includes:

[0053] The visual image acquisition module is used to acquire image sets during the intermittent periods of concrete pouring for precast railway beams based on a visual matrix sensing device.

[0054] The semantic segmentation processing module is used to perform semantic segmentation on the image set of the pouring interval period to obtain the image set of the reinforcing steel area and the image set of the concrete area.

[0055] The height anomaly identification module is used to load the three-dimensional model of the steel reinforcement of the precast railway beam, and combine it with the image set of the steel reinforcement area to obtain the concrete height distribution data of each steel reinforcement submerged in concrete.

[0056] The surface anomaly identification module is used to input the preprocessed image set of the concrete area into the anomaly identification model to obtain concrete surface anomaly analysis information, which is used to characterize the type and size of anomalies on the concrete surface.

[0057] The comprehensive quality assessment module is used to combine the concrete height distribution data and the concrete surface anomaly analysis information to calculate comprehensive quality monitoring parameters, and generate concrete quality monitoring result information based on the comprehensive quality monitoring parameters.

[0058] Compared with the prior art, the present invention has the following beneficial effects:

[0059] This invention provides a method and system for real-time monitoring of the concrete pouring quality of precast beams based on AI vision. By performing semantic segmentation on the image set during the pouring intervals, it can accurately separate the image set of the reinforcing bar area and the image set of the concrete area, and accurately identify the location of the reinforcing bar and the distribution of the concrete, thereby ensuring the accuracy and reliability of the quality monitoring results. By combining the three-dimensional model of the reinforcing bar of the railway precast beam and spatially aligning the image of the reinforcing bar area with the three-dimensional model of the railway precast beam reinforcing bar, it can accurately identify the submerged position of the reinforcing bar and the height distribution of the concrete, thereby better reflecting the actual concrete pouring situation and ensuring that the construction quality meets the design requirements. Attached Figure Description

[0060] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0061] Figure 1 This is a flowchart illustrating a method for real-time monitoring of precast beam concrete pouring quality based on AI vision, provided in one embodiment of the present invention.

[0062] Figure 2 A schematic diagram of the structure of a real-time monitoring system for concrete pouring quality of precast railway beams based on AI vision, provided in an embodiment of the present invention;

[0063] Figure 3 This is a schematic diagram of the structure of a semantic segmentation processing module provided in one embodiment of the present invention;

[0064] Figure 4 This is a schematic diagram of the structure of a semantic segmentation model provided in one embodiment of the present invention;

[0065] Figure 5 This is a schematic diagram of the structure of a high-frequency enhancement module according to an embodiment of the present invention;

[0066] Figure 6 This is a schematic diagram of the structure of another semantic segmentation processing module provided in one embodiment of the present invention;

[0067] Figure 7 This is a schematic diagram of the structure of a high anomaly identification module provided in one embodiment of the present invention;

[0068] Figure 8 This is a schematic diagram of the structure of a comprehensive quality assessment module provided in one embodiment of the present invention;

[0069] Figure 9 This is a schematic diagram of the structure of a visual image acquisition module according to an embodiment of the present invention;

[0070] Figure 10 This is a schematic diagram of the structure of a surface anomaly identification module provided in one embodiment of the present invention. Detailed Implementation

[0071] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0072] In view of this, the present solution provides the following embodiments to solve the above-mentioned technical problems:

[0073] One embodiment provides a method for real-time monitoring of the concrete pouring quality of precast beams based on AI vision, such as... Figure 1 As shown, the method includes:

[0074] S101, Obtain the image set of the intermittent pouring period of precast concrete for railway beams; this step specifically includes the following methods:

[0075] The original image data of the railway precast beam concrete pouring site is acquired by a visual matrix sensing device. The original image data is then processed by image preprocessing technology to remove noise and equalize brightness, resulting in a standardized image.

[0076] Based on the pixel grayscale distribution characteristics in the standardized image, the boundary contour between the concrete area and the pouring boundary area in the standardized image is identified, and a pouring boundary image is generated.

[0077] If the difference between each irrigation boundary image within the preset intermittent judgment period is less than the preset boundary contour change threshold, the standardized image is set as the irrigation intermittent period image in the irrigation intermittent period image set, and the irrigation intermittent period image set is constructed based on the irrigation intermittent period image.

[0078] S102, perform semantic segmentation on the image set during the pouring intervals to obtain image sets of the reinforcing steel area and concrete area; this step specifically includes the following methods:

[0079] The images of each pouring interval period in the image set are input into the semantic segmentation model to obtain the semantic segmentation results of the steel reinforcement and concrete in each pouring interval period image.

[0080] Based on the images from the intermittent pouring period and the semantic segmentation results of the reinforcing bars, images of each reinforcing bar region in the image set of reinforcing bar regions are generated, and a set of reinforcing bar region images is constructed based on the images of the reinforcing bar regions.

[0081] Based on the images from the intermittent pouring period and the semantic segmentation results of the concrete, images of each concrete region in the concrete region image set are generated, and a concrete region image set is constructed based on the concrete region images.

[0082] S103, combining the 3D model of the precast railway beam reinforcement and the image set of the reinforcement area, obtains the concrete height distribution data at each reinforcement submerged in concrete; this step specifically includes the following methods:

[0083] Load the preset 3D model of the steel reinforcement of the precast railway beam. The 3D model of the steel reinforcement of the precast railway beam includes the spatial position parameters of each steel bar.

[0084] Spatially align the images of each rebar region in the rebar region image set with the 3D model of the railway precast beam rebar, and register the positional relationship of each rebar in the rebar region images and the 3D model of the railway precast beam rebar.

[0085] Based on the image recognition of each steel bar region, the spatial location parameters of the submerged position of each steel bar in the 3D model of the precast railway beam steel bars are determined, and concrete height distribution data is set based on the spatial location parameters of the submerged position of each steel bar.

[0086] S104, After preprocessing the concrete area image set, input it into the anomaly recognition model to identify concrete surface anomaly analysis information. This information is used to characterize the type and size of anomalies on the concrete surface. This step specifically includes the following methods:

[0087] Concrete region discrete sub-image set is constructed based on discrete concrete region sub-images of each concrete region image in the concrete region image set, which are segmented by the steel bars.

[0088] The discrete sub-images of the concrete region in the discrete sub-image set of the concrete region are subjected to input preprocessing for the anomaly recognition model to obtain the standard discrete sub-images of the concrete region.

[0089] The standard discrete sub-image of the concrete area is input into the anomaly recognition model to obtain the concrete surface anomaly category information and concrete surface anomaly size information of each concrete area image; wherein, the concrete surface anomaly category information includes bubble anomaly category information, crack anomaly category information and texture anomaly category information, and the concrete surface anomaly size information includes the bubble anomaly size information corresponding to the bubble anomaly category information, the crack anomaly size information corresponding to the crack anomaly category information and the texture anomaly size information corresponding to the texture anomaly category information;

[0090] The concrete surface anomaly category information and the concrete surface anomaly size information of each concrete area image are summarized to generate the concrete surface anomaly analysis information.

[0091] S105, combining the concrete height distribution data and the concrete surface anomaly analysis information, calculate the comprehensive quality monitoring parameters, and generate concrete quality monitoring result information based on the comprehensive quality monitoring parameters. The comprehensive quality monitoring parameters include the comprehensive quality monitoring parameters of each sampling point corresponding to each of the steel bars submerged in concrete during each pouring interval; this step specifically includes the following sub-steps:

[0092] By combining the concrete height distribution data and the concrete surface anomaly analysis information, comprehensive quality monitoring parameters are calculated, and concrete quality monitoring results are generated based on the comprehensive quality monitoring parameters.

[0093] Based on the concrete surface anomaly analysis information of the image set of the pouring interval period in the sampling grid constructed by connecting the sampling points in each of the pouring interval periods, a grid surface anomaly score of each sampling grid is generated, and the concrete surface anomaly score of the sampling point is obtained by summing the grid surface anomaly scores of the sampling grids in the neighborhood of the sampling point.

[0094] The comprehensive quality monitoring parameters of the sampling point are calculated by weighting the concrete height distribution score and the concrete surface anomaly score.

[0095] Based on the comprehensive quality monitoring parameters of the sampling points in the key functional areas and the weights of the key functional areas, the concrete quality monitoring result information is generated.

[0096] In one embodiment, an AI vision-based real-time monitoring system 100 for the concrete pouring quality of precast beams is provided, such as... Figure 2As shown, a method for real-time monitoring of the concrete pouring quality of precast beams based on AI vision is described. The system includes: a visual image acquisition module 101, a semantic segmentation processing module 102, a height anomaly recognition module 103, a surface anomaly recognition module 104, and a comprehensive quality assessment module 105, wherein:

[0097] Specifically, the visual image acquisition module 101 can be used to acquire image sets of the intermittent pouring periods of railway precast beam concrete based on a visual matrix sensing device. The visual image acquisition module 101 can be connected to the semantic segmentation processing module 102, and can send the acquired image sets of the intermittent pouring periods of railway precast beam concrete to the semantic segmentation processing module 102 through a communication channel.

[0098] Optionally, the visual matrix sensing device in the visual image acquisition module 101 can be composed of multiple image sensors, which can simultaneously capture image data of the concrete pouring site from different angles.

[0099] Optionally, the visual image acquisition module 101 can determine whether the railway precast beam concrete pouring site is in a pouring interval period based on the original image data of the railway precast beam concrete pouring site acquired by the visual matrix sensing device. When the visual image acquisition module 101 determines that the railway precast beam concrete pouring site is in a pouring interval period, the visual image acquisition module 101 can construct an image set of the railway precast beam concrete pouring interval period after preprocessing the original image data acquired by the visual matrix sensing device during the pouring interval period.

[0100] Specifically, the semantic segmentation processing module 102 can be connected to the visual image acquisition module 101, the height anomaly recognition module 103, and the surface anomaly recognition module 104. After receiving the image set of the pouring interval period sent by the visual image acquisition module 101, the semantic segmentation processing module 102 can perform semantic segmentation on the image set of the pouring interval period to obtain an image set of the reinforcing steel area and an image set of the concrete area. The semantic segmentation processing module 102 can send the generated concrete area image set to the height anomaly recognition module 103 through a communication channel, and the semantic segmentation processing module 102 can send the generated reinforcing steel area image set to the surface anomaly recognition module 104 through a communication channel.

[0101] Optionally, the semantic segmentation processing module 102 can be equipped with a semantic segmentation model, and can input the images of each pouring interval period in the image set of pouring interval period into the semantic segmentation model to obtain the image set of the reinforcing steel area and the image set of the concrete area corresponding to each pouring interval period image.

[0102] Specifically, the height anomaly identification module 103 can be connected to the semantic segmentation processing module 102 and the comprehensive quality assessment module 105. The height anomaly identification module 103 can receive the set of rebar area images generated by the semantic segmentation processing module 102. After receiving the rebar area image set, the height anomaly identification module 103 can load a three-dimensional model of the railway precast beam rebar and, combined with the rebar area image set, obtain the concrete height distribution data at each rebar submerged in concrete. The height anomaly identification module 103 sends the generated concrete height distribution data to the comprehensive quality assessment module 105 via a communication channel.

[0103] Optionally, the height anomaly identification module 103 can load a three-dimensional model of the steel reinforcement of the precast railway beam, spatially align the image set of the steel reinforcement area with the three-dimensional model of the precast railway beam steel reinforcement, register the positional relationship of each steel reinforcement in the image set of the steel reinforcement area and the three-dimensional model of the precast railway beam steel reinforcement, and based on the images of each steel reinforcement area in the registered image set of the steel reinforcement area, identify the spatial position parameters of the submerged position of each steel reinforcement in the three-dimensional model of the precast railway beam steel reinforcement, and obtain the concrete height distribution data at each steel reinforcement submerged in concrete.

[0104] Optionally, the visual image acquisition module 101 can acquire point cloud data of the railway precast beam reinforcement before concrete pouring, combine it with the digital model of the railway precast beam project, construct a railway precast beam reinforcement building information model, and use the railway precast beam reinforcement building information model as a three-dimensional model of the railway precast beam reinforcement.

[0105] Specifically, the surface anomaly identification module 104 can be connected to the semantic segmentation processing module 102 and the comprehensive quality assessment module 105. The surface anomaly identification module 104 can receive a set of concrete area images generated by the semantic segmentation processing module 102. After receiving the set of concrete area images generated by the semantic segmentation processing module 102, the surface anomaly identification module 104 can preprocess the concrete area images and input them into the anomaly identification model to obtain concrete surface anomaly analysis information. The surface anomaly identification module 104 can then send the generated concrete surface anomaly analysis information to the comprehensive quality assessment module 105 through a communication channel.

[0106] Optionally, the concrete surface anomaly analysis information generated by the surface anomaly identification module 104 can be used to characterize the type and size of anomalies on the concrete surface.

[0107] Optionally, the concrete surface anomaly analysis information may include concrete surface anomaly category information and concrete surface anomaly size information. Specifically, the concrete surface anomaly category information may include, but is not limited to, one or more of the following: bubble anomaly category information, crack anomaly category information, aggregate anomaly category information, mortar anomaly category information, and texture anomaly category information. The concrete surface anomaly size information may include, but is not limited to, one or more of the following: bubble anomaly size information corresponding to the bubble anomaly category information, crack anomaly size information corresponding to the crack anomaly category information, aggregate anomaly size information corresponding to the aggregate anomaly category information, mortar anomaly size information corresponding to the mortar anomaly category information, and texture anomaly size information corresponding to the texture anomaly category information.

[0108] Optionally, the anomaly recognition model mounted on the surface anomaly recognition module 104 may include, but is not limited to, AlexNet, Visual Geometry Group (VGG), YOLO, Region Convolutional Neural Network (R-CNN) and Fully Convolutional Network (FCN).

[0109] Specifically, the comprehensive quality assessment module 105 can be connected to the height anomaly identification module 103 and the surface anomaly identification module 104. The comprehensive quality assessment module 105 can receive concrete height distribution data generated by the height anomaly identification module 103 and concrete surface anomaly analysis information generated by the surface anomaly identification module 104. The comprehensive quality assessment module 105 can be used to combine the concrete height distribution data and the concrete surface anomaly analysis information to calculate comprehensive quality monitoring parameters, and generate concrete quality monitoring results information based on the comprehensive quality monitoring parameters.

[0110] The aforementioned AI vision-based real-time monitoring system for precast beam concrete pouring quality acquires image sets during pouring intervals through a visual matrix sensing device in the visual image acquisition module. Combined with semantic segmentation processing, it can accurately distinguish and identify the reinforcing steel area and the concrete area. The system can identify the height distribution of reinforcing steel submerged in concrete through a height anomaly identification module, and identify the types and dimensions of anomalies on the concrete surface through a surface anomaly identification module. This enables comprehensive monitoring of key internal parameters and external surface conditions during the precast beam concrete pouring process. By acquiring and processing images during pouring intervals, it can obtain real-time concrete height distribution data and surface anomaly analysis information. The comprehensive quality assessment module quickly generates monitoring results, allowing for timely understanding of the pouring quality status and effectively preventing the escalation of quality problems, thus ensuring the stability of the pouring process.

[0111] In an optional embodiment of the invention, such as Figure 3As shown, the semantic segmentation processing module 102 may include a semantic segmentation reasoning submodule 210, a rebar region recognition submodule 220, and a concrete region recognition submodule 230, wherein:

[0112] Specifically, the semantic segmentation inference submodule 210 can be connected to the visual image acquisition module, the rebar region recognition submodule 220, and the concrete region recognition submodule 230. The semantic segmentation inference submodule 210 can input images from each pouring interval period in the image set acquired by the visual image acquisition module 101 into the semantic segmentation model mounted on the semantic segmentation inference submodule 210, obtaining the rebar semantic segmentation result information and the concrete semantic segmentation result information for each pouring interval period image. The semantic segmentation inference submodule 210 can transmit the generated rebar semantic segmentation result information to the rebar region recognition submodule 220 via a communication channel, and the semantic segmentation inference submodule 210 can transmit the generated concrete semantic segmentation result information to the concrete region recognition submodule 230 via a communication channel.

[0113] Specifically, the rebar region identification submodule 220 can be connected to the semantic segmentation inference submodule 210 and the height anomaly identification module. The rebar region identification submodule 220 can be used to obtain images of each rebar region in the rebar region image set based on the images from the intermittent pouring period and the rebar semantic segmentation results generated by the semantic segmentation inference submodule 210, and construct a rebar region image set based on these images. The rebar region identification submodule 220 can then send the constructed rebar region image set to the height anomaly identification module via a communication channel.

[0114] Specifically, the concrete region identification submodule 230 can be connected to the semantic segmentation inference submodule 210 and the surface anomaly identification module. The concrete region identification submodule 230 can generate images of each concrete region in a concrete region image set based on images from the pouring interval period and the concrete semantic segmentation results generated by the semantic segmentation inference submodule 210, and construct a concrete region image set based on these images. The concrete region identification submodule 230 can then send the constructed concrete region image set to the surface anomaly identification module via a communication channel.

[0115] The aforementioned AI-based vision-based real-time monitoring system for the concrete pouring quality of precast railway beams utilizes a semantic segmentation model to automatically identify and distinguish different regions in images, resulting in more accurate segmentation. This allows for precise identification of the location of reinforcing bars and the distribution of concrete, significantly improving the accuracy and reliability of quality monitoring. By generating images of the reinforcing bar region based on images from pouring intervals and semantic segmentation results of the reinforcing bars, precise integration and extraction of the reinforcing bar region can be achieved. This enables the subsequent height anomaly identification module to process the high-quality reinforcing bar region images, enhancing the reliability of quality monitoring. Furthermore, by generating images of the concrete region based on images from pouring intervals and semantic segmentation results of the concrete, mutual interference between the concrete region and other structural regions of the precast railway beam, such as reinforcing bars, can be suppressed, allowing for focused analysis of the concrete surface condition and thus improving the accuracy of concrete surface anomaly detection.

[0116] In an optional embodiment of the invention, such as Figure 4 and Figure 5 As shown, the semantic segmentation model carried by the semantic segmentation inference submodule can be improved by adding a high-frequency enhancement module after the convolutional module with the lowest downsampling factor in the traditional U-shaped neural network model.

[0117] Optionally, since the semantic segmentation model performs binary region contour semantic segmentation, the maximum downsampling factor of the traditional U-shaped neural network model can be reduced, thereby simplifying the network structure of the semantic segmentation model and improving its response speed.

[0118] Furthermore, the loss function of the semantic segmentation model can include a binary cross-entropy loss that uses pixels on the segmentation boundary as samples. The expression for the binary cross-entropy loss can be:

[0119] ;

[0120] In the formula, For binary cross-entropy loss, For the sample size, For the first The true label of each sample The model predicts the first The true label of each sample The positive class probability.

[0121] Optionally, the semantic segmentation inference submodule can input images from the intermittent pouring periods into the semantic segmentation model to obtain segmentation results. The semantic segmentation model implemented in the semantic segmentation inference submodule may include a convolutional module, a high-frequency enhancement module, a downsampling layer, an upsampling layer, and a K1 convolutional layer. The K1 convolutional layer has a kernel size of [missing information]. The convolutional layer.

[0122] Specifically, the high-frequency enhancement module includes an attention-enhanced high-frequency branch and a basic high-frequency extraction branch. The expressions for the attention-enhanced high-frequency branch and the basic high-frequency extraction branch can be:

[0123] ;

[0124] ;

[0125] ;

[0126] In the formula, and These are the attention output feature maps of the attention-enhanced high-frequency branch and the basic output feature maps of the basic high-frequency extraction branch, respectively. To extract feature maps at high frequencies, The input feature maps are used for the attention-enhanced high-frequency branch and the basic high-frequency extraction branch. , , and For activation layer, The kernel size is Convolutional layers, For splicing layers, and These are the global average pooling layer and the global max pooling layer, respectively. and These are the inverse discrete cosine transform and the discrete cosine transform, respectively. It is a high-pass filter.

[0127] Optional, Discrete Cosine Transformer It can transform an image from the spatial domain to the frequency domain, making high-frequency components easier to separate; high-pass filter It can effectively retain high-frequency information and filter low-frequency noise; inverse discrete cosine transformer The processed information can be converted back to the spatial domain to ensure that high-frequency features are accurately represented in the original image dimensions. The high-frequency enhancement module can extract and enhance high-frequency information through a combination of discrete cosine transform, high-pass filtering, and inverse discrete cosine transform.

[0128] Optionally, the expression for the high-frequency enhancement module can be: .in, For high frequency enhancement modules, The kernel size is The convolutional layer.

[0129] The aforementioned AI vision-based real-time monitoring system for the concrete pouring quality of precast railway beams adds a high-frequency enhancement module after the convolution module with the lowest sampling factor in the traditional U-shaped neural network model. This enhances high-frequency information in the early stages of network feature extraction, enabling targeted enhancement and preservation of high-frequency features in the image. Consequently, the edge contours between steel bars and concrete can be more clearly identified, improving the accuracy of semantic segmentation.

[0130] In an optional embodiment of the present invention, please refer to Figure 6 The rebar region recognition submodule 220 may include a rebar semantic segmentation unit 221 and a rebar comparison enhancement unit 222, wherein:

[0131] The rebar semantic segmentation unit 221 can be used to generate an initial rebar semantic segmentation image based on the image of the intermittent pouring period and the rebar semantic segmentation result information.

[0132] The rebar contrast enhancement unit 222 can be used to enhance the contrast of the initial rebar semantic segmentation image and generate a rebar region image.

[0133] In the aforementioned AI vision-based real-time monitoring system for the concrete pouring quality of precast railway beams, the initial semantic segmentation image of the reinforcing bars is generated by fusing the original image with the segmentation results through the reinforcing bar semantic segmentation unit. The reinforcing bar contrast enhancement unit directionally enhances the edge contrast of the reinforcing bar outline, which can significantly improve the boundary differentiation ability of densely arranged reinforcing bars, overcome the feature confusion error caused by the cross-over and overlap of the reinforcing bar mesh, ensure the accuracy of the submerged position of the reinforcing bars, and provide a reliable data foundation for concrete thickness measurement.

[0134] In an optional embodiment of the present invention, please refer to Figure 6 The concrete region identification submodule 230 may include a concrete semantic segmentation unit 231 and a concrete histogram enhancement unit 232, wherein:

[0135] The concrete semantic segmentation unit 231 can be used to generate an initial concrete semantic segmentation image based on the image of the pouring interval period and the concrete semantic segmentation result information.

[0136] The concrete histogram enhancement unit 232 can be used to perform histogram equalization processing on the initial concrete semantic segmentation image to generate a concrete region image.

[0137] In the aforementioned AI vision-based real-time monitoring system for the concrete pouring quality of precast railway beams, an initial concrete semantic segmentation image is constructed using a concrete semantic segmentation unit. The concrete histogram enhancement unit optimizes the surface texture distribution based on histogram equalization technology, which can eliminate artifact interference and enhance the grayscale features of concrete surface defects, thereby improving the accuracy of the anomaly recognition model in the surface anomaly recognition module.

[0138] In an optional embodiment of the invention, such as Figure 7 As shown, the height anomaly recognition module 103 may include a model loading and analysis unit 310, an image model registration unit 320, and a height distribution recognition unit 330, wherein:

[0139] The model loading analysis unit 310 can be used to load a preset three-dimensional model of the steel reinforcement of precast railway beams.

[0140] Optionally, the three-dimensional model of the steel reinforcement in the precast railway beam can include the spatial location parameters of each steel reinforcement.

[0141] The image model registration unit 320 can be used to spatially align the images of each steel reinforcement area in the steel reinforcement area image set with the 3D model of the steel reinforcement of the precast railway beam, and register the positional relationship of each steel reinforcement in the steel reinforcement area image and the 3D model of the steel reinforcement of the precast railway beam.

[0142] The height distribution recognition unit 330 can be used to recognize the spatial location parameters of the submerged position of each steel bar in the three-dimensional model of the precast railway beam steel bars based on the image of each steel bar area, and set the concrete height distribution data based on the spatial location parameters of the submerged position of each steel bar.

[0143] The aforementioned AI vision-based real-time monitoring system for the concrete pouring quality of precast railway beams, by loading a 3D model of the precast railway beam reinforcement, provides a precise 3D reference framework for reinforcement position matching and height calculation, clearly defining the theoretical spatial coordinates of each reinforcement bar in the precast beam and improving the accuracy of reinforcement bar spatial positioning. By using an image model registration unit to spatially align the images of each reinforcement bar area in the reinforcement bar image set with the 3D model of the precast railway beam reinforcement, registering the positional relationship of each reinforcement bar, it can achieve precise fusion of 2D image information and 3D model data, thereby eliminating spatial position deviations caused by factors such as image shooting angle and distance.

[0144] In an optional embodiment of the present invention, the comprehensive quality monitoring parameters may include the comprehensive quality monitoring parameters of each sampling point corresponding to each reinforcing bar submerged in concrete during each pouring interval. The three-dimensional model of the railway precast beam may include key functional areas and key functional area weights. Figure 7 and Figure 8 As shown, the model loading analysis unit 310 can also be used to load a three-dimensional model of a precast railway beam. The comprehensive quality assessment module 105 may include a height anomaly scoring unit 510, a surface anomaly scoring unit 520, a comprehensive parameter calculation unit 530, and a monitoring result generation unit 540, wherein:

[0145] The height anomaly scoring unit 510 can be used to calculate the concrete height distribution score of each sampling point based on the concrete height distribution data of the image set of each pouring interval period.

[0146] The surface anomaly scoring unit 520 can be used to generate a grid surface anomaly score for each sampling grid based on the concrete surface anomaly analysis information of the image set of the pouring interval period in the sampling grid constructed by connecting each sampling point in each pouring interval period, and to obtain the concrete surface anomaly score of the sampling point by summing the grid surface anomaly scores of the sampling grids in the neighborhood of the sampling point.

[0147] The comprehensive parameter calculation unit 530 can be used to calculate the comprehensive quality monitoring parameters of the sampling points by weighted sum of the concrete height distribution score and the concrete surface anomaly score.

[0148] The monitoring result generation unit 540 can be used to generate concrete quality monitoring result information based on the comprehensive quality monitoring parameters of sampling points in key functional areas and the weights of key functional areas.

[0149] Optionally, the weights of key functional areas can be calculated based on the load conditions, reinforcement density, and slope distribution of each area.

[0150] The aforementioned AI vision-based real-time monitoring system for the concrete pouring quality of precast railway beams can quantitatively assess the concrete height distribution by calculating the concrete height distribution score at each sampling point, analyzing the deviation of the concrete height at each sampling point, and thus accurately measuring the concrete quality represented by the concrete height distribution. By using the surface anomaly scoring unit to generate concrete surface anomaly scores for the sampling points, it can achieve a refined and correlated assessment of concrete surface anomalies, thereby comprehensively reflecting the quality status of the concrete surface. This provides accurate surface dimension data for comprehensive parameter calculation and improves the accuracy of concrete surface quality assessment.

[0151] Furthermore, in the aforementioned AI vision-based real-time monitoring system for the concrete pouring quality of precast railway beams, the comprehensive parameter calculation unit calculates the comprehensive quality monitoring parameters of the sampling points, enabling multi-dimensional comprehensive analysis of concrete quality. This allows for a more comprehensive and accurate assessment of the overall concrete pouring quality at each sampling point. The model loading analysis unit combines the comprehensive quality monitoring parameters and the weights of key functional areas to generate concrete quality monitoring results. This allows for a balanced assessment of the overall quality of the precast beams, highlighting key aspects while also encompassing the whole, thus more accurately reflecting the overall quality level of the precast beams and ensuring the overall quality and safety of railway precast beams.

[0152] In an optional embodiment of the present invention, the visual image acquisition module can acquire point cloud data of the precast railway beams before concrete pouring, combine it with the digital model of the precast railway beam project, construct a precast railway beam architectural information model, and use the precast railway beam architectural information model as a three-dimensional model of the precast railway beams. The precast railway beam architectural information model may include a precast railway beam reinforcement architectural information model.

[0153] In an optional embodiment of the present invention, the expressions for the concrete height distribution score and the concrete surface anomaly score can be:

[0154] ;

[0155] ;

[0156] In the formula, and The first The sequence number of the intermittent pouring period is The concrete height distribution score and concrete surface anomaly score at the sampling points. The plane conformity coefficient. For the first The sequence number of the intermittent pouring period is The total number of sampling points, For the first The sequence number of the intermittent pouring period is The spatial location parameters of the sampling points, For the first The fitted plane obtained from sampling points during the intermittent pouring period. For the first The fitted plane obtained from the sampling points during the intermittent pouring period is in the sequence number. The spatial location fitting parameters at the sampling points, For the first The sequence number of the intermittent pouring period is The total number of sampling grids in the neighborhood of the sampling point. For the first The sequence number of the intermittent pouring period is The sampling point in the neighborhood of the first The total number of anomaly types in the grid sampling grid. For the first Weighting coefficients for different anomaly types For the first The sequence number of the intermittent pouring period is The sampling point in the neighborhood of the first The first in the sampling grid The size of the anomaly for each type of anomaly. For the first The maximum allowable size for each anomaly type.

[0157] Optionally, when the reinforcement uses a grid layout, the total number of sampling grids in the neighborhood of the sampling point. It can be 4.

[0158] In an optional embodiment of the invention, such as Figure 9 As shown, the visual image acquisition module 101 may include a raw image acquisition unit 110, an irrigation boundary recognition unit 120, and an intermittent image construction unit 130, wherein:

[0159] The original image acquisition unit 110 can be used to acquire original image data of the railway precast beam concrete pouring site through a visual matrix sensing device, and use image preprocessing technology to filter noise and equalize brightness of the original image data to obtain a standardized image.

[0160] The pouring boundary recognition unit 120 can be used to identify the boundary contour between the concrete area and the pouring boundary area in the standardized image based on the pixel grayscale distribution characteristics in the standardized image, and generate a pouring boundary image.

[0161] The intermittent image construction unit 130 can be used to set the standardized image as the intermittent period image in the intermittent period image set if the difference between each irrigation boundary image within the preset intermittent judgment period is less than the preset boundary contour change threshold, and to construct the intermittent period image set based on the intermittent period image.

[0162] The aforementioned AI vision-based real-time monitoring system for the concrete pouring quality of precast railway beams uses a pouring boundary recognition unit to identify the boundary contour between the concrete area and the pouring boundary area. This enables precise definition of the pouring area boundary and accurate determination of whether the pouring is in an intermittent period, ensuring the relevance and accuracy of the judgment on the pouring process status. Furthermore, the intermittent image construction unit constructs an image set for the pouring intermittent period. By monitoring changes in the boundary contour, it effectively determines whether the pouring is in a relatively stable intermittent state, accurately distinguishing images during intermittent periods. This ensures that the constructed image set for the pouring intermittent period only contains valid images that reflect the intermittent state, thereby ensuring the reliability and relevance of the concrete pouring quality analysis results.

[0163] In an optional embodiment of the invention, such as Figure 10 As shown, the surface anomaly recognition module 104 may include a discrete image construction unit 410, a discrete image processing unit 420, a surface anomaly detection unit 430, and an analysis result generation unit 440, wherein:

[0164] The discrete image construction unit 410 can be used to construct a set of discrete sub-images of concrete regions based on discrete concrete region images segmented by steel bars in a set of concrete region images.

[0165] The discrete image processing unit 420 can be used to perform input preprocessing on the discrete sub-images of the concrete region in the discrete sub-image set of the concrete region for the anomaly recognition model, so as to obtain the standard discrete sub-image of the concrete region.

[0166] The surface anomaly detection unit 430 can be used to input standard discrete sub-images of concrete areas into the anomaly recognition model to obtain concrete surface anomaly category information and concrete surface anomaly size information for each concrete area image.

[0167] Optionally, the concrete surface anomaly category information includes bubble anomaly category information, crack anomaly category information, and texture anomaly category information. The concrete surface anomaly size information includes the bubble anomaly size information corresponding to the bubble anomaly category information, the crack anomaly size information corresponding to the crack anomaly category information, and the texture anomaly size information corresponding to the texture anomaly category information. At this point, the total number of anomaly types... It can be 3.

[0168] The analysis result generation unit 440 can be used to summarize the concrete surface anomaly category information and concrete surface anomaly size information of the images of each concrete area, and generate concrete surface anomaly analysis information.

[0169] The aforementioned AI-based vision-based real-time monitoring system for the concrete pouring quality of precast railway beams utilizes a discrete image construction unit to build discrete sub-image sets for each concrete region. This allows for precise segmentation and focusing of concrete surface images, setting each discrete concrete region as an independent analysis object. This avoids interference from reinforcing steel areas in the analysis of concrete surface anomalies, ensuring the accuracy of surface anomaly analysis. A discrete image processing unit preprocesses the discrete sub-image sets for anomaly recognition models, standardizing parameters such as size to better meet the model's input requirements. This ensures stable and efficient anomaly detection, improving the reliability of the surface anomaly recognition process. A surface anomaly detection unit identifies the type and size of concrete surface anomalies, accurately classifying and quantifying them, providing a comprehensive and accurate understanding of concrete surface quality parameters. Finally, an analysis result generation unit generates concrete surface anomaly analysis information, clearly displaying the distribution of anomalies and comprehensively assessing surface quality. This ensures the scientific rigor and completeness of the quality assessment for precast railway beam concrete pouring.

[0170] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the aforementioned AI vision-based real-time monitoring method for concrete pouring quality of precast railway beams.

[0171] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0172] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for real-time monitoring of the concrete pouring quality of precast beams based on AI vision, characterized in that, The methods include: Acquire image sets during the intermittent periods of concrete pouring for precast railway beams; Semantic segmentation is performed on the image set of the pouring interval period to obtain the image set of the reinforcing steel area and the image set of the concrete area; By combining the three-dimensional model of the precast railway beam reinforcement and the image set of the reinforcement area, the concrete height distribution data at each reinforcement submerged in concrete was obtained. The preprocessed image set of the concrete area is input into the anomaly recognition model to identify the anomaly analysis information of the concrete surface. The anomaly analysis information of the concrete surface is used to characterize the type and size of the anomaly on the concrete surface. By combining the concrete height distribution data and the concrete surface anomaly analysis information, comprehensive quality monitoring parameters are calculated, and concrete quality monitoring results are generated based on the comprehensive quality monitoring parameters. Semantic segmentation is performed on the image set of the pouring interval period to obtain the image set of the reinforcing steel area and the image set of the concrete area; Including methods: The images of each pouring interval period in the set of pouring interval period images are input into the semantic segmentation model to obtain the semantic segmentation result information of the steel reinforcement and the semantic segmentation result information of the concrete in each of the pouring interval period images; Based on the images of the intermittent pouring period and the semantic segmentation results of the reinforcing bars, images of each reinforcing bar region in the image set of the reinforcing bar region are generated, and the image set of the reinforcing bar region is constructed based on the images of the reinforcing bar region. Based on the images of the pouring intervals and the semantic segmentation results of the concrete, images of each concrete region in the concrete region image set are generated, and the concrete region image set is constructed based on the concrete region images. The semantic segmentation model is improved by adding a high-frequency enhancement module after the convolutional module with the lowest downsampling factor in the traditional U-shaped neural network model; The high-frequency enhancement module includes an attention-enhanced high-frequency branch and a basic high-frequency extraction branch, the expressions of which are: ； ； ； In the formula, and These are the attention output feature maps of the attention-enhanced high-frequency branch and the basic output feature maps of the basic high-frequency extraction branch, respectively. To extract feature maps at high frequencies, The input feature maps for the attention-enhanced high-frequency branch and the basic high-frequency extraction branch are... , , and For activation layer, The kernel size is Convolutional layers, For splicing layers, and These are the global average pooling layer and the global max pooling layer, respectively. and These are the inverse discrete cosine transform and the discrete cosine transform, respectively. It is a high-pass filter; For function composition; This is an element-wise multiplication operation.

2. The method for real-time monitoring of precast beam concrete pouring quality based on AI vision according to claim 1, characterized in that, Acquire a set of images during the intermittent pouring periods of precast concrete for railway beams; including method: The original image data of the railway precast beam concrete pouring site is acquired by a visual matrix sensing device. The original image data is then processed by image preprocessing technology to remove noise and equalize brightness, resulting in a standardized image. Based on the pixel grayscale distribution characteristics in the standardized image, the boundary contour between the concrete area and the pouring boundary area in the standardized image is identified, and a pouring boundary image is generated. If the difference between the irrigation boundary images within the preset intermittent judgment period is less than the preset boundary contour change threshold, the standardized image is set as the irrigation intermittent period image in the irrigation intermittent period image set, and the irrigation intermittent period image set is constructed based on the irrigation intermittent period image.

3. The method for real-time monitoring of precast beam concrete pouring quality based on AI vision according to claim 1, characterized in that, The method for generating the image of the reinforcing steel area includes: Based on the images from the pouring intervals and the semantic segmentation results of the reinforcing bars, an initial semantic segmentation image of the reinforcing bars is generated; The initial semantic segmentation image of the reinforcing bars is contrast-enhanced to generate the image of the reinforcing bar region.

4. The method for real-time monitoring of precast beam concrete pouring quality based on AI vision according to claim 1, characterized in that, The method for generating the image of the concrete area includes: An initial concrete semantic segmentation image is generated based on the images from the pouring interval period and the concrete semantic segmentation results. Histogram equalization is performed on the initial concrete semantic segmentation image to generate the concrete region image.

5. The method for real-time monitoring of precast beam concrete pouring quality based on AI vision according to claim 1, characterized in that, By combining a 3D model of the precast railway beam reinforcement and an image set of the reinforcement area, the concrete height distribution data at each reinforcement submerged in concrete was obtained; including the following methods: Load the preset three-dimensional model of the railway precast beam reinforcement, which includes the spatial position parameters of each reinforcement bar; Spatially align the images of each reinforcing bar region in the image set with the three-dimensional model of the reinforcing bar of the precast railway beam, and register the positional relationship of each reinforcing bar in the images of the reinforcing bar regions and the three-dimensional model of the reinforcing bar of the precast railway beam; Based on the image recognition of each of the steel reinforcement regions, the spatial position parameters of the submerged positions of each steel reinforcement in the three-dimensional model of the railway precast beam steel reinforcement are identified, and the concrete height distribution data is set based on the spatial position parameters of the submerged positions of each steel reinforcement.

6. The method for real-time monitoring of precast beam concrete pouring quality based on AI vision according to claim 5, characterized in that, The comprehensive quality monitoring parameters include the comprehensive quality monitoring parameters of each sampling point corresponding to each of the steel bars submerged in concrete during each pouring interval; including the method: By combining the concrete height distribution data and the concrete surface anomaly analysis information, comprehensive quality monitoring parameters are calculated, and concrete quality monitoring results are generated based on the comprehensive quality monitoring parameters. Including methods: By combining the concrete height distribution data and the concrete surface anomaly analysis information, comprehensive quality monitoring parameters are calculated, and concrete quality monitoring results are generated based on the comprehensive quality monitoring parameters. Based on the concrete surface anomaly analysis information of the image set of the pouring interval period in the sampling grid constructed by connecting the sampling points in each of the pouring interval periods, a grid surface anomaly score of each sampling grid is generated, and the concrete surface anomaly score of the sampling point is obtained by summing the grid surface anomaly scores of the sampling grids in the neighborhood of the sampling point. The comprehensive quality monitoring parameters of the sampling point are calculated by weighting the concrete height distribution score and the concrete surface anomaly score. Based on the comprehensive quality monitoring parameters of the sampling points in the key functional areas and the weights of the key functional areas, the concrete quality monitoring results are generated.

7. The method for real-time monitoring of precast beam concrete pouring quality based on AI vision according to claim 6, characterized in that, The expressions for the concrete height distribution score and the concrete surface anomaly score are as follows: ； ； In the formula, and The first The sequence number of the intermittent pouring period is The concrete height distribution score and concrete surface anomaly score at the sampling points. The plane conformity coefficient. For the first The sequence number of the intermittent pouring period is The total number of sampling points, For the first The sequence number of the intermittent pouring period is The spatial location parameters of the sampling points, For the first The fitted plane obtained from sampling points during the intermittent pouring period. For the first The fitted plane obtained from the sampling points during the intermittent pouring period is in the sequence number. The spatial location fitting parameters at the sampling points, For the first The sequence number of the intermittent pouring period is The total number of sampling grids in the neighborhood of the sampling point. For the first The sequence number of the intermittent pouring period is The sampling point in the neighborhood of the first The total number of anomaly types in the grid sampling grid. For the first Weighting coefficients for different anomaly types For the first The sequence number of the intermittent pouring period is The sampling point in the neighborhood of the first The first in the sampling grid The size of the anomaly for each type of anomaly. For the first The maximum allowable size for each anomaly type.

8. A real-time monitoring system for the concrete pouring quality of precast beams based on AI vision, characterized in that, The system is used to implement the AI ​​vision-based real-time monitoring method for precast beam concrete pouring quality as described in any one of claims 1-7, the system comprising: The visual image acquisition module is used to acquire image sets during the intermittent periods of concrete pouring for precast railway beams based on a visual matrix sensing device. The semantic segmentation processing module is used to perform semantic segmentation on the image set of the pouring interval period to obtain the image set of the reinforcing steel area and the image set of the concrete area. The height anomaly identification module is used to load the three-dimensional model of the steel reinforcement of the precast railway beam, and combine it with the image set of the steel reinforcement area to obtain the concrete height distribution data of each steel reinforcement submerged in concrete. The surface anomaly identification module is used to input the preprocessed image set of the concrete area into the anomaly identification model to obtain concrete surface anomaly analysis information, which is used to characterize the type and size of anomalies on the concrete surface. The comprehensive quality assessment module is used to combine the concrete height distribution data and the concrete surface anomaly analysis information to calculate comprehensive quality monitoring parameters, and generate concrete quality monitoring result information based on the comprehensive quality monitoring parameters.

Images