Method and system for comprehensive evaluation of concrete appearance quality

Abstract

The application discloses a kind of concrete appearance quality comprehensive evaluation method, comprising the following steps: S1, collect concrete appearance image data and carry out data labeling of quality defect area;S2, based on the concrete appearance image data obtained in S1, construct deep learning instance segmentation network and train, obtain concrete appearance comprehensive defect segmentation model;Subsequently, the concrete appearance picture is input to the concrete appearance comprehensive defect segmentation model, and the defect area parameters are obtained by image segmentation model inference and image analysis, and data table is made;S3, the appearance quality score of each concrete appearance photo is used as the target value, and the defect area parameters obtained based on S2 are used as the characteristic value, to establish a concrete appearance quality evaluation machine learning model, output evaluation result, the application can solve the problem that the information obtained in the current comprehensive evaluation of concrete appearance quality is often not comprehensive enough, secondary integration is needed, and the corresponding evaluation information of engineering personnel cannot be given in time.

Description

Technical Field

[0001] This invention relates to the field of comprehensive evaluation methods for the appearance quality of concrete. More specifically, this invention relates to a comprehensive evaluation method and system for the appearance quality of concrete based on computer vision and numerical analysis. Background Technology

[0002] With societal development, the significance of concrete structures has transcended mere spatial requirements. Design now places greater emphasis on aesthetic appeal, demanding not only practicality but also visual appeal. From ordinary buildings to bridges spanning mountains and seas, the standards for concrete appearance are high regardless of the type of structure. The beauty of concrete structures requires not only creative design but also a reflection of the overall quality of the structure itself. Appearance assessment is a crucial aspect of quality inspection; however, due to regional variations and differences in construction skill levels, concrete appearance quality remains a challenge to manage during construction. Relevant departments have issued specific requirements. For instance, Chinese standard JGJ 169-2009 "Technical Specification for Application of Fair-faced Concrete" and American standard ACI 347.3R-13 "Guide to Formed Concrete Surfaces" stipulate that concrete surfaces should be smooth, uniform in color, and free from defects such as exposed reinforcement, honeycombing, voids, and slag inclusions. They also specify parameters such as the pore area for certain types of concrete.

[0003] Despite continuous research and optimization by technical personnel, the numerous factors influencing concrete appearance quality and the complex and varied control methods mean that some problems remain unresolved, causing considerable difficulties for relevant organizations. Current research attempts to identify concrete appearance quality issues using computer vision and numerical analysis algorithms. These algorithms can perform high-dimensional quantitative analysis of concrete appearance quality based on the digital features of acquired images, offering high accuracy and reliability, and have therefore been applied in some engineering projects. However, the information obtained from the comprehensive evaluation of concrete appearance quality using these technologies is often incomplete, requiring secondary integration, and timely provision of evaluation information to engineers is crucial. Furthermore, a mature system is lacking for engineers to use and make subsequent adjustments. Summary of the Invention

[0004] To achieve these objectives and other advantages according to the present invention, in one aspect, a preferred embodiment of the present invention provides a method for comprehensive evaluation of the appearance quality of concrete, comprising the following steps:

[0005] S1. Collect concrete appearance image data and annotate the data of quality defect areas;

[0006] S2. Based on the concrete appearance image data obtained in S1, a deep learning instance segmentation network is constructed and trained to obtain a comprehensive concrete appearance defect segmentation model. Subsequently, the concrete appearance image is input into the comprehensive concrete appearance defect segmentation model. The defect area parameters are obtained through image segmentation model inference and image analysis, and a data table is generated.

[0007] S3. Score the appearance quality of each concrete appearance photo as the target value, and establish a machine learning model for concrete appearance quality evaluation based on the defect area parameters obtained in S2 as feature values, and output the evaluation results.

[0008] Preferably, in step S1, image data is acquired using a camera device. During the acquisition process, the distance between the camera device and the concrete surface is determined, and the original image captured by the camera device is pre-adjusted according to the calibration conditions.

[0009] The types of markings for the quality defect areas include: color difference areas, crack areas, honeycomb areas, and pitted areas, all of which are marked using the polygon marking method.

[0010] Preferably, in step S2, after inputting the concrete appearance image into the comprehensive defect segmentation model of the concrete appearance, a target mask is obtained, and image processing is performed on it to extract quantitative numerical information; wherein,

[0011] For the crack region, the length and maximum width of the crack are obtained using a skeleton-based elongated numerical extraction method.

[0012] For color difference areas, a conventional numerical extraction method based on circumscribed polygons is used to extract quantized numerical information.

[0013] Preferably, for the crack region, the length and maximum width of the crack are obtained using a skeleton-based elongated numerical extraction method, including the following steps:

[0014] S21. Using the image skeleton extraction method, the connected regions of the crack mask are refined to the width of one pixel, and an undirected cyclic skeleton line graph is established accordingly.

[0015] S22. The length information of the crack can be directly obtained from the undirected cyclic skeleton line diagram. The maximum inscribed circle at each point of the crack can be obtained by traversing the undirected cyclic skeleton line diagram with a circle, which is the maximum width at that point.

[0016] Preferably, for color difference areas, honeycomb areas, and pitted areas, a conventional numerical extraction method based on circumscribed polygons is used to extract quantitative numerical information. By finding the smallest circumscribed rectangle or ellipse of the irregular area, the length, width, and area of ​​the area are obtained. For color difference areas, the CIE average difference between the color difference area and the normal color of the concrete surface is calculated.

[0017] Preferably, for the crack region, the length and maximum width of the crack are obtained using a skeleton-based elongated numerical extraction method; and,

[0018] For color difference areas, a conventional numerical extraction method based on circumscribed polygons is used to extract quantified numerical information. This is achieved by using the distance between the shooting device (which collects data via S1) and the concrete surface, and then converting the image pixel values ​​into corresponding real values ​​proportionally. The calculated values ​​are obtained using the following formula:

[0019]

[0020] Where: for the length of any line segment in the image,

[0021] P real , which represents the distance in pixels between two endpoints in the image in real-world conditions;

[0022] D real This represents the actual distance between the camera and the concrete surface.

[0023] L base This represents the distance in pixels between the two endpoints in reality under the calibration condition.

[0024] P base , which represents the distance in pixels between the two endpoints in the image under the calibration condition;

[0025] D base , represents the distance between the camera and the concrete surface under calibration conditions.

[0026] Preferably, in step S3, the appearance quality score is divided into five levels from good to bad: A, B, C, D, and E, which are converted into numerical values ​​of 1, 2, 3, 4, and 5.

[0027] The characteristic values ​​obtained by S2 include: the number, average length, and average maximum width of cracked areas; the number, average area, and average CIE color difference between color difference areas and the concrete's own appearance color; and the number and average area of ​​honeycomb and pitted areas. The formula for calculating the CIE color difference is as follows:

[0028]

[0029] In the formula:

[0030] L a ,L b - The brightness values ​​of the normal color of the concrete surface and the color of the color difference area of ​​the concrete in the CIE color space;

[0031] A aA b - The red and green axis values ​​of the normal color of the concrete surface and the color difference area of ​​the concrete in the CIE color space;

[0032] B a B b - The blue and yellow axis values ​​of the normal color of the concrete surface and the color difference area of ​​the concrete in the CIE color space.

[0033] Preferably, the step S3, which involves establishing a machine learning model for evaluating the appearance quality of concrete, includes the following steps:

[0034] S31. The concrete appearance feature value data obtained from S2 will be divided into training set, validation set and test set;

[0035] S32. Select six machine learning algorithms, including CatBoost, XgbBoost, LightGBM, ExtraTrees, RandomTree, and DecisionTree. Train each of these six algorithms on the training set using the default hyperparameters, and select the three algorithms with the highest scores on the validation set to proceed to the next stage.

[0036] S33. Set the range of optimized hyperparameters, and use the 5-fold cross-validation method and the Bayesian hyperparameter optimization method to adjust them on the subsets composed of the training set and the validation set, respectively, to obtain the best hyperparameter combination for each of the three algorithms. Finally, introduce the optimized hyperparameter combination settings into the three algorithms obtained in S32, and select the model with the highest prediction score on the test set as the machine learning model for evaluating the appearance quality of concrete.

[0037] On the other hand, another technical solution of this application provides an evaluation system for the comprehensive evaluation method of concrete appearance quality according to any one of claims 1-8, comprising:

[0038] Data acquisition, addition, and preprocessing module: used to acquire concrete appearance image data and annotate the areas with quality defects;

[0039] Model training and inference module: This module is used to construct and train a deep learning instance segmentation network based on concrete appearance image data to obtain a comprehensive concrete appearance defect segmentation model. Subsequently, concrete appearance images are input into this comprehensive concrete appearance defect segmentation model. The defect area parameters are obtained through image segmentation model inference and image analysis, and a data table is generated. The appearance quality of each concrete appearance photo is scored as the target value, and a machine learning model for evaluating concrete appearance quality is established based on the defect area parameters as feature values, outputting the evaluation results.

[0040] Adjustment and Management Module: Used to add, delete, and view the images and quantitative data required for the model, and to adjust the output types of expert base scores and quantitative numerical results of concrete appearance quality.

[0041] This invention offers at least the following advantages: It can quickly extract and integrate various concrete appearance defect areas from different concrete appearance images, reducing the need for experience-based inspections, equipment requirements, and repetitive testing, thus improving efficiency and effectively avoiding subjectivity issues during the inspection process. Furthermore, it boasts advantages such as short inspection time, high inspection accuracy, and significantly enhanced concrete appearance defect detection capabilities. Simultaneously, it rapidly assesses the overall quality level of concrete appearance based on an expert scoring database, and provides accurate quantitative and evaluation information on appearance defects through masking results obtained from image analysis. Moreover, based on the above technology, a data management and technology usage platform is provided, facilitating engineers to quickly apply the relevant technologies and make timely decisions.

[0042] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Attached Figure Description

[0043] Figure 1 This is a flowchart of the comprehensive evaluation method for concrete appearance quality based on computer vision and numerical analysis according to an embodiment of the present invention.

[0044] Figure 2 This is a schematic diagram illustrating the general target analysis method based on circumscribed polygons and the slender target analysis method based on skeleton model, according to embodiments of the present invention.

[0045] Figure 3 This is a schematic diagram of the loss curve during the training process of the comprehensive defect segmentation model for concrete appearance according to an embodiment of the present invention.

[0046] Figure 4 This is a diagram illustrating the prediction results obtained from the comprehensive defect segmentation model for concrete appearance according to an embodiment of the present invention.

[0047] Figure 5 This is a comparative diagram showing the prediction results obtained by the machine learning model for evaluating the appearance quality of concrete in an embodiment of the present invention. Detailed Implementation

[0048] The present invention will now be described in further detail with reference to the accompanying drawings, so that those skilled in the art can implement it based on the description.

[0049] The following description is intended to disclose the present invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious modifications will occur to those skilled in the art. The basic principles of the invention defined in the following description can be applied to other embodiments, modifications, improvements, equivalents, and other technical solutions that do not depart from the spirit and scope of the invention.

[0050] Those skilled in the art should understand that, in the disclosure of this invention, the terms "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the above terms should not be construed as limiting this invention.

[0051] It is understood that the term "a" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of an element can be one, while in another embodiment, the number of the element can be multiple, and the term "a" should not be understood as a limitation on the number.

[0052] like Figure 1-5 As shown, in one preferred embodiment of the present invention, a method for comprehensive evaluation of the appearance quality of concrete is provided, comprising the following steps:

[0053] S1. Collect concrete appearance image data and annotate the quality defect areas; the types of quality defect areas to be annotated include: color difference areas, crack areas, honeycomb areas, and pitted areas, and all of them are annotated using the polygon annotation method.

[0054] In step S1, the data acquisition process includes manual and drone photography. During the photography process, the distance between the photography device and the concrete surface is measured manually or recorded using infrared light. For the same concrete surface area, two images are taken. The first image shows the original surface area, referred to as K1; the second image is taken using a calibration plate placed close to the concrete surface to measure the shooting angle, referred to as K2. Based on this information, the original image is pre-adjusted to a standard parallel forward image, referred to as K3. In this embodiment, if a depth camera is used for direct photography, there is no need to record distance and calibration information; the image can be directly adjusted using the depth matrix obtained from the depth camera, equivalent to K3. In this embodiment, the HikVision series camera was selected for the monocular camera, and the Intel RealSense series camera was selected for the binocular depth camera. The recommended shooting distance is between 50-350cm.

[0055] After the images were taken, all K3-class images and distance information were collected to construct a concrete appearance dataset. The image quality defect areas in the dataset are labeled in four ways: color difference areas, crack areas, and other defect areas, all using polygon annotation methods.

[0056] Meanwhile, due to the limited number of concrete appearance quality images, their quality is significantly affected by environmental conditions and shooting skills, and they differ considerably from natural images. Therefore, before formally starting model training, additional high-quality concrete appearance images were captured under laboratory conditions to obtain a pre-trained model. The aforementioned concrete appearance dataset was then used for model fine-tuning. Finally, all processed image experimental data was divided into three datasets: a training set, a validation set, and a prediction set. All data in the prediction set originated from the concrete appearance dataset.

[0057] S2. Based on the concrete appearance image data obtained in S1, a deep learning instance segmentation network is constructed and trained to obtain a comprehensive defect segmentation model for concrete appearance. Subsequently, the concrete appearance image is input into the comprehensive defect segmentation model for concrete appearance. The defect region parameters, such as the area and length of the defect region, are obtained through image segmentation model inference and image analysis, and the results are compiled into a data table.

[0058] In this embodiment, the network used in the comprehensive defect segmentation model for concrete appearance is an instance segmentation network based on the improved ConvNeXt-V2 backbone network. The backbone network uses the ConvNeXt-V2-Tiny version, the Neck part has 3 FPNs, and the Head part uses a conventional one-stage segmentation head.

[0059] Specifically, in this embodiment, the optimizer is SGD, the training iterations are 200, and the initial learning rate is set to 10. -3 It decreases linearly with increasing training iterations, eventually dropping to 10. -6 Focal Loss was chosen as the loss function to address the long-tail imbalance problem. The loss curve during training is shown below. Figure 3 .

[0060] After obtaining the comprehensive defect segmentation model for concrete appearance, input a concrete appearance image. The model will output the defect area mask and category determined from the image. See the schematic diagram for specific results. Figure 4 After obtaining the output mask, image processing is needed to extract quantized numerical information, such as area, length, and width. For crack-like targets, a skeleton-based elongated numerical extraction method is used to obtain the crack length and maximum width. For other defect areas, including color difference regions, a conventional numerical extraction method based on circumscribed polygons is used. See the schematic diagram below. Figure 2 .

[0061] S3. Score the appearance quality of each concrete appearance photo as the target value, and establish a machine learning model for concrete appearance quality evaluation based on the defect area parameters obtained in S2 as feature values, and output the evaluation results.

[0062] In this embodiment, to establish a machine learning model for evaluating the appearance quality of concrete, a feature set and a target value are needed. Based on the appearance quality of concrete, concrete designers and on-site project engineers are invited to score the images in the concrete appearance image dataset. The scores are divided into five levels from best to worst: A, B, C, D, and E, which are represented by the numerical values ​​1, 2, 3, 4, and 5. The maximum and minimum values ​​are discarded, and the average is used to obtain the appearance quality score for each image as the target value. The feature set is obtained from the quantified values ​​in S2 and includes: the number of cracks, their average length, and average maximum width; the number of color difference areas, their average area, and the average CIE difference between these areas and the concrete's original color; and the number and average area of ​​other defective areas.

[0063] In another technical solution, in step S2, after inputting the concrete appearance image into the comprehensive defect segmentation model of the concrete appearance, a target mask is obtained, and image processing is performed on it to extract quantitative numerical information; wherein,

[0064] For the crack region, the length and maximum width of the crack are obtained using a skeleton-based elongated numerical extraction method.

[0065] For color difference areas, a conventional numerical extraction method based on circumscribed polygons is used to extract quantized numerical information.

[0066] in,

[0067] For the crack region, the length and maximum width of the crack are obtained using a skeleton-based elongated numerical extraction method, including the following steps:

[0068] S21. Using the image skeleton extraction method, the connected regions of the crack mask are refined to the width of one pixel, and an undirected cyclic skeleton line graph is established accordingly.

[0069] S22. The length information of the crack can be directly obtained from the undirected cyclic skeleton line diagram. The maximum inscribed circle at each point of the crack can be obtained by traversing the undirected cyclic skeleton line diagram with a circle, which is the maximum width at that point.

[0070] In another technical solution, for color difference areas, honeycomb areas, and pitted areas, a conventional numerical extraction method based on circumscribed polygons is used to extract quantitative numerical information. By finding the smallest circumscribed rectangle or ellipse of the irregular area, the length, width, and area of ​​the area are obtained. For color difference areas, the CIE average difference between the color difference area and the normal color of the concrete is calculated.

[0071] In another technical solution, for the crack region, a skeleton-based elongated numerical extraction method is used to obtain the crack length and maximum width; and,

[0072] For color difference areas, a conventional numerical extraction method based on circumscribed polygons is used to extract quantified numerical information. This is achieved by using the distance between the shooting device (which collects data via S1) and the concrete surface, and then converting the image pixel values ​​into corresponding real values ​​proportionally. The calculated values ​​are obtained using the following formula:

[0073]

[0074] Where: for the length of any line segment in the image,

[0075] P real , which represents the distance in pixels between two endpoints in the image in real-world conditions;

[0076] D real This represents the actual distance between the camera and the concrete surface.

[0077] L base This represents the distance in pixels between the two endpoints in reality under the calibration condition.

[0078] P base , which represents the distance in pixels between the two endpoints in the image under the calibration condition;

[0079] D base , represents the distance between the camera and the concrete surface under calibration conditions.

[0080] The results of comparing the numerical information of concrete appearance quality defects obtained by image analysis with the corresponding manual measurement information are shown in Table 1, which verifies the accuracy of the image analysis method.

[0081] Table 1 Comparison of numerical information analysis and measured values ​​of concrete appearance quality defects.

[0082]

[0083] In another technical solution, in step S3, the appearance quality score is divided into five levels from good to bad: A, B, C, D, and E, which are converted into numerical values ​​of 1, 2, 3, 4, and 5.

[0084] The characteristic values ​​obtained by S2 include: the number, average length, and average maximum width of cracked areas; the number, average area, and average CIE color difference between color difference areas and the concrete's own appearance color; and the number and average area of ​​honeycomb and pitted areas. The formula for calculating the CIE color difference is as follows:

[0085]

[0086] In the formula:

[0087] L a ,L b - The brightness values ​​of the normal color of the concrete surface and the color of the color difference area of ​​the concrete in the CIE color space;

[0088] A a A b - The red and green axis values ​​of the normal color of the concrete surface and the color difference area of ​​the concrete in the CIE color space;

[0089] B a B b - The blue and yellow axis values ​​of the normal color of the concrete surface and the color difference area of ​​the concrete in the CIE color space.

[0090] In another technical solution, S3 specifically includes the following steps:

[0091] S31. The concrete appearance feature data obtained from S2 will be divided into training set, validation set and test set; the ratio is approximately 7:2:1.

[0092] S32. Since it was unknown in advance which machine learning algorithm was suitable for building a machine learning model for evaluating the appearance quality of concrete, six machine learning algorithms, including CatBoost, XgbBoost, LightGBM, ExtraTrees, RandomTree, and DecisionTree, were introduced at the same time. The above six algorithms were trained on the training set using default hyperparameters, and the three algorithms with the highest scores on the validation set were selected to enter the next stage.

[0093] S33. Set the range of optimized hyperparameters, and use the 5-fold cross-validation method and the Bayesian hyperparameter optimization method to adjust them on the subsets composed of the training set and the validation set, respectively, to obtain the best hyperparameter combination for each of the three algorithms. Finally, introduce the optimized hyperparameter combination settings into the three algorithms obtained in S32, and select the model with the highest prediction score on the test set as the machine learning model for evaluating the appearance quality of concrete.

[0094] In this embodiment, a total of 310 sets of data were used to construct the machine learning model for evaluating the appearance quality of concrete. The CatBoost algorithm was ultimately selected as the best algorithm, with an accuracy of 0.86. Figure 5 The test results for 31 sets of test data are shown.

[0095] On the other hand, another technical solution of this application provides an evaluation system for a comprehensive evaluation method of concrete appearance quality, comprising:

[0096] Data acquisition, addition, and preprocessing module: used to acquire concrete appearance image data and annotate the areas with quality defects;

[0097] Model training and inference module: This module is used to construct and train a deep learning instance segmentation network based on concrete appearance image data to obtain a comprehensive concrete appearance defect segmentation model. Subsequently, concrete appearance images are input into this comprehensive concrete appearance defect segmentation model. The defect area parameters are obtained through image segmentation model inference and image analysis, and a data table is generated. The appearance quality of each concrete appearance photo is scored as the target value, and a machine learning model for evaluating concrete appearance quality is established based on the defect area parameters as feature values, outputting the evaluation results.

[0098] Adjustment and Management Module: Used to add, delete, and view the images and quantitative data required for the model, and to adjust the output types of expert base scores and quantitative numerical results of concrete appearance quality.

[0099] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.

Claims

1. A comprehensive evaluation method for the appearance quality of concrete, characterized in that, Includes the following steps: S1. Collect concrete appearance image data and annotate the data of quality defect areas; S2. Based on the concrete appearance image data obtained in S1, a deep learning instance segmentation network is constructed and trained to obtain a comprehensive concrete appearance defect segmentation model. Subsequently, the concrete appearance image is input into the comprehensive concrete appearance defect segmentation model. The defect area parameters are obtained through image segmentation model inference and image analysis, and a data table is generated. S3. Score the appearance quality of each concrete appearance photo as the target value, and use the defect area parameters obtained in S2 as the feature values ​​to establish a machine learning model for concrete appearance quality evaluation and output the evaluation results. For the crack region, the length and maximum width of the crack are obtained using a skeleton-based elongated numerical extraction method, including the following steps: S21. Using the image skeleton extraction method, the connected regions of the crack mask are refined to the width of one pixel, and an undirected cyclic skeleton line graph is established accordingly. S22. The length information of the crack can be obtained directly from the undirected cyclic skeleton line diagram. The maximum inscribed circle at each point of the crack can be obtained by traversing the undirected cyclic skeleton line diagram with a circle, which is the maximum width at that point. For color difference areas, honeycomb areas, and pitted areas, a conventional numerical extraction method based on circumscribed polygons is used to extract quantitative numerical information. By finding the minimum circumscribed rectangle or ellipse of the irregular area, the length, width, and area of ​​the area are obtained. For color difference areas, the CIE average difference between the color difference area and the normal color of the concrete surface is calculated. For the crack region, a skeleton-based elongated numerical extraction method is used to obtain the crack length and maximum width; and, For color difference areas, a conventional numerical extraction method based on circumscribed polygons is used to extract quantified numerical information. This is achieved by using the distance between the shooting device (which collects data via S1) and the concrete surface, and then converting the image pixel values ​​into corresponding real values ​​proportionally. The calculated values ​​are obtained using the following formula: Where: for the length of any line segment in the image, , which represents the distance in pixels between two endpoints in the image in real-world conditions; This represents the actual distance between the camera and the concrete surface. This represents the distance in pixels between the two endpoints in reality under the calibration condition. , which represents the distance in pixels between the two endpoints in the image under the calibration condition; , represents the distance between the camera and the concrete surface under calibration conditions.

2. The comprehensive evaluation method for concrete appearance quality according to claim 1, characterized in that, In step S1, image data is collected using a camera device. During the collection process, the distance between the camera device and the concrete surface is determined, and the original image captured by the camera device is pre-adjusted according to the calibration. The types of markings for the quality defect areas include: color difference areas, crack areas, honeycomb areas, and pitted areas, all of which are marked using the polygon marking method.

3. The comprehensive evaluation method for concrete appearance quality according to claim 2, characterized in that, In step S2, after inputting the concrete appearance image into the comprehensive defect segmentation model for concrete appearance, a target mask is obtained, and image processing is performed on it to extract quantitative numerical information; wherein, For the crack region, the length and maximum width of the crack are obtained using a skeleton-based elongated numerical extraction method. For color difference areas, a conventional numerical extraction method based on circumscribed polygons is used to extract quantized numerical information.

4. The comprehensive evaluation method for concrete appearance quality according to claim 1, characterized in that, In S3, the appearance quality score is divided into five levels from good to bad: A, B, C, D, and E, which are converted into numerical values ​​of 1, 2, 3, 4, and 5. The characteristic values ​​obtained by S2 include: the number, average length, and average maximum width of cracked areas; the number, average area, and average CIE color difference between color difference areas and the concrete's own appearance color; and the number and average area of ​​honeycomb and pitted areas. The formula for calculating the CIE color difference is as follows: In the formula: - Brightness values ​​of the normal color and the color difference area of ​​the concrete surface in the CIE color space; - The red and green axis values ​​of the normal color of the concrete surface and the color difference area of ​​the concrete in the CIE color space; - The blue and yellow axis values ​​of the normal color of the concrete surface and the color difference area of ​​the concrete in the CIE color space.

5. The comprehensive evaluation method for concrete appearance quality based on computer vision and numerical analysis according to claim 1, characterized in that, In step S3, a machine learning model for evaluating the appearance quality of concrete is established, including the following steps: S31. The concrete appearance feature value data obtained from S2 will be divided into training set, validation set and test set; S32. Select six machine learning algorithms, including CatBoost, XgbBoost, LightGBM, ExtraTrees, RandomTree, and DecisionTree. Train each of these six algorithms on the training set using the default hyperparameters, and select the three algorithms with the highest scores on the validation set to proceed to the next stage. S33. Set the range of optimized hyperparameters, and use the 5-fold cross-validation method and the Bayesian hyperparameter optimization method to adjust them on the subsets composed of the training set and the validation set, respectively, to obtain the best hyperparameter combination for each of the three algorithms. Finally, introduce the optimized hyperparameter combination settings into the three algorithms obtained in S32, and select the model with the highest prediction score on the test set as the machine learning model for evaluating the appearance quality of concrete.

6. An evaluation system for the comprehensive evaluation method of concrete appearance quality according to any one of claims 1-5, characterized in that: include: Data acquisition, addition, and preprocessing module: used to acquire concrete appearance image data and annotate the areas with quality defects; Model training and inference module: This module is used to construct and train a deep learning instance segmentation network based on concrete appearance image data to obtain a comprehensive concrete appearance defect segmentation model. Subsequently, concrete appearance images are input into this comprehensive concrete appearance defect segmentation model. The defect area parameters are obtained through image segmentation model inference and image analysis, and a data table is generated. The appearance quality of each concrete appearance photo is scored as the target value, and a machine learning model for evaluating concrete appearance quality is established based on the defect area parameters as feature values, outputting the evaluation results. Adjustment and Management Module: Used to add, delete, and view the images and quantitative data required for the model, and to adjust the output types of expert base scores and quantitative numerical results of concrete appearance quality.

Images