A mockup-based digital evaluation method and system for fair-faced concrete

Abstract

This invention provides a digital evaluation method and system for fair-faced concrete based on mockups, belonging to the field of building construction. The technical solution is as follows: Surface images of fair-faced concrete mockup templates are acquired; baseline Lab color values ​​and baseline texture features are extracted from the acquired template images to construct an evaluation baseline model; surface images of the fair-faced concrete component to be evaluated are acquired, generating image data; defect area identification is performed on the image data, extracting valid image areas; and Lab color values ​​and texture features to be evaluated are extracted from the valid image areas; comparison results are generated based on the evaluation baseline model; and evaluation output information is generated based on the comparison results. The beneficial effects of this invention are: by constructing a dual-feature model of color and texture, introducing a defect identification and masking mechanism, and adopting a structured comparison algorithm and output system, the entire process from data acquisition, feature extraction, defect avoidance to intelligent comparison and evaluation output is automated.

Description

Technical Field

[0001] This invention relates to the field of building construction, and in particular to a digital evaluation method and system for fair-faced concrete based on Mockup. Background Technology

[0002] With the continuous improvement of urban building quality, exposed concrete has become one of the important materials for modern building facade decoration due to its natural texture and minimalist aesthetics. To ensure the consistency and quality control of its large-scale construction, the construction industry has widely introduced the "Mockup first" mechanism, which involves first making and accepting standard templates, and then using them to guide the construction of subsequent components. However, in practical applications, how to compare and evaluate the surface consistency between the mockup template and the actual components in a quantifiable and traceable way remains one of the key issues in engineering quality control. Traditional evaluation methods are mainly based on manual visual inspection, which is easily affected by the observation angle, lighting environment and individual experience. They are highly subjective and difficult to achieve stable and reliable quality judgment in large-scale construction. Therefore, digital evaluation methods that integrate computer vision and image recognition algorithms have gradually become a research hotspot.

[0003] However, existing digital image processing methods are mostly limited to color distribution analysis or single texture quantization, and have not yet formed a dedicated technical framework for the evaluation of fair-faced concrete surfaces. On the one hand, most existing technologies are based on directly calculating color or texture features from the entire image, lacking a mechanism for identifying and masking surface defect areas, which leads to abnormal areas interfering with the overall evaluation accuracy. On the other hand, evaluation indicators often use simple color difference thresholds or coarse texture matching algorithms, which cannot truly reflect the microscopic differences of fair-faced concrete surfaces at the visual perception level. In addition, most current systems lack a structured output method, making it difficult to achieve standardized use in construction supervision, defect tracing, and quality archiving. Therefore, how to construct a digital evaluation method that combines color and texture dimensions, has a defect tolerance mechanism, and can output structured results has become a technical bottleneck that urgently needs to be overcome. Summary of the Invention

[0004] The purpose of this invention is to provide a digital evaluation method and system for fair-faced concrete based on Mockup, which automates the entire process from data acquisition, feature extraction, defect avoidance to intelligent comparison and evaluation output by constructing a dual feature model of color and texture, introducing a defect identification and masking mechanism, and adopting a structured comparison algorithm and output system.

[0005] This invention is achieved through the following measures:

[0006] A digital evaluation method based on fair-faced concrete mockup, characterized by comprising:

[0007] Surface images of fair-faced concrete mockup samples were collected, and baseline Lab color values ​​and baseline texture features were extracted from the collected sample images to construct an evaluation baseline model.

[0008] Collect surface images of the fair-faced concrete component to be evaluated and generate image data;

[0009] Defect region identification is performed on the image data, a mask map is generated, and the defect region is masked using the mask map to extract the effective image region. The Lab color value and texture feature to be evaluated are extracted from the effective image region.

[0010] Based on the evaluation benchmark model, the similarity of the texture features to be evaluated in the effective image region is compared with the benchmark texture features of the sample image, and the color difference between the Lab color value to be evaluated in the effective image region and the benchmark Lab color value of the sample image is calculated to generate the comparison result.

[0011] The invention also has the following specific features:

[0012] Surface images of fair-faced concrete mockup samples were acquired, and baseline Lab color values ​​and baseline texture features were extracted from the acquired sample images, including:

[0013] The surface image of the fair-faced concrete mockup sample is acquired by an image acquisition device. The surface image is preprocessed and then converted to the CIE Lab color space to extract the baseline Lab color value. At the same time, the baseline texture feature of the image is extracted by the gray-level co-occurrence matrix. Finally, the baseline Lab color value and baseline texture feature are obtained for constructing the evaluation baseline model.

[0014] The construction of the evaluation benchmark model includes: constructing corresponding color feature datasets and texture feature datasets based on the benchmark Lab color values ​​and benchmark texture features extracted from the sample image, respectively;

[0015] Statistical processing and standardization are performed on the color feature dataset and texture feature dataset respectively to obtain color description vectors and texture description vectors. The color description vectors and texture description vectors are combined according to a fixed format to construct a feature matrix for evaluating the benchmark model. The construction of the benchmark model is completed when the preset trigger conditions are met.

[0016] The process of acquiring surface images of the fair-faced concrete component to be evaluated and generating image data includes: obtaining an image sequence covering the surface of the fair-faced concrete component through an image acquisition device, and performing format conversion, illumination adjustment and noise suppression operations on the image sequence after acquisition to obtain structured image data.

[0017] Performing defect region identification on image data, generating a mask image, and using the mask image to mask defect regions in order to extract valid image regions includes: performing pixel-level defect identification operations on the image data to generate a mask image representing defect regions and non-defect regions;

[0018] Based on the information marked by the mask image, the corresponding defective areas in the image data are masked, while the pixel content corresponding to the non-defective areas in the image data is retained, and the effective image area is extracted.

[0019] Extracting the Lab color value and texture feature to be evaluated from the effective image region includes: performing image preprocessing, image color space conversion and texture feature calculation operations on the image frames in the effective image region to obtain the Lab color value and texture feature to be evaluated.

[0020] Based on the evaluation benchmark model, the similarity of the texture features to be evaluated in the effective image region is compared with the benchmark texture features in the sample image. The color difference between the Lab color values ​​to be evaluated in the effective image region and the benchmark Lab color values ​​in the sample image is calculated, and the comparison results are generated as follows:

[0021] The evaluation process involves calling the color description vector and texture description vector from the benchmark model, and calculating them against the Lab color value and texture feature extracted from the effective image region. During texture feature comparison, a preset similarity calculation method is used to match the texture feature to be evaluated with the benchmark texture feature, outputting a texture similarity score. During color value comparison, a standard color difference formula is used to compare the Lab color value to be evaluated with the benchmark Lab color value, outputting a color difference value. Finally, the texture similarity score and color difference value are combined and the comparison result is output.

[0022] Based on the comparison results, the evaluation output information related to the fair-faced concrete component to be evaluated is generated, including: performing classification judgment operations based on the comparison results, and generating structured evaluation output information according to the preset evaluation rules.

[0023] A digital evaluation system based on fair-faced concrete mockup, characterized in that it includes:

[0024] Evaluation benchmark model construction module: Collect surface images of fair-faced concrete mockup samples, extract benchmark Lab color values ​​and benchmark texture features from the collected sample images, and construct the evaluation benchmark model;

[0025] Image acquisition module for evaluation: Acquires surface images of the fair-faced concrete component to be evaluated and generates image data;

[0026] Image preprocessing and effective region extraction module: performs defect region identification on image data, generates a mask image, and uses the mask image to mask defect regions in order to extract effective image regions. It then extracts the Lab color value and texture features to be evaluated from the effective image regions.

[0027] Texture and color comparison calculation module: Based on the evaluation benchmark model, the similarity of the texture features to be evaluated in the effective image area is compared with the benchmark texture features of the sample image, and the color difference of the Lab color value to be evaluated in the effective image area is calculated with the benchmark Lab color value of the sample image to generate the comparison result.

[0028] Evaluation output generation module: Based on the comparison results, it generates evaluation output information related to the fair-faced concrete component to be evaluated.

[0029] The beneficial effects of this invention are as follows: By constructing a dual-feature model of color and texture, introducing a defect identification and masking mechanism, and adopting a structured comparison algorithm and output system, this invention achieves full automation of the entire process from data acquisition, feature extraction, defect avoidance to intelligent comparison and evaluation output.

[0030] Compared with existing traditional methods that rely mainly on manual visual inspection or single-dimensional comparison, this invention not only significantly improves the objectivity and consistency of the evaluation, but also enhances the quantifiability, repeatability and traceability of the comparison process. Attached Figure Description

[0031] Figure 1 The present invention provides an overall flowchart of a digital evaluation method based on fair-faced concrete mockup. Detailed Implementation

[0032] To clearly illustrate the technical features of this solution, the following detailed implementation method will be used to explain the solution.

[0033] Example 1

[0034] See Figure 1 A digital evaluation method based on fair-faced concrete mockup, characterized by comprising:

[0035] Step S1: Acquire a surface image of a fair-faced concrete mockup sample, and extract the baseline Lab color value and baseline texture features from the acquired sample image, including:

[0036] The surface image of the fair-faced concrete mockup sample is acquired by an image acquisition device. The surface image is preprocessed and then converted to the CIE Lab color space to extract the baseline Lab color value. At the same time, the baseline texture feature of the image is extracted by the gray-level co-occurrence matrix. Finally, the baseline Lab color value and baseline texture feature are obtained for constructing the evaluation baseline model.

[0037] The image acquisition device is used to acquire surface images of fair-faced concrete mockup samples. It preferably uses a high-resolution industrial camera with an industrial-grade image sensor, an industrial lens, and a constant-current driven LED array light source to provide a stable and uniform imaging environment. To ensure the consistency of the acquired images, the image acquisition device is fixedly installed by a structural support or track platform, the sampling angle and sampling distance are kept constant, and the triggering timing is controlled by the acquisition control module.

[0038] The acquisition control module is used to control the start time of image acquisition. It can be implemented based on an embedded system, such as Jetson Nano, Raspberry Pi or STM32 microcontroller unit. The module monitors the exposure stability, average brightness and sampling alignment status of the image sensor in real time. If multiple consecutive frames of images meet the preset trigger threshold (such as brightness standard deviation less than the set value, image center of gravity drift less than 1 pixel, etc.), the acquisition command is automatically sent to ensure that the sampled images are comparable and representative.

[0039] The acquired image is input into the image preprocessing module, where image denoising (using Gaussian filtering), white balance correction, and illumination normalization are performed sequentially to enhance color reproduction and boundary information representation. After preprocessing, the color space conversion module is called to perform color space mapping on the image, converting it from the RGB color space to the CIE Lab color space. The conversion process follows the CIE standard color management process, using the intermediate XYZ space as a transition and combining it with the D65 reference white point for nonlinear transformation to extract the three components L (luminance), a (red-green axis), and b (yellow-blue axis) of each pixel in the image, thereby constructing the baseline Lab color value distribution of the entire image.

[0040] In addition, to extract baseline texture features, the preprocessed image is first converted into a grayscale image, and then a gray-level co-occurrence matrix (GLCM) is constructed by setting a local sliding window (e.g., 7×7 pixels) on the image. Within each window, the co-occurrence frequency of different gray-level pairs is statistically analyzed according to a fixed spatial direction (0°, 45°, 90°, 135°), and four texture statistical indicators are calculated, including contrast, energy, entropy, and correlation. These indicators reflect the differences and structure of the image in spatial gray-level changes. Finally, the indicator set is standardized and aggregated into a unified texture description vector according to the window moving strategy, which is the baseline texture feature.

[0041] The aforementioned baseline Lab color values ​​and baseline texture features serve as the foundational data for constructing subsequent evaluation baseline models, providing an objective quantitative basis for evaluating the digital similarity between Mockup templates and actual components. This combined method exhibits strong stability and engineering adaptability, effectively enhancing the reliability and consistency of fair-faced concrete components during the digital comparison process.

[0042] The construction of the evaluation benchmark model includes: constructing corresponding color feature datasets and texture feature datasets based on the benchmark Lab color values ​​and benchmark texture features extracted from the sample image, respectively;

[0043] Statistical processing and standardization are performed on the color feature dataset and texture feature dataset respectively to obtain color description vectors and texture description vectors. The color description vectors and texture description vectors are combined according to a fixed format to construct a feature matrix for evaluating the benchmark model. The construction of the benchmark model is completed when the preset trigger conditions are met.

[0044] The evaluation benchmark model was constructed based on benchmark Lab color values ​​and benchmark texture features extracted from images of fair-faced concrete mockup templates. Color benchmark datasets and texture benchmark datasets were constructed separately. To improve the stability and representativeness of the model, cluster analysis was performed on the Lab values ​​of each image pixel in the color benchmark dataset. K-means or Mean Shift clustering algorithms were preferably used to extract the main color distribution regions, forming clustered color feature centers. This clustering process not only eliminates background interference but also more accurately captures the main color regions of the fair-faced concrete surface. Furthermore, statistical parameters such as color mean, color standard deviation, and main color distribution density were calculated for the clustered samples, and these were combined to construct a color description vector. This description vector reflects the typical color distribution characteristics of the mockup template in Lab space.

[0045] For the construction of the texture benchmark dataset, the texture statistical indicators extracted from the gray-level co-occurrence matrix are used as the original feature dimensions, including image contrast, entropy, energy, and correlation. To enhance the model's sensitivity to differences in surface structure among different sample plates, feature selection strategies (such as variance filtering or mutual information ranking) are introduced to remove feature components with poor stability or weak correlation. The Z-score normalization method is then used to unify the numerical scale of different statistics. Finally, a stable and quantifiable texture description vector is formed.

[0046] Color and texture description vectors are concatenated according to a fixed-dimensional rule to construct a feature matrix, which serves as the core parameter structure of the evaluation benchmark model. This dimensional combination ensures that color and texture information have equal weight in the model and can be evaluated independently, avoiding biased judgments caused by information imbalance. The feature matrix is ​​highly structured and reconstructable, making it suitable for subsequent similarity calculations and color comparisons.

[0047] The generation of the evaluation benchmark model is automatically completed by the model training control module. This module can control the modeling process through dynamic monitoring of internal measurement parameters. For example, when the standard deviation of the color distribution drops to a stable threshold and the correlation between texture features remains within an acceptable range, the module determines that the feature extraction process has reached convergence and automatically triggers the model building program. This mechanism can effectively avoid human intervention and model instability, ensuring that the generated model has high representativeness and engineering reusability.

[0048] Step S2: Acquire surface images of the fair-faced concrete component to be evaluated. The image data generation includes: acquiring an image sequence covering the surface of the fair-faced concrete component through an image acquisition device, and performing format conversion, illumination adjustment and noise suppression operations on the image sequence after acquisition to obtain structured image data.

[0049] Image data acquisition of the fair-faced concrete components to be evaluated was performed using a mobile image acquisition device. This device integrates a high-resolution industrial camera, a uniform brightness LED array light source, and a three-dimensional attitude control system to meet the requirements for accurate imaging of complex component surfaces in different engineering scenarios. During image acquisition, the acquisition device performs multi-angle scanning sampling under the guidance of the control program, that is, periodically adjusting the viewing angle and shooting angle on the set path to ensure that the surface texture and color details of the component are completely covered. After each image acquisition, the system automatically records the acquisition time (timestamp) and spatial location information (which can be obtained through IMU or visual odometry) and binds them to the corresponding image frame to form a time-series numbered image sequence, providing the necessary basis for subsequent area positioning and data alignment operations.

[0050] The acquired raw image frames are then input into the image preprocessing module. This module first converts different image formats (such as RAW, JPEG, etc.) into a unified structured intermediate format (such as TIFF or PNG uncompressed format) to ensure data consistency. Secondly, uneven lighting is prone to occur in actual construction sites, so a lighting normalization operation is introduced. Based on Retinex theory, global brightness adjustment is performed on each frame of the image to eliminate reflection interference caused by lighting changes. In addition, for edge blurring or dust contamination that may exist on the surface of concrete components, an edge enhancement algorithm (such as Laplacian enhancement) is used to highlight the construction boundary. At the same time, median filtering and bilateral filtering algorithms are combined to suppress high-frequency noise, improve the overall image clarity and robustness of subsequent processing.

[0051] After preprocessing, the image data is organized into structured data by the data generation module. The structured format includes not only the processed image frame data itself, but also metadata such as the spatial location information of the image, acquisition parameters (such as exposure time and focal length settings), shooting time, and component number. This organization can support subsequent targeted operations such as defect area identification and effective area extraction, ensuring that the analysis is based on an image with consistent temporal and spatial distribution, effectively improving the recognition accuracy and evaluation stability.

[0052] Step S3, performing defect region identification on the image data, generating a mask image, and using the mask image to mask defect regions to extract valid image regions includes: performing pixel-level defect identification operations on the image data to generate a mask image representing defect regions and non-defect regions;

[0053] First, for each frame of the original image data, input images... Perform semantic-level pixel classification processing by calling a trained image segmentation model (such as DeepLabV3+ or U-Net) for each pixel in the image. The corresponding image content is used to predict semantic labels to determine whether it belongs to a defect area;

[0054] This segmentation model is based on a convolutional neural network, integrates a multi-scale receptive field structure, and incorporates contextual information to extract features from the input image at multiple levels. Its classification output is a logistic tensor. , represents the predicted probability that a pixel is a defect;

[0055] The segmentation model adopts an encoder-decoder structure. DeepLabV3+ introduces atrous convolution and spatial pyramid pooling (ASPP) to capture multi-scale contextual features, making it suitable for large-area continuous texture analysis. U-Net, on the other hand, adopts a symmetrical structure and skip connection mechanism, which has good detail restoration capabilities and is particularly suitable for identifying fine-grained defect features on the surface of fair-faced concrete. The model training data comes from fair-faced concrete image samples collected from engineering sites. The image content covers multiple typical defect types such as cracks, honeycombing, pitting, pollution, and discoloration. All training images are pixel-level defect annotations completed by an expert team to construct a high-quality semantic segmentation dataset.

[0056] The training adopts a supervised learning method, and the loss function is class-balanced cross-entropy loss to ensure that the model does not ignore rare defect types when the sample distribution is uneven.

[0057] After training, the model can generate a set of probability values ​​for each pixel in the input image. Classified into the predicted category with the highest probability Finally, all defect classes are uniformly assigned a value of 0 in the mask image, and non-defect classes are assigned a value of 1, generating a binary mask image for masking processing. .

[0058] Based on this probability diagram By setting classification thresholds Perform a binarization operation to obtain the mask image. ,Right now:

[0059]

[0060] in When the probability of a pixel is less than the threshold When the threshold is met, it is judged as a non-defect area (retained); when the threshold is greater than or equal to the threshold, it is judged as a defect area (masked).

[0061] Then, using the generated mask image The original image frame is processed through pixel-by-pixel logical operations. Perform masking operations to extract non-defect regions and generate valid image regions. Its definition is:

[0062]

[0063] in This represents the matrix of pixel values ​​retained after masking. Unmasked pixels retain their original values ​​in the original image, while masked areas are set to zero or filled with background values. Furthermore, to monitor image quality and determine whether to perform image resampling, the system calculates the defect ratio and defect area ratio on the mask image. The definition is as follows:

[0064]

[0065] Where D is the total number of defective pixels. The system calculates the total number of pixels in the image based on the defect area ratio R and a set threshold. Comparison:

[0066]

[0067] in, This is a Boolean logic state indicating whether to initiate image resampling. The system has a preset resampling threshold. Image resampling specifically includes: moving the image acquisition position, optimizing the shooting angle, or adjusting the image resolution settings to obtain new image data with less defect interference, thereby ensuring the accuracy of subsequent evaluation and analysis and the reliability of the data basis.

[0068] when When the percentage is 40%, it indicates that there is too much defect interference in the current image, which may affect the objectivity and accuracy of the assessment. The system will automatically trigger the image re-acquisition module. The re-acquisition module can control the image acquisition device to perform one or a combination of the following operations: change the acquisition angle, move the camera position, adjust the focal length or exposure parameters, enable the high-resolution acquisition mode, etc., to generate new image data with clearer image and less defect interference for subsequent processing.

[0069] In summary, this step, through pixel-level defect identification using a semantic segmentation model and masking extraction using a masking mechanism, not only clearly eliminates image interference factors but also achieves structured and high-quality extraction of the effective evaluation region. This establishes a stable and reliable image data source for subsequent color value and texture feature calculations, enhancing the repeatability and engineering applicability of the system evaluation results.

[0070] Extracting the Lab color value and texture feature to be evaluated from the effective image region includes: performing image preprocessing, image color space conversion and texture feature calculation operations on the image frames in the effective image region to obtain the Lab color value and texture feature to be evaluated.

[0071] After defect masking and effective region extraction are completed, the effective image region enters the image analysis module for further data extraction processing:

[0072] The image analysis module first performs image preprocessing operations, including the following steps:

[0073] Brightness equalization processing: The image brightness distribution is adjusted based on the adaptive histogram equalization (CLAHE) method to reduce the image color difference caused by uneven local illumination;

[0074] Color correction processing: A mapping correction method based on a color reference card is used to correct color shifts and eliminate color deviations caused by changes in lighting conditions during shooting;

[0075] Edge enhancement processing: Use the Laplacian or Sobel operator to enhance the visual features of image edges to highlight the boundary information of the surface structure of fair-faced concrete components.

[0076] After preprocessing, the image data undergoes color space conversion from RGB to CIE Lab color space. This conversion is achieved through a non-linear mapping defined by the CIE standard, resulting in a color expression that better matches human visual perception. Specifically, the L (lightness channel), a (green-red axis), and b (blue-yellow axis) channel values ​​of each pixel are extracted, and these values ​​are statistically calculated to form the Lab color value to be evaluated, serving as the basic parameter for objectively quantifying the color consistency of the concrete surface.

[0077] At the same time, the image analysis module performs texture feature calculations on the image:

[0078] First, the preprocessed color image is converted into a grayscale image using a weighted method. Then, a grayscale co-occurrence matrix (GLCM) is constructed using the grayscale image. This matrix describes the grayscale relationship of image pixels using specific distances (e.g., 1 pixel) and orientations (e.g., 0°, 45°, 90°, 135°).

[0079] Then, based on GLCM, four texture statistics indicators—energy, contrast, entropy, and correlation—are calculated to quantify the texture feature distribution of the image surface. The resulting statistical indicators are then integrated into a texture feature vector to be evaluated.

[0080] Finally, the Lab color values ​​to be evaluated and the texture feature vectors to be evaluated are combined into a complete structured dataset according to a preset unified structural format. This dataset is clearly associated with the valid image region number and the identifier of the fair-faced concrete component to be evaluated. This standardized organization and storage model can effectively ensure the traceability and repeatability of the data, and provide an objective and standardized data input source for subsequent texture similarity calculation and color comparison operations. This improves the accuracy and robustness of the digital evaluation process for fair-faced concrete construction quality, and completely solves the fundamental technical problems of traditional manual evaluation methods, such as strong subjectivity, poor repeatability, and susceptibility to human interference.

[0081] Step S4: Based on the evaluation benchmark model, compare the similarity between the texture features to be evaluated in the effective image region and the benchmark texture features in the sample image, and calculate the color difference between the Lab color values ​​to be evaluated in the effective image region and the benchmark Lab color values ​​in the sample image. The comparison results include:

[0082] The evaluation process involves calling the color description vector and texture description vector from the benchmark model, and calculating them against the Lab color value and texture feature extracted from the effective image region. During texture feature comparison, a preset similarity calculation method is used to match the texture feature to be evaluated with the benchmark texture feature, outputting a texture similarity score. During color value comparison, a standard color difference formula is used to compare the Lab color value to be evaluated with the benchmark Lab color value, outputting a color difference value. Finally, the texture similarity score and color difference value are combined and the comparison result is output.

[0083] During the comparison phase, the corresponding baseline Lab color value and baseline texture feature vector are first retrieved from the feature matrix of the evaluation baseline model based on the component identifier of the component to be evaluated.

[0084] The specific calculation process for texture feature comparison is as follows: Define the texture feature vector to be evaluated as... Define the baseline texture feature vector as The vectors represent four standardized statistical indicators: contrast, entropy, energy, and correlation.

[0085] The difference between the two is calculated using the Euclidean distance method, expressed as:

[0086]

[0087] After obtaining the Euclidean distance of the texture features, it is further converted into a standardized texture similarity score through Gaussian mapping. The calculation formula is as follows:

[0088]

[0089] in, The texture feature vector of the region to be evaluated. The texture feature vector of the reference region, , , , These are the contrast, entropy, energy, and correlation index values ​​of the region to be evaluated. , , , These are the indicator values ​​corresponding to the benchmark region. This represents the Euclidean distance between texture feature vectors. The distance sensitivity adjustment coefficient was obtained through on-site experimental calibration. This represents texture similarity, with a value ranging from 0 to 1. The closer the value is to 1, the closer the texture of the area to be evaluated is to the reference area, reflecting better surface quality.

[0090] The specific steps for comparing and calculating color values ​​are as follows: Define the Lab color value of the image region to be evaluated as a vector. Define the reference area as a vector corresponding to the Lab color value. Calculate the differences for each channel in sequence:

[0091]

[0092] Subsequently, considering the local color difference sensitivity of the concrete surface, a nonlinear local sensing adjustment function is introduced:

[0093]

[0094] Redefine the interaction terms between color channels to reflect the mutual influence of hue shifts between channels a and b:

[0095]

[0096] Then determine the weighted correction terms for the colors of each channel:

[0097]

[0098] The final improved color difference evaluation value formula is:

[0099]

[0100] in, , , These represent the Lab values ​​for the brightness, red-green channel, and yellow-blue channel of the area to be evaluated. , , The Lab value corresponding to the baseline region. , , These are the channel differences between the two regions. , , Let L, a, and b represent the nonlinear local sensing modulation functions, respectively. , , These are the local sensitivity parameters for channels L, a, and b, respectively, obtained through data statistics. This is the color interaction item between channels a and b. To act as a stabilizing factor and prevent the denominator from approaching zero, thus causing computational instability. , , Weighted correction value for channel color difference , , These are the normalized weights for each channel, satisfying... , This is the channel interaction effect moderating coefficient. This is the final color difference evaluation value; the smaller the value, the closer the color of the area being evaluated is to the reference area. After the above calculations are completed, the resulting texture similarity score will be... and color difference evaluation value Unified and integrated into a structured comparison result data format:

[0101]

[0102] in, Identification of the component to be evaluated. Image regions are assigned numbers, and the integrated data is archived by binding the component identifier with the image region number to ensure that the data has traceability, accuracy and consistency;

[0103] The entire comparison process, through the innovative application of structured data integration and quantitative calculation methods, solves the technical problems of strong subjectivity and poor repeatability in the existing evaluation of concrete surface quality, significantly improving the accuracy and objectivity of automated evaluation of fair-faced concrete construction quality, and is suitable for a wide range of engineering practice applications.

[0104] Step S5: Based on the comparison results, generate evaluation output information related to the fair-faced concrete component to be evaluated, including: performing classification judgment operations based on the comparison results, and generating structured evaluation output information according to preset evaluation rules.

[0105] In the evaluation output, texture similarity scores are obtained. Color difference evaluation value Based on this, a structured and standardized evaluation output information generation mechanism was further proposed to effectively guide the actual fair-faced concrete construction and quality control.

[0106] Specifically, this invention designs a set of clear classification judgment and evaluation rules to achieve a precise transformation process from quantitative comparison results to operational and implementable evaluation output information on the construction site;

[0107] The specific implementation process for generating the evaluation output information is as follows:

[0108] First, this invention constructs a preset evaluation rule for texture similarity score and color difference evaluation value. The rule is calibrated and determined based on the quality standards of fair-faced concrete construction and a large amount of historical data from field tests, demonstrating clear objectivity and applicability.

[0109] The texture similarity score The classification threshold can be set to two critical values: 0.90 and 0.75.

[0110] The specific classification rules are defined as follows: when the texture similarity is ≥0.90, it means that the texture quality is very close to the sample and is rated as "excellent"; when the texture similarity score is between 0.75 and 0.90, it means that the texture quality is acceptable and is rated as "good"; when the texture similarity score is below 0.75, it means that the texture quality deviates significantly and is rated as "poor", requiring processing or rework.

[0111] Meanwhile, color difference evaluation value The classification thresholds can be set to two key values: 1.0 and 3.0. Specifically, when the color difference evaluation value is less than or equal to 1.0, it indicates that there is no obvious color difference in vision, which is "excellent"; when the color difference evaluation value is between 1.0 and 3.0, it indicates that there is a slight but acceptable color difference, which is "good"; when the color difference evaluation value is greater than or equal to 3.0, the color difference is obviously unacceptable, which is "poor", and measures need to be taken to correct or rework.

[0112] Secondly, during the classification and judgment operation, the system automatically reads the texture similarity score and color difference evaluation value recorded in the comparison results according to the aforementioned threshold rules, and determines the specific quality category to which the texture and color difference belong. Then, according to a predefined preset mapping table, it automatically calls up the evaluation suggestion items that strictly correspond to the category. These evaluation suggestion items are prepared in clear and concise language beforehand, including but not limited to: "Accepted, no further processing required," "Minor corrections are recommended on site, such as local sanding or cleaning," and "Obviously does not meet the standards, rework or adjustment of the construction plan is necessary," etc., so that on-site technicians can directly understand and effectively implement them. Finally, the specific evaluation suggestion items are bound with the component identifier, texture classification result, and color classification result of the component to be evaluated to form the final evaluation output information in a structured standard format.

[0113] Finally, the specific evaluation suggestions mentioned above are bound to the component identifier, texture classification result and color classification result of the component to be evaluated to form the final evaluation output information in a structured standard format.

[0114] The final structured evaluation output information includes: ① the component identifier of the component to be evaluated (uniquely identifying the fair-faced concrete component to be evaluated), ② the classification result of texture similarity (excellent, good, or poor), ③ the classification result of color difference (excellent, good, or poor), and ④ specific evaluation suggestions that strictly correspond to the above classification results (such as acceptance, on-site correction, or rework), to ensure that on-site personnel can intuitively and clearly obtain detailed, targeted, and implementable quality control instructions.

[0115] Example 2

[0116] A digital evaluation system based on fair-faced concrete mockup, characterized in that it includes:

[0117] Evaluation benchmark model construction module: Collect surface images of fair-faced concrete mockup samples, extract benchmark Lab color values ​​and benchmark texture features from the collected sample images, and construct the evaluation benchmark model;

[0118] Image acquisition module for evaluation: Acquires surface images of the fair-faced concrete component to be evaluated and generates image data;

[0119] Image preprocessing and effective region extraction module: performs defect region identification on image data, generates a mask image, and uses the mask image to mask defect regions in order to extract effective image regions. It then extracts the Lab color value and texture features to be evaluated from the effective image regions.

[0120] Texture and color comparison calculation module: Based on the evaluation benchmark model, the similarity of the texture features to be evaluated in the effective image area is compared with the benchmark texture features of the sample image, and the color difference of the Lab color value to be evaluated in the effective image area is calculated with the benchmark Lab color value of the sample image to generate the comparison result.

[0121] Evaluation output generation module: Based on the comparison results, it generates evaluation output information related to the fair-faced concrete component to be evaluated.

[0122] The technical features of this invention not described can be implemented by or using existing technology, and will not be repeated here. Of course, the above description is not a limitation of this invention, and this invention is not limited to the examples above. Any changes, modifications, additions or substitutions made by those skilled in the art within the scope of this invention should also be within the protection scope of this invention.

Claims

1. A digital evaluation method based on fair-faced concrete mockups, characterized in that, include: Surface images of fair-faced concrete mockup samples were collected, and baseline Lab color values ​​and baseline texture features were extracted from the collected sample images to construct an evaluation baseline model. The construction of the evaluation benchmark model includes: constructing corresponding color feature datasets and texture feature datasets based on the benchmark Lab color values ​​and benchmark texture features extracted from the sample image, respectively; Statistical processing and standardization are performed on the color feature dataset and texture feature dataset respectively to obtain color description vectors and texture description vectors. The color description vectors and texture description vectors are combined according to a fixed format to construct a feature matrix for evaluating the benchmark model. The construction of the benchmark model is completed when the preset trigger conditions are met. Cluster analysis is performed on the Lab value of each image pixel in the color feature dataset. The K-means clustering algorithm is used to extract the main color distribution regions to form color feature centers with aggregation. The color mean, color standard deviation and main color distribution density statistical parameters are calculated for the clustered samples. The color mean, color standard deviation and main color distribution density statistical parameters are combined to construct a color description vector. For the texture feature dataset, texture statistical indicators extracted from the gray-level co-occurrence matrix are used as the original feature dimensions. The texture statistical indicators include the contrast, entropy, energy and correlation of the image. A variance screening strategy is introduced to remove feature components with low stability and weak correlation. The Z-score standardization method is used to unify the numerical scale of different texture statistical indicators to form a texture description vector. The color description vector and the texture description vector are concatenated according to a fixed dimension rule to construct a feature matrix, which is then used as the core parameter structure of the evaluation benchmark model. The preset triggering conditions include the standard deviation of the color distribution decreasing to a stable threshold and the correlation between texture features remaining within an acceptable range. Collect surface images of the fair-faced concrete component to be evaluated and generate image data; Defect region identification is performed on the image data, a mask map is generated, and the defect region is masked using the mask map to extract the effective image region. The Lab color value and texture feature to be evaluated are extracted from the effective image region. Using the generated mask image The original image frame is processed through pixel-by-pixel logical operations. Perform masking operations to extract non-defect regions and generate valid image regions. Its definition is: To monitor image quality and determine whether to perform image resampling, the system calculates the defect ratio and defect area ratio of the mask image. The definition is as follows: Where D is the total number of defective pixels. The total number of pixels in the image. The defect area ratio is calculated by the system based on the defect area ratio R and a set threshold. Comparison: in, This is a Boolean logic state indicating whether to initiate image resampling. The system has a preset resampling threshold. Image resampling specifically includes: moving the image acquisition position, optimizing the shooting angle, or adjusting the image resolution settings to obtain new image data with less defect interference, thereby ensuring the accuracy of subsequent evaluation and analysis and the reliability of the data basis. The color description vector and texture description vector in the evaluation benchmark model are called and calculated against the Lab color value and texture feature to be evaluated extracted from the effective image region, respectively. When performing texture feature comparison, the texture feature vector to be evaluated is defined as... Define the baseline texture feature vector as The vectors represent four standardized statistical indicators: contrast, entropy, energy, and correlation, respectively. The difference between the two is calculated using the Euclidean distance method, expressed as: After obtaining the Euclidean distance of the texture features, it is further converted into a standardized texture similarity score through Gaussian mapping. The calculation formula is as follows: in, The texture feature vector of the region to be evaluated. The texture feature vector of the reference region, , , , These are the contrast, entropy, energy, and correlation index values ​​of the region to be evaluated. , , , These are the indicator values ​​corresponding to the benchmark region. This represents the Euclidean distance between texture feature vectors. This is the distance sensitivity adjustment coefficient. Indicates texture similarity; When performing color value comparison, the Lab color value of the image region to be evaluated is defined as a vector. Define the reference area as a vector corresponding to the Lab color value. Calculate the differences for each channel in sequence: Subsequently, considering the local color difference sensitivity of the concrete surface, a nonlinear local sensing adjustment function is introduced: Redefine the interaction terms between color channels to reflect the mutual influence of hue shifts between channels a and b: Then determine the weighted correction terms for the colors of each channel: The final improved color difference evaluation value formula is: in, , , These represent the Lab values ​​for the brightness, red-green channel, and yellow-blue channel of the area to be evaluated. , , The Lab value corresponding to the baseline region. , , These are the channel differences between the two regions. , , Let L, a, and b represent the nonlinear local sensing modulation functions, respectively. , , These are the local sensitivity parameters for channels L, a, and b, respectively. This is the color interaction item between channels a and b. As a stabilizing factor, , , Weighted correction value for channel color difference , , These are the normalized weights for each channel, satisfying... , This is the channel interaction effect moderating coefficient. This is the final color difference evaluation value; The obtained texture similarity score and color difference evaluation value Unified and integrated into a structured comparison result data format: in, Identification of the component to be evaluated. Image regions are assigned numbers, and the integrated data is archived by binding the component identifier with the image region number; Based on the comparison results, evaluation output information related to the fair-faced concrete component to be evaluated is generated.

2. The digital evaluation method based on fair-faced concrete mockup according to claim 1, characterized in that, Surface images of fair-faced concrete mockup samples were acquired, and baseline Lab color values ​​and baseline texture features were extracted from the acquired sample images, including: The surface image of the fair-faced concrete mockup sample is acquired by an image acquisition device. The surface image is preprocessed and then converted to the CIE Lab color space to extract the baseline Lab color value. At the same time, the baseline texture feature of the image is extracted by the gray-level co-occurrence matrix. Finally, the baseline Lab color value and baseline texture feature are obtained for constructing the evaluation baseline model.

3. The digital evaluation method based on fair-faced concrete mockup according to claim 2, characterized in that, Surface images of the fair-faced concrete component to be evaluated were acquired, and the generated image data included: Image sequences covering the surface of fair-faced concrete components are acquired using an image acquisition device. After acquisition, the image sequences are processed for format conversion, illumination adjustment, and noise suppression to obtain structured image data.

4. The digital evaluation method based on fair-faced concrete mockup according to claim 3, characterized in that, Defect region identification is performed on image data, a mask image is generated, and the defect region is masked using the mask image to extract the valid image region, including: Perform pixel-level defect identification on the image data to generate a mask image representing defective and non-defective regions; Based on the information marked by the mask image, the corresponding defective areas in the image data are masked, while the pixel content corresponding to the non-defective areas in the image data is retained, and the effective image area is extracted.

5. The digital evaluation method based on fair-faced concrete mockup according to claim 4, characterized in that, Extracting the Lab color values ​​and texture features to be evaluated from the effective image region includes: Image preprocessing, color space conversion, and texture feature calculation are performed on the image frames in the effective image region to obtain the Lab color value and texture feature to be evaluated.

6. The digital evaluation method based on fair-faced concrete mockup according to claim 5, characterized in that, Based on the comparison results, the evaluation output information related to the fair-faced concrete component to be evaluated includes: Based on the comparison results, a classification judgment operation is performed, and structured evaluation output information is generated according to the preset evaluation rules.

7. A digital evaluation system based on fair-faced concrete mockup using the method described in any one of claims 1-6, characterized in that, include: Evaluation benchmark model construction module: Collect surface images of fair-faced concrete mockup samples, extract benchmark Lab color values ​​and benchmark texture features from the collected sample images, and construct the evaluation benchmark model; Image acquisition module for evaluation: Acquires surface images of the fair-faced concrete component to be evaluated and generates image data; Image preprocessing and effective region extraction module: performs defect region identification on image data, generates a mask image, and uses the mask image to mask defect regions in order to extract effective image regions. It then extracts the Lab color value and texture features to be evaluated from the effective image regions. Texture and color comparison calculation module: Based on the evaluation benchmark model, the similarity of the texture features to be evaluated in the effective image area is compared with the benchmark texture features of the sample image, and the color difference of the Lab color value to be evaluated in the effective image area is calculated with the benchmark Lab color value of the sample image to generate the comparison result. Evaluation output generation module: Based on the comparison results, it generates evaluation output information related to the fair-faced concrete component to be evaluated.

Images