Image analysis method and system for concrete apparent quality defect detection

Abstract

The invention discloses an image analysis method and system for concrete apparent quality defect detection, and relates to the technical field of concrete apparent quality defect detection.The method comprises the steps that a camera device is used for collecting concrete surface images in a specified distance interval, and the optical axis of a camera lens is controlled to be perpendicular to the concrete surface; carrying out image preprocessing on the collected concrete surface image, and carrying out defect type image marking; carrying out key feature extraction on the preprocessed concrete surface image by adopting a convolutional neural network to obtain key feature information; constructing a multi-algorithm comparison verification framework, performing cross validation in combination with the key feature information to obtain defect detection results and evaluation results of a plurality of defect detection models, optimizing model parameters in combination with a confusion matrix, and determining a final concrete surface detection model; obtaining an apparent quality defect detection result based on the defect detection result and a preset quality defect grading threshold value; the efficiency of concrete apparent defect detection is improved.

Description

Technical Field

[0001] The present application relates to the technical field of concrete surface quality defects, and in particular to an image analysis method and system for detecting concrete surface quality defects. Background Art

[0002] External defects in reinforced concrete are very common in engineering, and many reinforced concrete structures often fail due to these defects. The main impacts of these defects on the structure include reduced concrete strength, leakage, corrosion of steel bars, and overall structural durability.

[0003] Currently, the assessment of concrete surface quality defects in China primarily relies on manual inspection and scoring. While simple, this assessment method is labor-intensive and inefficient. It is also prone to unnecessary errors due to subjectivity, ambiguity, and randomness, introducing significant uncertainty into practical applications. Therefore, traditional concrete surface quality defect detection suffers from high labor intensity and low efficiency due to subjective judgments that are prone to errors, leaving room for improvement. Summary of the Invention

[0004] In order to solve the problems of high labor intensity and low efficiency of subjective judgment that are prone to errors in traditional concrete surface quality defect detection, the present application provides an image analysis method and system for concrete surface quality defect detection.

[0005] In the first aspect, the invention objectives of this application are achieved by adopting the following technical solutions: Image analysis methods for detecting concrete surface quality defects include: Use a camera to capture images of the concrete surface at a specified distance, and control the optical axis of the camera lens to be perpendicular to the concrete surface; perform image preprocessing on the captured concrete surface images and mark the defect types; Convolutional neural network is used to extract key features from pre-processed concrete surface images to obtain key feature information; Constructing a multi-algorithm comparison and verification framework based on multiple defect detection models, performing cross-validation in combination with the key feature information, and obtaining defect detection results and evaluation results of the multiple defect detection models; Based on the evaluation results and the confusion matrix optimization model parameters, a final concrete surface detection model is determined; in the concrete surface detection model, an apparent quality defect detection result is obtained based on the defect detection result and a preset quality defect classification threshold.

[0006] By adopting the above technical solution, an image analysis method with high efficiency and automated detection of concrete surface defects is provided, which is conducive to improving the detection efficiency and detection accuracy of concrete surface defects. Specifically, this application combines image preprocessing and defect type image labeling by limiting the distance between the camera and the concrete surface (e.g., 150-200 cm) and the vertical angle of the optical axis (e.g., ≤4°) to perform preliminary detection and localization of concrete surface quality defects, which is more efficient and accurate than manual visual inspection. A convolutional neural network is also used to automatically extract key features of the concrete surface, replacing traditional manual standard features. The performance of each defect detection model in terms of defect localization accuracy, recall rate, and real-time performance is evaluated through a multi-algorithm comparison and verification framework for multiple defect detection models. The optimal model is then selected based on the evaluation results. The model parameters are optimized by combining confusion matrices (e.g., transfer learning fine-tuning, gradient boosting decision trees, etc.) to enhance the robustness and generalization ability of the concrete surface detection model. This helps maintain the detection accuracy of the concrete surface detection model in complex scenarios such as strong light and shadow interference, and supports rapid adaptation to different engineering scenarios (e.g., bridges, tunnels, and factory buildings).

[0007] In a preferred embodiment of the present application, the method further includes: Acquire a concrete surface area, and divide the concrete surface area into a first detection area and a second detection area; Obtaining, based on the first detection area and the second detection area, apparent quality defect factors related to apparent quality defects of concrete, wherein the apparent quality defect factors include a first apparent quality defect factor and a second apparent quality defect factor; obtaining, based on the apparent quality defect factors, apparent defect influencing factors indicating a quality defect grade, wherein the apparent defect influencing factors include a first apparent defect influencing factor and a second apparent defect influencing factor; dividing the first inspection area and the second inspection area according to the apparent defect influencing factor; A division result is obtained, and an apparent defect quality detection parameter is obtained according to the division result.

[0008] By adopting the above technical solution, based on the first detection area (such as natural weathering) and the second detection area (such as artificial pouring), apparent quality defect factors related to apparent quality defect detection are obtained respectively. For example, the first factor includes weathering degree, crack distribution, and color unevenness; the second factor includes material type, construction defects, porosity, and edge sharpness. By distinguishing between natural weathering and artificial pouring areas, the interference of apparent quality defects of different causes can be avoided. The first factor is used to reflect material aging and environmental effects, and the second factor is used to characterize construction technology and material properties. Combining the two can comprehensively cover the causes of concrete apparent quality defects; through influencing factor calculation and regional redivision, apparent defect quality detection parameters (such as SA value, SB value, grading evaluation, three-dimensional model) are obtained to achieve quantitative evaluation and reduce human subjective judgment errors.

[0009] In a preferred example of the present application, the step of obtaining a concrete surface area and dividing the concrete surface area to obtain a first detection area and a second detection area includes: According to the concrete surface area, obtaining a naturally weathered surface area and an artificially poured surface area; Obtaining a defect characteristic area according to the naturally weathered surface area and the artificially poured surface area; A first detection area and a second detection area are acquired according to the defect characteristic area and the concrete surface area.

[0010] By adopting the above technical solution, combined with the hierarchical area division of natural weathering area-artificial casting area-defect feature area-final detection area, the defect feature area (such as honeycomb surface and hollowing) can be accurately identified and located for defects, effectively avoiding the interference of natural weathering; further, after dividing the surface into two types of areas, naturally weathered and artificially cast, different image processing parameters can be used respectively (for example, the weathered area needs to enhance the crack contrast, and the artificial area needs to suppress light reflection interference) to optimize the calculation efficiency.

[0011] In a preferred example of the present application, obtaining an apparent defect impact factor indicating a quality defect grade based on the apparent quality defect factor includes: Obtaining preset defect quality association parameters, and screening the first apparent quality defect factor based on the preset defect quality association parameters to obtain a first quality defect feature, wherein the first quality defect feature includes a first texture complexity feature, a first crack distribution feature, and a first color non-uniformity feature; Obtaining a first apparent defect influencing factor by calculating an image gray level co-occurrence matrix according to the first quality defect feature; Screening the second apparent quality defect factor according to the preset defect quality association parameter to obtain a second quality defect feature, wherein the second quality defect feature includes a second surface morphology feature, a second porosity feature, a second edge sharpness feature, and a second light reflection feature; According to the second quality defect feature, a second apparent defect influencing factor is obtained through an edge detection algorithm and color histogram analysis.

[0012] By adopting the above technical solution, the grayscale co-occurrence matrix is ​​suitable for quantifying crack distribution (spatial texture complexity) and color unevenness (grayscale change). The grayscale co-occurrence matrix is ​​insensitive to illumination changes (suitable for outdoor weathered areas), for example, reflecting crack density through contrast characteristics; this application analyzes the second apparent defect influencing factor through edge detection combined with color histogram, where edge sharpness can characterize the flatness of the construction process (such as template splicing marks), and color histogram analysis can detect color anomalies caused by fading or pollution (common defects in artificial pouring); in order to further improve the detection accuracy of concrete surface quality defect detection, this application separates intrinsic influencing factors and extrinsic influencing factors for evaluation, where the separation of intrinsic factors (inherent properties of materials) and extrinsic factors (environmental / human influences) supports differentiated maintenance strategies. For example, areas with high intrinsic factors require material reinforcement, and areas with high extrinsic factors require cleaning or repainting.

[0013] In a preferred example of the present application, the dividing of the first detection area and the second detection area according to the apparent defect influencing factor includes: Obtaining first apparent quality inspection points through K-means clustering according to the first apparent defect influencing factor, and dividing the first inspection area according to the first apparent quality inspection points; According to the second apparent defect influencing factor, obtaining a second apparent quality detection point by using a support vector machine classifier, and dividing the detection area according to the second apparent quality detection point; and / or, The obtaining of apparent defect quality detection parameters includes: Calculating the surface roughness SA value and SB value according to the division result of the first detection area; According to the division result of the second detection area, an apparent quality grading evaluation result and a three-dimensional morphology reconstruction model are generated.

[0014] By adopting the above technical solution, the support vector machine classifier is a support vector machine (SVM) classifier; specifically, K-means clustering is suitable for locating apparent quality inspection points in naturally weathered areas, and can quickly identify areas with similar textures (such as weathering crack clusters); the SVM classifier is suitable for locating defect inspection points in artificially poured areas, and can accurately distinguish construction defects (such as honeycomb surfaces and air holes): because K-means processes unsupervised features and SVM processes supervised classification, the combination of the two is adaptable to complex scenarios (such as the intersection of mixed weathering and artificial defects); and the SA value (international standard ISO 4287) provides quantitative apparent quality defect evaluation, and the SB value (such as the arithmetic mean deviation) supplements the micro-geometric features.

[0015] In a preferred example of the present application, obtaining apparent quality defect factors related to apparent quality defects of concrete based on the first detection area and the second detection area includes: Acquire a first apparent quality defect factor according to the first detection area, wherein the first apparent quality defect factor includes a weathering degree characteristic, a crack distribution characteristic, and a color unevenness characteristic; According to the second detection area, a second apparent quality defect factor is obtained, wherein the second apparent quality defect factor includes a material type feature, a construction defect feature, a porosity feature, and an edge sharpness feature.

[0016] By adopting the above technical solutions, the naturally weathered areas focus on environmental erosion effects (such as crack expansion rate and color fading); the artificial casting areas focus on process defects (such as excessive porosity leading to water seepage, and poor edge sharpness reflecting the quality of formwork splicing). Through targeted testing of different areas, the accuracy of analysis of apparent quality defect factors can be improved.

[0017] In a preferred example of the present application, the first apparent defect influencing factor includes a first apparent defect intrinsic influencing factor and a first apparent defect extrinsic influencing factor. Obtaining the first apparent defect influencing factor according to the first quality defect characteristic includes: According to the weathering degree characteristics and crack distribution characteristics, obtaining the first apparent defect intrinsic influencing factor through fractal dimension calculation and morphological analysis; Obtaining a first apparent defect external influencing factor through color histogram contrast analysis according to the color non-uniformity characteristics; and / or, The second apparent defect influencing factor includes a second apparent defect intrinsic influencing factor and a second extrinsic influencing factor. The obtaining of the second apparent defect influencing factor according to the second quality defect characteristic specifically includes: According to the material type characteristics, porosity characteristics and edge sharpness characteristics, the intrinsic influencing factor of the second apparent defect is obtained through spectral reflectance and three-dimensional point cloud reconstruction; According to the construction defect characteristics, a second apparent defect external influencing factor is obtained through morphological operation and defect area segmentation.

[0018] By adopting the above technical solution, based on the first apparent defect influencing factor (natural weathering), the fractal dimension calculation is used to analyze the intrinsic influencing factors to quantify the complexity of the crack texture, and morphological analysis is used to analyze the extrinsic influencing factors, such as detecting the sharpening of the crack edge; based on the second apparent defect influencing factor (artificial pouring), the spectral refractive index is used to analyze the intrinsic influencing factors to identify the material type, and three-dimensional point cloud reconstruction is used to quantify the intrinsic influencing factors, such as porosity and surface undulation; morphological operations combined with defect segmentation are conducive to analyzing construction defects (such as bubbles and cold joints) and defect location; further, the fractal dimension is robust to noise and suitable for the non-uniform texture of the weathered area; three-dimensional point cloud reconstruction combined with spectral reflectivity can penetrate surface stains and extract the intrinsic characteristics of the material.

[0019] In the second aspect, the invention objective of this application is achieved by adopting the following technical solutions: An image analysis system for detecting apparent quality defects of concrete, configured to execute the image analysis method for detecting apparent quality defects of concrete as described above; the system comprises: An image acquisition module is used to use a camera to capture images of the concrete surface at a specified distance interval and control the optical axis of the camera lens to be perpendicular to the concrete surface; An image preprocessing and marking module is used to preprocess the collected concrete surface images and mark the defect types; A key feature extraction module is used to extract key features from the pre-processed concrete surface image using a convolutional neural network to obtain key feature information; A multi-algorithm comparison and verification framework module is used to construct a multi-algorithm comparison and verification framework based on multiple defect detection models, and perform cross-validation in combination with the key feature information to obtain defect detection results and evaluation results of the multiple defect detection models; a model optimization and selection module is used to optimize model parameters based on the evaluation results and the confusion matrix to determine the final concrete surface detection model; in the concrete surface detection model, the apparent quality defect detection results are obtained based on the defect detection results and the preset quality defect classification threshold.

[0020] In a third aspect, the invention objective of this application is achieved by adopting the following technical solutions: A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the above-mentioned image analysis method for detecting concrete surface quality defects.

[0021] Fourthly, the invention objectives of this application are achieved by adopting the following technical solutions: A computer program product includes a computer program / instruction, which, when executed by a processor, implements the steps of the image analysis method for detecting apparent quality defects of concrete.

[0022] In summary, this application includes at least one of the following beneficial technical effects: 1. Precisely controlling the camera equipment to capture images of the concrete surface ensures image quality, laying the foundation for subsequent analysis. Image preprocessing and defect marking improve the pertinence and accuracy of detection. Using a convolutional neural network to extract key features enhances the ability to understand complex surface characteristics. A multi-algorithm comparison and verification framework is constructed, and cross-validation is performed based on key features to ensure the robustness and generalization of the model. 2. Obtaining specific apparent quality defect factors (such as texture complexity, crack distribution, etc.) for different inspection areas helps to more accurately assess the actual condition of the concrete surface; calculating the corresponding apparent quality defect influencing factors based on the obtained apparent quality defect factors can quantify the specific impact of these factors on the apparent quality defects. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a flow chart of an image analysis method for detecting concrete surface quality defects in one embodiment of the present application; Figure 2 This is a flowchart after step S1 in the image analysis method for detecting apparent quality defects of concrete in one embodiment of the present application. DETAILED DESCRIPTION

[0024] The present application is further described in detail below with reference to the accompanying drawings.

[0025] In one embodiment, if Figure 1 As shown, the present application discloses an image analysis method for detecting apparent quality defects of concrete, which specifically includes the following steps: S1: Use a camera to collect concrete surface images at a specified distance interval, and control the camera lens optical axis to be perpendicular to the concrete surface; perform image preprocessing on the collected concrete surface images and mark the defect type images.

[0026] In this embodiment, the specified distance interval is based on the concrete surface material and lighting conditions, setting the distance between the camera and the surface (e.g., 150-200 cm) and the vertical angle of the optical axis (e.g., ≤4°); image preprocessing includes Gaussian filtering for denoising, histogram equalization for contrast enhancement, and morphological operations (closing operations to fill holes); defect type marking refers to the preliminary marking of cracks, holes, and color difference areas based on traditional threshold segmentation (e.g., Otsu algorithm) or lightweight CNN (e.g., MobileNet).

[0027] S2: Use convolutional neural network to extract key features of preprocessed concrete surface images and obtain key feature information.

[0028] In this embodiment, the convolutional neural network adopts the ResNet-50 pre-training model, fine-tunes the last fully connected layer, and outputs multi-scale features (texture, shape, color); the key features include crack texture complexity (feature extraction based on LBP histogram), porosity (obtained by calculating and statistically analyzing the area of ​​the connected domain), and color distribution (extracted by generating an HSV spatial histogram).

[0029] S3: Build a multi-algorithm comparison and verification framework based on multiple defect detection models, perform cross-validation based on key feature information, and obtain defect detection results and evaluation results of multiple defect detection models.

[0030] In this embodiment, the defect detection model integrates YOLOv5s, Mask R-CNN, and EfficientDet-D1. By fusing the defect detection results of multiple models, it detects the overlapping areas of defects. YOLOv5s is a lightweight real-time detection that outputs bounding boxes and confidence levels; Mask R-CNN is based on instance segmentation (ResNeXt-101 backbone) and outputs pixel-level defect masks. EfficientDet-D1 is used to balance accuracy and speed (input size 512×512) and supports multi-scale prediction.

[0031] Specifically, a weighted fusion algorithm is used, such as YOLOv5s with a weight of 0.4, Mask R-CNN with a weight of 0.3, and EfficientDet with a weight of 0.3, to output the final defect detection results. The cross-validation process includes k-fold splitting and evaluation indicator verification, where k in the k-fold split is 5, and each fold accounts for 80% of the training set and 20% of the validation set. The evaluation indicators include F1-score and mAP@0.5. F1-score is used to integrate the model's precision and recall rate ((target ≥ 0.85), and the evaluation indicator mAP@0.5 is used to measure the mean average precision (target ≥ 0.78). Evaluation results are obtained after cross-validation based on the evaluation indicators.

[0032] In actual application development, the multi-algorithm comparison verification box of multiple defect detection models achieved F1-score = 0.89 and mAP@0.5 = 0.82, which is a 12% improvement over the single model.

[0033] S4: Based on the evaluation results and the confusion matrix, the model parameters are optimized to determine the final concrete surface detection model; in the concrete surface detection model, the apparent quality defect detection results are obtained based on the defect detection results and the preset quality defect classification threshold.

[0034] In this embodiment, the confusion matrix optimization includes error analysis and sample weight parameter adjustment of the Focal Loss parameter, where error analysis refers to the statistical model misdetecting samples (such as misjudging stains as cracks), and targeted adjustment of the error loss. The Focal Loss parameter α = 0.25 (suppressing the weight of easily classified samples) and γ = 2 (focusing on difficult-to-classify samples); the apparent quality defect detection results include SA value, SB value (root mean square roughness) and a three-dimensional morphology heat map containing highlighted defect areas, with highlighted content such as red = crack, yellow = hole, etc.

[0035] Specifically, the preset quality defect classification thresholds include high apparent defects, medium apparent defects and low apparent defects; SA value calculation refers to the calculation based on ISO 4287 standard, (n is the number of sampling points, i is the sampling point index; z i is the profile height; is the average height of the contour of n sampling points); when setting the quality defect classification threshold, SA ≥ 2.5 μm can be determined as a high apparent defect (AUC = 0.91) based on ROC curve analysis (Receiver Operating Characteristic Curve).

[0036] For example, during the inspection of concrete guardrails on a highway bridge, drones collected images (at a distance of 1 meter, with the optical axis vertical), marked defects after preprocessing, ResNet-50 extracted features, and YOLOv5s detected crack areas (confidence level 0.87). After multi-model fusion, a SA value distribution map was generated (SA = 2.8μm, which was determined to be a high apparent defect).

[0037] In one embodiment, in actual applications, such as in a large-scale water conservancy project, the surface of the concrete dam has both naturally weathered areas (cracks and fading caused by long-term exposure) and artificially poured areas (honeycombed surfaces and pores left over from construction); it is necessary to quickly detect the dam surface, distinguish between natural erosion and construction defects, and generate a regional quality defect detection report. Therefore, in order to further improve the efficiency of concrete surface quality defect detection, a regional processing strategy is adopted based on the differentiated characteristics of natural weathering and artificial defects to avoid misjudgment by a single model (such as misjudging weathering cracks as construction pores).

[0038] like Figure 2 As shown, after step S1, the image analysis method for detecting concrete surface quality defects further includes: S10: acquiring a concrete surface area, dividing the concrete surface area, and obtaining a first detection area and a second detection area.

[0039] In this embodiment, the concrete surface area is the complete image area after pre-processing in step S1; the first detection area refers to the naturally weathered area, such as cracks and fading caused by long-term exposure; the second detection area is the artificial pouring area, such as the honeycombed surface and pores left over from construction.

[0040] Specifically, step S10 includes: S101: According to the concrete surface area, a naturally weathered surface area and an artificially poured surface area are obtained.

[0041] In this embodiment, the naturally weathered surface area refers to the degraded area of ​​the concrete surface caused by long-term exposure to the environment (such as rain erosion, ultraviolet aging, freeze-thaw cycles), and its characteristics include fading, network cracks, peeling, uneven color, etc.; the artificially cast surface area refers to the defective area formed during the construction process, such as honeycombed surface, cold joints, air holes, formwork splicing marks, etc., which are usually accompanied by sharp edges, uniform color or local protrusions.

[0042] Specifically, in the color feature extraction stage: first perform color space conversion, that is, convert the image from RGB to HSV color space, separate brightness (V channel), saturation (S channel) and hue (H channel); low saturation (S≤0.3) areas (i.e. faded areas, such as pigment loss caused by long-term wind and sun exposure) and low brightness (V≤0.4) areas (i.e. stains or pollutant covered areas) are judged as naturally weathered areas; high saturation (S≥0.6, referring to high pigment concentration in newly poured concrete) areas and high brightness areas (V≥0.6, referring to unaffected areas) are judged as naturally weathered areas. Clean surfaces affected by environmental erosion are identified as artificially poured areas). Morphological operations (closing and opening) are then performed. Closing involves using a 3×3 rectangular kernel to fill tiny holes in naturally weathered areas (such as residual stains in cracks). Opening involves using a 5×5 rectangular kernel to remove isolated noise points (such as reflective spots or small particles) in artificially poured areas. A watershed algorithm is then used to segment overlapping areas based on gradient amplitude (Sobel operator) to generate a binary mask to mark naturally weathered and artificially poured areas. Morphological filtering can eliminate the effects of uneven illumination and improve feature stability.

[0043] S102: Obtain defect characteristic areas based on the naturally weathered surface area and the artificially poured surface area.

[0044] In this embodiment, the defect feature area refers to a local area with significant geometric or texture anomalies in a naturally weathered or artificially cast area, such as cracks, holes, edge mutations, etc.

[0045] Specifically, for the extraction of natural weathering defects, Canny edge detection and morphological gradients were used for crack detection. The Canny edge detection threshold was 50, and the Gaussian filter σ was 1.5 to locate linear cracks. Morphological gradients were used to enhance crack edges, and area threshold filtering (≥50 pixels) was used to remove artifacts. For defect extraction in artificial building areas, honeycomb surface detection and edge sharpness analysis were used. Honeycomb surface detection used the Laplacian operator to calculate gradient amplitude and locate uneven surface areas. A watershed algorithm was used to segment dense defects (such as honeycomb pores). Morphological closing (3×3 rectangular kernel) was then used to fuse the honeycomb surface detection results with the edge sharpness analysis results to connect discrete defect regions and form a continuous defect mask.

[0046] S103: Acquire a first detection area and a second detection area according to the defect characteristic area and the concrete surface area.

[0047] In this embodiment, the degree of environmental erosion needs to be assessed in the first detection area, and the construction quality needs to be repaired in the second detection area. The natural weathering defect mask is superimposed on the original surface area based on the mask logic operation to generate the first detection area. The artificial casting defect area is expanded using morphological dilation (3×3 rectangular kernel) to ensure boundary coverage to generate the second detection area. The first detection area and the second detection area are distinguished and marked based on different marking information.

[0048] S20: Acquire apparent quality defect factors related to apparent quality defects of concrete according to the first detection area and the second detection area, wherein the apparent quality defect factors include a first apparent quality defect factor and a second apparent quality defect factor.

[0049] In this embodiment, the first apparent quality defect factor refers to the key characteristics of the naturally weathered area (weathering degree, crack distribution, color unevenness), and the second apparent quality defect factor refers to the key characteristics of the artificially cast area (material type, construction defects, porosity, edge sharpness).

[0050] Specifically, step S20 includes: S201: Acquire a first apparent quality defect factor according to a first detection area, wherein the first apparent quality defect factor includes a weathering degree characteristic, a crack distribution characteristic, and a color non-uniformity characteristic.

[0051] In this embodiment, the weathering degree characteristic refers to the degree of surface degradation such as fading, peeling, and holes caused by environmental erosion; the crack distribution characteristic refers to the density, length, width, and spatial distribution pattern of the cracks; and the color unevenness characteristic refers to the uneven color distribution (such as color difference and mottled) caused by fading or pollution.

[0052] Specifically, the treated concrete surface image was converted to the LAB color space, and the standard deviation of the lightness channel (L) was calculated (σ ≥ 15 was considered high weathering). Morphological operations were then performed, and the watershed algorithm was used to segment large faded areas, generate weathering degree masks, and extract weathering degree features. Canny edge detection (threshold = 50, Gaussian filter σ = 1.5) was used to locate cracks. The crack edges were enhanced by combining morphological gradients, and the crack density (crack length per unit area ≥ 2 mm / m) was calculated. 2 For color inhomogeneity feature extraction, color histogram analysis and local contrast enhancement are used, and the histogram kurtosis of the HSV-S channel is counted (for example, Kurtosis ≥ 3 is determined to be high inhomogeneity); for local contrast enhancement, CLAHE (contrast-limited adaptive histogram equalization) is used to highlight the color difference area. The LAB color space of this application has a tolerance of ±40% brightness fluctuation to illumination changes, and has good feature extraction accuracy and efficiency.

[0053] S202: Acquire a second apparent quality defect factor according to the second detection area, wherein the second apparent quality defect factor includes a material type feature, a construction defect feature, a porosity feature, and an edge sharpness feature.

[0054] In this embodiment, the material type feature refers to the difference in concrete composition (such as aggregate type and additive ratio); the construction defect feature refers to process defects such as honeycomb surface, cold joints, and air holes; the porosity feature refers to the volume ratio of tiny holes on the surface; and the edge sharpness feature refers to the clarity of the defect edge (such as template splicing marks).

[0055] Specifically, the material type characteristics are obtained through spectral reflectance analysis and color matching algorithm. That is, a portable spectrometer is used to obtain RGB-NIR data, and the PLS regression model is used to match the concrete composition (such as Ca(OH)2 content ≥8% is high-alkalinity concrete), while the color matching algorithm is based on the comparison of RGB mean with the preset concrete color card (ΔE<5 is judged to be the same batch of materials). Construction defect feature extraction uses the Laplacian operator to calculate gradient amplitude and locate uneven areas. The watershed algorithm is then used to segment dense defects (such as honeycomb pores with a pore size of 1-3 mm) to detect honeycombed surface structures on the slow-set surface. Morphological closing operations (5×5 rectangular kernels) are used to connect linear defects, and cold joints are identified if their length is ≥10 mm. Porosity feature extraction uses 3D point cloud reconstruction. A Kinect v2 depth camera is used to acquire surface point clouds, and the PCL library is used to calculate porosity (formula: porosity = total area volume / pore volume). Edge sharpness feature extraction uses the mean gradient amplitude to calculate the mean gradient amplitude. A mean gradient amplitude ≥15 is considered a sharp edge. The Laplacian operator is then used to quantify the strength of edge mutations, with λ ≥0.5 identifying high sharpness.

[0056] S30: Obtaining an apparent defect influence factor indicating a quality defect grade according to the apparent quality defect factor, wherein the apparent defect influence factor includes a first apparent defect influence factor and a second apparent defect influence factor.

[0057] In this embodiment, the first apparent defect influencing factor refers to the comprehensive apparent quality defect quantitative index of the naturally weathered area (such as fractal dimension, GLCM contrast); the second apparent defect influencing factor refers to the comprehensive apparent quality defect quantitative index of the artificially poured area (such as spectral reflectance, three-dimensional point cloud reconstruction).

[0058] Specifically, step S30 includes: According to the apparent quality defect factors, the apparent defect influencing factors indicating the quality defect level are obtained, including: S301: Obtain preset defect quality association parameters, and screen a first apparent quality defect factor based on the preset defect quality association parameters to obtain a first quality defect feature, wherein the first quality defect feature includes a first texture complexity feature, a first crack distribution feature, and a first color non-uniformity feature.

[0059] In this embodiment, the preset defect quality association parameter refers to a pre-set threshold or weight, which is used to screen key features related to apparent quality defects (such as texture complexity and crack density); the first quality defect feature refers to the key apparent quality defect index of the naturally weathered area; the first texture complexity feature refers to the irregularity of the surface texture (such as the random distribution of cracks), where the texture complexity is GLCM contrast ≥ 0.8 or entropy ≥ 1.5; the first crack distribution feature refers to the density, length, width and spatial distribution of cracks, where the threshold of cracks is set to crack density ≥ 2 mm / m 2 ; The first color unevenness characteristic refers to uneven color distribution (such as color difference, mottled) caused by fading or pollution.

[0060] Specifically, when setting the threshold or weight of the preset defect quality associated parameter, the screening threshold (such as crack density ≥ 2mm / m 2 =High apparent defect); Texture complexity is obtained by calculating contrast and entropy through gray-level co-occurrence matrix (GLCM); Crack distribution is obtained by Canny edge detection (threshold = 50) and then statistically analyzing crack length and density; Color non-uniformity is obtained by the standard deviation of the lightness channel (L) in the LAB color space (σ ≥ 15 is judged as high non-uniformity); In this embodiment, the texture complexity (weight 0.4), crack density (weight 0.4), and color non-uniformity (weight 0.2) are weighted and summed to generate the first quality defect feature value, and the feature overlap rate between the naturally weathered area and the artificially poured area is made ≤10% through feature discrimination, and the tolerance of the concrete surface image to illumination changes is measured.

[0061] S302: According to the first quality defect feature, obtain a first apparent defect influencing factor by calculating the image gray level co-occurrence matrix.

[0062] In this embodiment, the gray level co-occurrence matrix (GLCM) refers to the spatial distribution of gray values ​​of pixel pairs in a statistical image to quantify texture features; the first apparent defect influencing factor refers to a quantitative indicator calculated based on the GLCM, reflecting the apparent quality defect level of the naturally weathered area.

[0063] Specifically, the distance parameter in the gray-level co-occurrence matrix (GLCM) is set to 1 pixel, and the angle is set to four directions: 0°, 45°, 90°, and 135°; the normalization parameters include contrast, correlation, energy, and homogeneity.

[0064] In the gray-level co-occurrence matrix (GLCM), contrast: Contrast = ∑ i，j |ij| 2 ×P(i, j), where i and j represent the gray level indices in the gray level co-occurrence matrix (GLCM) (usually 0-255); P(i, j) is the probability of the gray level pair (i, j) in the gray level co-occurrence matrix.

[0065] Furthermore, the first apparent defect influencing factor includes a first apparent defect intrinsic influencing factor and a first apparent defect extrinsic influencing factor. According to the first quality defect characteristic, obtaining the first apparent defect influencing factor includes: S3021: Based on the weathering degree characteristics and crack distribution characteristics, the intrinsic influencing factor of the first apparent defect is obtained through fractal dimension calculation and morphological analysis.

[0066] In this embodiment, the first intrinsic factor of apparent defects is used to quantify the apparent quality defects of the material in naturally weathered areas, reflecting the inherent properties of the material. It is independent of the external environment and is applicable to naturally weathered areas (such as cracks and faded areas). The fractal dimension is used to describe the self-similarity complexity of the surface texture (the higher the fractal dimension, the rougher the surface). Morphological analysis extracts the geometric characteristics of the cracks (such as edge sharpness and hole distribution) through operations such as corrosion and expansion. Specifically, the fractal dimension calculation (Box-counting algorithm) is based on the grayscale image of the naturally weathered area (resolution ≥512×512), and the calculation is performed from 1 pixel to 1 / 4 of the image size. The grid size is gradually reduced (such as from 1×1 to 32×32), and the number of black pixels in each grid is counted. The fractal dimension D is calculated by the formula log(N(∈))=-D×log(∈)+C, where N(∈) is the number of pixels under the grid size ∈, and the fractal dimension D (unit: dimensionless) is output; C is a proportional constant, which is related to the object geometry, noise level and image resolution; when analyzing and calculating the intrinsic influencing factor of the first apparent defect, the fractal dimension weight is 0.7 (dominant factor); the edge density weight is 0.3 (auxiliary factor).

[0067] S3022: Obtain a first apparent defect external influencing factor through color histogram contrast analysis based on the color non-uniformity characteristics.

[0068] In this embodiment, the first apparent defect external influencing factor is used to reflect the quantitative index of the uneven color distribution on the surface of the naturally weathered area, which is related to the external environment (such as stains and light) and is suitable for naturally weathered areas such as faded areas and stained areas; the color histogram contrast is used to count the discrete degree of pixel distribution of the color channel (such as the brightness distribution of the HSV-S channel).

[0069] Specifically, the RGB image is converted to the HSV color space, the brightness (V channel) is extracted, the V channel value is divided into 16 bins (0-255), and then the mean is calculated. Standard deviation Then calculate the contrast, normalize the extreme contrast, and get the value of the first apparent defect external influence factor FI 外在 , and then perform threshold determination, such as FI 外在 ≥0.7: High color unevenness (such as severe fading); 0.4≤FI 外在 <0.7: medium color unevenness; FI 外在 <0.4: low color unevenness.

[0070] S303: Screening a second apparent quality defect factor according to preset defect quality association parameters to obtain a second quality defect feature, wherein the second quality defect feature includes a second surface morphology feature, a second porosity feature, a second edge sharpness feature, and a second light reflection feature.

[0071] In this embodiment, the preset defect quality association parameter is a screening threshold for artificial defects (such as porosity ≥10% is a high apparent defect); the second quality defect feature is a key apparent quality defect indicator of the artificial casting area; the second surface morphology feature refers to the geometric irregularity of the surface undulations (such as honeycomb surface); the second porosity feature refers to the volume proportion of tiny holes on the surface; the second edge sharpness feature refers to the clarity of the defect edge (such as template splicing marks); the second light reflection feature refers to the surface reflection intensity distribution (such as uneven paint).

[0072] Specifically, during feature extraction, the root mean square roughness (SB value) of the surface morphology is calculated through three-dimensional point cloud reconstruction (Kinect v2); the porosity is extracted by morphological top-hat transformation, and the volume ratio ≥10% is marked as high porosity; the edge sharpness is calculated by the gradient mean using the Sobel operator (≥15 is a sharp edge); the light reflection is determined by the standard deviation of the lightness channel (V) in the HSV color space (σ≥20 is judged as high reflection); the corresponding thresholds are set as porosity ≥10%; edge gradient mean ≥15; HSV-V standard deviation σ≥20; and then feature fusion is performed: the surface morphology (weight 0.5), porosity (weight 0.3), and edge sharpness (weight 0.2) are weighted association calculated.

[0073] S304: Obtain a second apparent defect influencing factor based on the second quality defect feature through edge detection algorithm and color histogram analysis.

[0074] In this embodiment, the edge detection algorithm locates the defect edge through the Sobel or Canny operator; the color histogram analysis statistically analyzes the color distribution characteristics (such as saturation unevenness); the second apparent defect influencing factor is based on the quantitative indicators of geometric and color characteristics, reflecting the apparent quality defect level of the artificial casting area.

[0075] Specifically, the contrast calculation formula of the HSV-S channel histogram is: Where n1 is the number of samples, that is, the number of pixels involved in the calculation (or the number of bins in the histogram); i is the image sample index; h i is the number of pixels or frequency of the i-th bin; μ is the number of pixels or frequency of all h i The average value of , that is, the mean of the histogram; Contrast is used to quantify the degree of discreteness of color distribution.

[0076] Entropy = -∑ i，j P(i, j)logP(i, j), where i and j represent the grayscale indices in the gray-level co-occurrence matrix (GLCM) (typically 0-255); P(i, j) is the probability of the grayscale pair (i, j) in the GLCM. The second apparent defect influencing factor = w1 × SB value + w2 × edge sharpness + w3 × color contrast (weights w1 = 0.5, w2 = 0.3, w3 = 0.2).

[0077] Furthermore, the second apparent defect influencing factor includes a second apparent defect intrinsic influencing factor and a second extrinsic influencing factor. The second apparent defect influencing factor is obtained according to the second quality defect characteristic, specifically including: S3041: Based on the material type characteristics, porosity characteristics and edge sharpness characteristics, the intrinsic influencing factors of the second apparent defect are obtained through spectral reflectance and 3D point cloud reconstruction.

[0078] Specifically, the second intrinsic influencing factor of apparent defects reflects the quantitative index of apparent quality defects of the inherent properties of the material in the artificial casting area. It is unrelated to the external environment and is applicable to artificial casting areas such as honeycomb surfaces and cold joints. Spectral reflectivity refers to the material's ability to reflect light of a specific wavelength and is used to analyze composition uniformity (such as Ca(OH)2 content). Three-dimensional point cloud reconstruction is to obtain surface geometry data through a depth camera and generate a point cloud model to quantify morphological features.

[0079] For example, porosity = total area volume / pore volume; porosity ≥ 10% is marked as a high apparent defect area; spectral weight is 0.6 (dominant factor); porosity weight is 0.4 (auxiliary factor); FI 内在 =0.6×spectral matching+0.4×porosity.

[0080] S3042: According to the construction defect characteristics, the second apparent defect external influencing factor is obtained through morphological operation and defect area segmentation.

[0081] Specifically, the second apparent defect external influencing factor reflects the quantitative index of surface construction defects in the artificial pouring area, is related to the external process, and is applicable to artificial pouring areas such as template splicing marks and rough surfaces; defect area segmentation locates construction defects (such as cold joints and air holes) through edge detection or watershed algorithm, and the defect area can be pre-marked by Otsu threshold segmentation; the watershed algorithm based on distance transformation is used to segment the defect area and suppress over-segmentation (merging rate ≥70%); the cold joint has a length ≥10mm and an edge sharpness ≥15; the honeycomb rough surface has a pore diameter of 1-3mm and an area ratio ≥5%; for example, FI external = 0.5×edge density + 0.5×defect area ratio.

[0082] S40: Divide the first inspection area and the second inspection area according to the apparent defect influencing factor.

[0083] Specifically, step S40 includes: S401: According to the first apparent defect influencing factor, first apparent quality detection points are obtained through K-means clustering, and a first detection area is divided according to the first apparent quality detection points.

[0084] In this embodiment, the K-means clustering unsupervised algorithm divides the detection points according to the similarity of the influencing factors; the first apparent defect influencing factor refers to the characteristic quantitative index of the natural weathering area (such as texture complexity, crack density, and color unevenness); the first apparent quality detection point is a candidate point of the high apparent defect area obtained by clustering, which is used for subsequent area division.

[0085] Specifically, the number of clusters of K-means clustering is K=3, where the number of clusters K is set according to the actual scenario (natural weathering areas are usually divided into three categories: high, medium, and low apparent defects, K=3). The Euclidean distance between each detection point and the cluster center is calculated and assigned to the nearest cluster for cluster assignment. Then cluster labels are generated, such as the high apparent defect cluster is marked in red, the medium apparent defect is yellow, and the low apparent defect is green; only the high apparent defect cluster area with a density ≥0.5 points / pixel is retained, and the discrete detection points are connected based on morphological close connection to form a continuous area.

[0086] S402: According to the second apparent defect influencing factor, a second apparent quality detection point is obtained by a support vector machine classifier, and a detection area is divided according to the second apparent quality detection point.

[0087] In this embodiment, a support vector machine (SVM) supervised algorithm is used to divide high / low apparent defect areas based on the feature space; the second apparent defect influencing factor refers to the characteristic quantitative index of the artificial casting area (such as material type, porosity, and edge sharpness); the second apparent quality detection point refers to the high apparent defect sample point after SVM classification, which is used to generate the detection area.

[0088] Specifically, the penalty parameter C of the support vector machine (SVM) is 1.0, balancing the classification interval and the false detection rate; the kernel parameter γ is 0.1 (controlling the overfitting risk of the RBF kernel). In this embodiment, the LIBSVM library is used to train the model, and the input image features are classified by SVM. The probability value is output (high apparent defect probability ≥0.8 is positive), and the detection points with a probability ≥0.8 are retained. The coverage of the detection points is expanded based on the morphological dilation operation to ensure the integrity of the boundary.

[0089] S50: Obtaining a segmentation result, and obtaining an apparent defect quality detection parameter according to the segmentation result.

[0090] Specifically, step S50 includes: S501: Calculating the surface roughness SA value and SB value according to the division result of the first detection area.

[0091] Specifically, SA value refers to the surface roughness parameter defined in ISO 4287 standard, the arithmetic mean deviation, which is used to reflect the average value of the surface micro-geometric roughness; SB value is the root mean square roughness, which is used to reflect the root mean square value of the surface micro-geometric roughness, reflecting the micro-geometric roughness. Among them, z i is the surface profile height; is the average height; n is the number of sampling points; the unit of SA value and SB value is micrometer (μm).

[0092] S502: Generate an appearance quality grading evaluation result and a three-dimensional shape reconstruction model according to the division result of the second detection area.

[0093] Specifically, the surface quality grading evaluation refers to dividing the artificial casting area into high, medium and low surface defect levels based on the SB value (for example, SB ≥ 3.0 μm is a high surface defect); the three-dimensional morphology reconstruction model is a surface visualization model based on point cloud data, which marks the defect location and severity.

[0094] For example, the apparent quality grading threshold is set as high apparent defect (red): SB ≥ 3.0 μm; medium apparent defect (yellow): 2.0 μm ≤ SB < 3.0 μm; low apparent defect (green): SB < 2.0 μm; the high apparent defect area is highlighted in the three-dimensional morphology reconstruction model.

[0095] It should be understood that the serial numbers of the steps in the above embodiments do not imply the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0096] In one embodiment, an image analysis system for detecting apparent quality defects of concrete is provided. The image analysis system for detecting apparent quality defects of concrete corresponds to the image analysis method for detecting apparent quality defects of concrete in the above embodiment.

[0097] The image analysis system for concrete surface quality defect detection includes an image acquisition module, an image preprocessing and labeling module, a key feature extraction module, a multi-algorithm comparison and verification framework module, and a model optimization and selection module. A detailed description of each functional module is as follows: An image acquisition module is used to use a camera to capture images of the concrete surface at a specified distance interval and control the optical axis of the camera lens to be perpendicular to the concrete surface; An image preprocessing and marking module is used to preprocess the collected concrete surface images and mark the defect types; A key feature extraction module is used to extract key features from the pre-processed concrete surface image using a convolutional neural network to obtain key feature information; The multi-algorithm comparison and verification framework module is used to build a multi-algorithm comparison and verification framework based on multiple defect detection models, and perform cross-validation in combination with key feature information to obtain defect detection results and evaluation results of multiple defect detection models; The model optimization and selection module is used to optimize the model parameters based on the evaluation results and the confusion matrix to determine the final concrete surface detection model; in the concrete surface detection model, the apparent quality defect detection results are obtained based on the defect detection results and the preset quality defect classification threshold.

[0098] For the specific limitations of the image analysis system for detecting apparent quality defects of concrete, please refer to the limitations of the image analysis method for detecting apparent quality defects of concrete mentioned above, which will not be repeated here; the various modules in the above-mentioned image analysis system for detecting apparent quality defects of concrete can be implemented in whole or in part through software, hardware and their combination; the above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above-mentioned modules.

[0099] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented: S1: Use a camera to collect images of the concrete surface at a specified distance interval, and control the optical axis of the camera lens to be perpendicular to the concrete surface; perform image preprocessing on the collected concrete surface images and mark the defect types; S2: Use a convolutional neural network to extract key features from pre-processed concrete surface images to obtain key feature information; S3: Build a multi-algorithm comparison and verification framework based on multiple defect detection models, combine key feature information for cross-validation, and obtain defect detection results and evaluation results of multiple defect detection models; S4: Based on the evaluation results and the confusion matrix, the model parameters are optimized to determine the final concrete surface detection model; in the concrete surface detection model, the apparent quality defect detection results are obtained based on the defect detection results and the preset quality defect classification threshold.

[0100] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0101] In one embodiment, particularly according to an embodiment of the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present invention includes a computer program product comprising a computer program / instructions that, when executed by a processor, implements the steps of the image analysis method for detecting surface quality defects of concrete. In such an embodiment, the computer program can be downloaded and installed from a network via a communication module and / or installed from removable media. When executed by a central processing unit (CPU), the computer program performs the various functions defined in the present invention.

[0102] Those skilled in the art will clearly understand that for the sake of convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0103] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, it should be understood by those skilled in the art that the technical solutions described in the aforementioned embodiments may still be modified, or some of the features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present application, and should all be included in the scope of protection of the present application.

Claims

1. An image analysis method for detecting apparent quality defects of concrete, characterized in that: include: Use a camera to capture images of the concrete surface at a specified distance, and control the optical axis of the camera lens to be perpendicular to the concrete surface; Perform image preprocessing on the collected concrete surface images and mark the defect types; Convolutional neural network is used to extract key features from pre-processed concrete surface images to obtain key feature information; Constructing a multi-algorithm comparison and verification framework based on multiple defect detection models, performing cross-validation in combination with the key feature information, and obtaining defect detection results and evaluation results of the multiple defect detection models; Based on the evaluation results and the confusion matrix optimization model parameters, a final concrete surface detection model is determined; in the concrete surface detection model, an apparent quality defect detection result is obtained based on the defect detection result and a preset quality defect classification threshold.

2. The image analysis method for detecting concrete surface quality defects according to claim 1, characterized in that: The method further comprises: Acquire a concrete surface area, divide the concrete surface area into a first detection area and a second detection area; Acquire, based on the first detection area and the second detection area, apparent quality defect factors related to apparent quality defects of concrete, wherein the apparent quality defect factors include a first apparent quality defect factor and a second apparent quality defect factor; Obtaining an apparent defect impact factor indicating a quality defect grade according to the apparent quality defect factor, wherein the apparent defect impact factor includes a first apparent defect impact factor and a second apparent defect impact factor; dividing the first inspection area and the second inspection area according to the apparent defect influencing factor; A division result is obtained, and an apparent defect quality detection parameter is obtained according to the division result.

3. The image analysis method for detecting concrete surface quality defects according to claim 2, characterized in that: The acquiring of the concrete surface area and dividing the concrete surface area to obtain the first detection area and the second detection area include: According to the concrete surface area, obtaining a naturally weathered surface area and an artificially poured surface area; Obtaining a defect characteristic area according to the naturally weathered surface area and the artificially poured surface area; A first detection area and a second detection area are acquired according to the defect characteristic area and the concrete surface area.

4. The image analysis method for detecting concrete surface quality defects according to claim 2, characterized in that: The step of obtaining an apparent defect influencing factor indicating a quality defect grade based on the apparent quality defect factor includes: Obtaining preset defect quality association parameters, and screening the first apparent quality defect factor based on the preset defect quality association parameters to obtain a first quality defect feature, wherein the first quality defect feature includes a first texture complexity feature, a first crack distribution feature, and a first color non-uniformity feature; Obtaining a first apparent defect influencing factor by calculating an image gray level co-occurrence matrix according to the first quality defect feature; Screening the second apparent quality defect factor according to the preset defect quality association parameter to obtain a second quality defect feature, wherein the second quality defect feature includes a second surface morphology feature, a second porosity feature, a second edge sharpness feature, and a second light reflection feature; According to the second quality defect feature, a second apparent defect influencing factor is obtained through an edge detection algorithm and color histogram analysis.

5. The image analysis method for detecting concrete surface quality defects according to claim 2 or 4, characterized in that: The dividing the first detection area and the second detection area according to the apparent defect influencing factor includes: Obtaining first apparent quality inspection points through K-means clustering according to the first apparent defect influencing factor, and dividing the first inspection area according to the first apparent quality inspection points; According to the second apparent defect influencing factor, obtaining a second apparent quality detection point by using a support vector machine classifier, and dividing the detection area according to the second apparent quality detection point; and / or, The obtaining of apparent defect quality detection parameters includes: Calculating the surface roughness SA value and SB value according to the division result of the first detection area; According to the division result of the second detection area, an apparent quality grading evaluation result and a three-dimensional morphology reconstruction model are generated.

6. The image analysis method for detecting concrete surface quality defects according to claim 4, characterized in that: The obtaining of apparent quality defect factors related to apparent quality defects of concrete based on the first detection area and the second detection area includes: Acquire a first apparent quality defect factor according to the first detection area, wherein the first apparent quality defect factor includes a weathering degree characteristic, a crack distribution characteristic, and a color unevenness characteristic; According to the second detection area, a second apparent quality defect factor is obtained, wherein the second apparent quality defect factor includes a material type feature, a construction defect feature, a porosity feature, and an edge sharpness feature.

7. The image analysis method for detecting concrete surface quality defects according to claim 6, characterized in that: The first apparent defect influencing factor includes a first apparent defect intrinsic influencing factor and a first apparent defect extrinsic influencing factor. Obtaining the first apparent defect influencing factor according to the first quality defect characteristic includes: According to the weathering degree characteristics and crack distribution characteristics, obtaining the first apparent defect intrinsic influencing factor through fractal dimension calculation and morphological analysis; Obtaining a first apparent defect external influencing factor through color histogram contrast analysis according to the color non-uniformity characteristic; and / or, The second apparent defect influencing factor includes a second apparent defect intrinsic influencing factor and a second extrinsic influencing factor. The obtaining of the second apparent defect influencing factor according to the second quality defect characteristic specifically includes: According to the material type characteristics, porosity characteristics and edge sharpness characteristics, the intrinsic influencing factor of the second apparent defect is obtained through spectral reflectance and three-dimensional point cloud reconstruction; According to the construction defect characteristics, a second apparent defect external influencing factor is obtained through morphological operation and defect area segmentation.

8. An image analysis system for detecting apparent quality defects of concrete, characterized in that: Used to perform the image analysis method for detecting concrete surface quality defects according to any one of claims 1 to 7; the system comprises: An image acquisition module is used to use a camera to capture images of the concrete surface at a specified distance interval and control the optical axis of the camera lens to be perpendicular to the concrete surface; An image preprocessing and marking module is used to preprocess the collected concrete surface images and mark the defect types; A key feature extraction module is used to extract key features from the pre-processed concrete surface image using a convolutional neural network to obtain key feature information; A multi-algorithm comparison and verification framework module is used to build a multi-algorithm comparison and verification framework based on multiple defect detection models, perform cross-validation in combination with the key feature information, and obtain defect detection results and evaluation results of the multiple defect detection models; The model optimization and selection module is used to optimize the model parameters based on the evaluation results and the confusion matrix to determine the final concrete surface detection model; in the concrete surface detection model, the apparent quality defect detection results are obtained based on the defect detection results and the preset quality defect classification threshold.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the image analysis method for detecting apparent quality defects of concrete as claimed in any one of claims 1 to 7 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the image analysis method for detecting apparent quality defects of concrete as claimed in any one of claims 1 to 7 are implemented.

Images