Deep learning-based concrete scouring quality visual inspection and grading system

Abstract

The invention provides a concrete wool washing quality visual detection and grading system based on deep learning, and is suitable for the technical field of concrete wool washing. The system comprises a data acquisition module, a dynamic preprocessing module, a deep learning model identification module and a decision interaction module, three-dimensional space coordinate information is converted into concrete burr quality evaluation indexes, quantitative quality grading is carried out, and accurate classification of undershoot / qualified / overshoot is achieved; brenner function verification and liquid focusing lens control are adopted, rough surface automatic focusing is achieved, rough surface image definition is improved, and the shooting speed is increased; according to the model, through ImageNet pre-training and Focal Loss optimization, the quality level can be rapidly output after single shooting, and image splicing and rework area labeling are achieved by combining improved SIFT feature matching.

Description

Technical Field

[0001] The present invention relates to the technical field of concrete roughening, and in particular to a visual inspection and grading system for concrete roughening quality based on deep learning. Background Art

[0002] During the concrete pouring process, it is usually not possible to complete the entire pouring in one go, and it needs to be carried out in stages. It is inevitable to consider the bearing capacity and durability of the interface between the new and old concrete. Concrete structures in special locations also need to consider the impermeability and erosion resistance of the interface between the new and old concrete. Therefore, before pouring new concrete at the interface between the new and old concrete, it is necessary to chisel or flush the interface of the old concrete to increase the roughness of the interface and enable the new and old concrete at the interface to better combine. The traditional chiseling method requires workers to use chiseling equipment, which is time-consuming and labor-intensive, and is gradually being replaced by high-pressure water gun flushing. However, the flushing method needs to be performed after the old concrete has solidified to a certain strength and before it is completely hardened, which requires relatively high processing time.

[0003] To ensure a reliable bond between new and old concrete, existing specifications stipulate that, from a quantitative perspective, the depth of the rough surface should be no less than 6mm, and from a qualitative perspective, the rough surface needs to have a slight amount of coarse sand exposed. Generally speaking, the rough surface of concrete is directly observed by construction personnel based on their construction experience to determine whether the rough surface of the concrete after roughening is qualified. For projects that require precise quantitative analysis, methods such as sand filling method, roughness meter method, fractal dimension method, needle touch method, and fine iron bead measurement method can be used for measurement. However, these methods all have the problem of complex operation, which will greatly affect construction efficiency during actual project application. In addition, there is also a method based on three-dimensional point cloud scanning to evaluate the roughness of the rough surface of concrete after roughening, but the scanning and data processing process of this method is not efficient. Therefore, there is an urgent need for a method that can quickly and accurately detect the quality of concrete roughening. Summary of the Invention

[0004] The purpose of this invention is to solve the problem of low efficiency of existing concrete roughening quality detection and impact on on-site construction progress. A concrete roughening quality visual detection and grading system based on deep learning is proposed, which can be widely used in the field of concrete roughening technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A deep learning-based visual inspection and grading system for concrete roughening quality, including:

[0007] The data acquisition module uses an autofocus industrial camera with a ring light source to collect raw image data of the rough surface of the concrete construction joint after roughening;

[0008] Dynamic preprocessing module for optimizing the quality of raw image data in real time;

[0009] A deep learning model recognition module uses a concrete roughening quality visual inspection model to identify the quality level of the rough surface of concrete construction joints after roughening;

[0010] The decision interaction module is used to display real-time detection results, stitch the original image data to form an overall image of the scanned area, and mark the rework area.

[0011] As a preferred technical solution of the present invention, the auto-focus industrial camera realizes auto-focus based on the defocus distance prediction algorithm of deep learning. The specific implementation process includes model architecture design, data set construction, model optimization training and focus strategy implementation;

[0012] The model architecture is designed by using a lightweight CNN network ShuffleNetV2 to extract image features and adding a three-layer MLP to map the features output by ShuffleNetV2 to defocus distance.

[0013] The dataset is constructed by using a liquid focusing lens. The voltage value is adjusted by a fixed step size to change the refractive index of the liquid. Image data at different defocus distances are collected. The image dataset collected at each data collection point covers the range from severe defocus to optimal focus. The Brenner clarity evaluation function is used to determine the optimal focus position, and the defocus distance of the optimal focus position is marked as 0. The defocus distance l of other images is calculated using formula (1).

[0014] l=|V i -V b | (1)

[0015] Where V i -The voltage of the liquid focusing lens corresponding to the i-th shooting, V b -The voltage corresponding to the liquid focus lens when shooting at optimal focus; all image data sets collected at different data collection points are divided into training set, validation set and test set according to the ratio of 7:2:1;

[0016] Model optimization training uses Huber loss function, L2 regularization term, Adam optimizer to train the defocus distance prediction model, with an initial learning rate of 0.0001; the mean absolute error MAE and determination coefficient R 2 Measure the predictive performance of the model;

[0017] Focus strategy is implemented based on the predicted defocus distance l p Divided into three levels, l p ∈(10,+∞) is a large defocus, and a large step size is used to quickly approach the optimal focus position. p∈(1,10] is medium defocus, using a medium step size to balance speed and accuracy, l p ∈(0,1] is a small defocus, and micro-step length is used for fine adjustment; once the image is acquired, the model predicts the defocus distance l p Then, the camera dynamically matches the step size and moves directly to the target area. The focus quality is verified by combining the Brenner function. If the quality does not meet the standard, a secondary fine-tuning of the focus is triggered.

[0018] As a preferred technical solution of the present invention, the real-time optimization includes using a multi-scale Retinex algorithm to decompose the incident light and reflected light components, and eliminating uneven illumination through multi-scale spatial convolution of a Gaussian filter kernel.

[0019] As a preferred technical solution of the present invention, the training process of the concrete roughening quality visual detection model includes collecting training sample data sets, data preprocessing, model architecture design and model training;

[0020] A training sample dataset was collected. The shooting distance between the camera and the rough surface of the concrete construction joint was fixed to ensure that the shooting range was consistent with the single scanning range of the 3D line laser scanner. Each scan corresponded to capturing an image of the rough surface of the concrete construction joint. The 3D line laser scanner was used to collect the 3D spatial coordinate information of the rough surface of the concrete construction joint after the concrete structure was roughened. The quality level of the corresponding image was evaluated based on the 3D spatial coordinate information.

[0021] Data preprocessing uses a multi-scale Retinex algorithm to decompose the incident and reflected light components, multi-scale spatial convolution with a Gaussian filter kernel to eliminate uneven illumination, and rotation, flipping, and brightness / contrast jittering to enhance data diversity. A hybrid strategy of oversampling and undersampling is used to generate synthetic samples based on feature space interpolation, and the K-nearest neighbor rule is applied to screen boundary samples for interpolation. A dynamic weight allocation mechanism is introduced to adjust the regional weights of generated samples based on category density, prioritizing the addition of synthetic samples to classification boundary areas. A generative adversarial network is combined to perform data augmentation in the latent space, and the discriminator-generator game is used to improve sample authenticity. Noise filtering is performed simultaneously, and anomalies in overlapping areas are removed using the edited nearest neighbor algorithm.

[0022] In the model architecture design, the feature extraction layer uses an improved ResNet-34 as the backbone network, embedding the ECA attention module to enhance feature extraction; a multi-scale fusion layer is constructed, introducing bidirectional cross-layer connections and deformable convolution modules; the hierarchical decision layer integrates a spatial pyramid pooling layer to distinguish between under-brush, qualified, and over-brush quality of concrete;

[0023] For model training, the backbone network was initialized based on the ImageNet pre-trained weights, the parameters of the first three convolutional layers were frozen, and the high-level modules were fine-tuned using a cosine annealing learning rate. The parameters of the first three convolutional layers were then unfrozen, and Focal Loss was introduced to balance the distribution of undershoot / overshoot samples. γ was set to 2.0 to suppress the weights of easily classified samples. The AdamW optimizer was used for training, and the learning rate was decayed in stages. Early stopping on the validation set was performed every 10 epochs, and a model rollback was triggered when the validation loss did not decrease for three consecutive times.

[0024] As a preferred technical solution of the present invention, the quality level evaluation is:

[0025] Q=w1H1+w2S1+w3N1 (2)

[0026] Where Q is the concrete roughening quality evaluation index, w1, w2, w3 are weight coefficients, H1 is the normalized maximum vertical height difference between adjacent peaks and troughs, S1 is the normalized slope between the maximum peak and the maximum trough, and N1 is the normalized number of peaks per unit area. The normalization is performed based on the maximum and minimum reference values ​​of each index.

[0027]

[0028]

[0029] Where H is the maximum vertical height difference between adjacent peaks and troughs in a single scan area, mm; S is the slope between the maximum peak and the maximum trough in a single scan area; and N is the number of peaks per unit area, z. p,i -The height of the i-th peak, mm, z v,i -The height of the valley adjacent to the i-th peak, mm, (x mp ,y mp , z mp )-coordinates of the maximum peak, (x mv ,y mv , z mv )-coordinates of the maximum trough, n-number of peaks in a single scan area, s-area of ​​a single scan area; the maximum and minimum reference values ​​are calculated based on historical quality inspection results; when Q≤0.2, the quality level is evaluated as undershoot; when 0.2<Q≤0.8, the quality level is evaluated as qualified; when Q>0.8, the quality level is evaluated as overshoot.

[0030] As a preferred technical solution of the present invention, the spliced ​​original image data adopts improved SIFT feature matching combined with RANSAC algorithm, sets the matching point threshold ≥50 pairs, and the homography matrix reprojection error ≤1.5 pixels.

[0031] The beneficial effects of the present invention are: 1. Quantitative quality grading, which converts three-dimensional spatial coordinate information into a concrete roughening quality evaluation index to achieve accurate classification of under- / qualified / over-shooting; 2. Brenner function verification and liquid focus lens control are used to achieve automatic focusing of rough surfaces, improve the clarity of rough surface images, and speed up shooting; 3. The model is pre-trained on ImageNet and optimized with Focal Loss, and can quickly output the quality level after a single shot. Combined with improved SIFT feature matching, image stitching and rework area labeling are achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 This is the workflow diagram of the concrete roughening quality visual inspection and grading system based on deep learning of the present invention. DETAILED DESCRIPTION

[0033] The following describes in detail specific embodiments of the present invention in conjunction with the accompanying drawings. It should be understood that the specific embodiments provided herein are intended only to illustrate and explain the present invention and are not intended to limit the present invention. It should be noted that many specific details are set forth in the following description to facilitate a full understanding of the present invention. However, the present invention may also have other embodiments and variations thereof. Therefore, the scope of protection of the present invention is not limited by the specific embodiments disclosed below.

[0034] Embodiment 1, a concrete roughening quality visual inspection and grading system based on deep learning, includes a data acquisition module, a dynamic preprocessing module, a deep learning model recognition module and a decision interaction module.

[0035] Module 1, the data acquisition module, uses an autofocus industrial camera paired with a ring light source to capture raw image data of the rough surface of a concrete construction joint after roughening. The autofocus industrial camera uses a deep learning-based defocus distance prediction algorithm to achieve automatic focus, quickly achieving precise focus on rough surfaces while ensuring image clarity and capture speed. The specific implementation process includes model architecture design, dataset construction, model optimization training, and focus strategy implementation.

[0036] Among them, the model architecture design uses the lightweight CNN network ShuffleNetV2 to extract image features, and adds a 3-layer MLP to map the features output by ShuffleNetV2 to defocus distance.

[0037] Dataset construction: A liquid-focus lens is used. The voltage value is adjusted by a fixed step of 0.2V to change the refractive index of the liquid, and image data at different defocus distances are collected. The image dataset collected at each data acquisition point covers the range from severe defocus to optimal focus. The Brenner sharpness evaluation function is used to determine the optimal focus position, and the defocus distance corresponding to the optimal focus position is marked as 0. The defocus distance l of other images is calculated by formula (1).

[0038] l = |V i - V b | (1)

[0039] In the formula, V i - The voltage of the liquid-focus lens corresponding to the i-th shot, V b - The voltage of the liquid-focus lens corresponding to the optimal focus shot; The dataset is divided by acquisition points to ensure that the training set / validation set / test set = 7:2:1, avoiding misallocation of data at the same point to different datasets.

[0040] Model optimization training: The Huber loss function, L2 regularization term, and Adam optimizer are used to train the defocus distance prediction model. The initial learning rate is 0.0001, and it decays by 50% every 20 epochs. Training terminates when the validation loss does not decrease for 5 consecutive times; The mean absolute error MAE and the coefficient of determination R 2 are used to measure the prediction performance of the model.

[0041] Focusing strategy implementation: The current frame image is collected and input into the defocus distance prediction model to obtain l p , and the voltage adjustment step ΔV is selected according to the l p level:

[0042] If l p ∈(10, +∞) is large defocus, a large step ΔV = 1V is used to quickly approach the optimal focus position;

[0043] If l p ∈(1, 10] is medium defocus, a medium step ΔV = 0.5V is used to balance speed and accuracy;

[0044] If l p ∈(0, 1] is small defocus, a micro step ΔV = 0.1V is used for fine adjustment.

[0045] After adjustment, a new image is collected, and the Brenner value is calculated to verify the sharpness. If Brenner ≥ threshold T, the focusing is completed. If Brenner < T: Trigger secondary small defocus fine-tuning until the threshold T is reached or the maximum adjustment次数 is exceeded; Record the final V b value, further update the dataset, and train the defocus distance prediction model.

[0046] Module 2, the dynamic preprocessing module, is used to optimize the quality of the original image data in real time. This real-time optimization includes using a multi-scale Retinex algorithm to decompose the incident and reflected light components, and eliminating uneven illumination through multi-scale spatial convolution of a Gaussian filter kernel.

[0047] Module three, the deep learning model recognition module, uses a concrete roughening quality visual inspection model to identify the quality level of the rough surface of concrete construction joints after roughening. The training process of this concrete roughening quality visual inspection model includes collecting training sample datasets, data preprocessing, model architecture design, and model training.

[0048] A training sample data set is collected, and the shooting distance between the shooting equipment and the rough surface of the concrete construction joint is fixed so that the shooting range is consistent with the single scanning range of the 3D line laser scanner. Each scan corresponds to taking an image of the rough surface of the concrete construction joint. The 3D line laser scanner is used to collect the three-dimensional spatial coordinate information of the rough surface of the concrete construction joint after the concrete structure is roughened. The quality level of the corresponding image is evaluated based on the three-dimensional spatial coordinate information.

[0049] For data preprocessing, a multi-scale Retinex algorithm is used to decompose the incident and reflected light components, multi-scale spatial convolution with a Gaussian filter kernel is used to eliminate uneven illumination, and rotation, flipping, and brightness / contrast jitter are applied to enhance data diversity. A hybrid strategy of oversampling and undersampling is adopted to generate synthetic samples based on feature space interpolation, and the K-nearest neighbor rule is applied to screen boundary samples for interpolation. A dynamic weight allocation mechanism is introduced to adjust the regional weights of generated samples according to the category density, and synthetic samples are preferentially added to the classification boundary area. A generative adversarial network is combined to perform data augmentation in the latent space, and the discriminator-generator game is used to improve the sample authenticity. Noise filtering is performed simultaneously, and an abnormal sample in the overlapping area is removed using the edited nearest neighbor algorithm.

[0050] In the model architecture design, the feature extraction layer uses an improved ResNet-34 as the backbone network, embeds the ECA attention module to enhance feature extraction, and adjusts the first-layer convolution kernel to 5×5 size; constructs a multi-scale fusion layer, introduces bidirectional cross-layer connections and deformable convolution modules; the hierarchical decision layer integrates the spatial pyramid pooling layer to distinguish the three categories of concrete roughening quality: under-punching, qualified, and over-punching.

[0051] For model training, the backbone network was initialized based on the ImageNet pre-trained weights, the parameters of the first three convolutional layers were frozen, and the high-level modules were fine-tuned using a cosine annealing learning rate. The parameters of the first three convolutional layers were then unfrozen, and Focal Loss was introduced to balance the distribution of undershoot / overshoot samples. γ was set to 2.0 to suppress the weights of easily classified samples. The AdamW optimizer was used for training, and the learning rate was decayed in stages. Early stopping on the validation set was performed every 10 epochs, and a model rollback was triggered when the validation loss did not decrease for three consecutive times.

[0052] When collecting training sample data sets, it is necessary to evaluate the quality level based on the three-dimensional spatial coordinate information. The specific evaluation formula is:

[0053] Q=w1H1+w2S1+w3N1 (2)

[0054] Where Q is the concrete roughening quality evaluation index, w1, w2, w3 are weight coefficients, H1 is the normalized maximum vertical height difference between adjacent peaks and troughs, S1 is the normalized slope between the maximum peak and the maximum trough, and N1 is the normalized number of peaks per unit area. The normalization is performed based on the maximum and minimum reference values ​​of each index.

[0055]

[0056]

[0057] Where H is the maximum vertical height difference between adjacent peaks and troughs in a single scan area, mm; S is the slope between the maximum peak and the maximum trough in a single scan area; and N is the number of peaks per unit area, z. p,i -The height of the i-th peak, mm, z v,i -The height of the valley adjacent to the i-th peak, mm, (x mp ,y mp , z mp )-coordinates of the maximum peak, (x mv ,y mv , z mv )-coordinates of the maximum trough, n-number of peaks in a single scan area, s-area of ​​a single scan area; the maximum and minimum reference values ​​are calculated based on historical quality inspection results; when Q≤0.2, the quality level is evaluated as undershoot; when 0.2<Q≤0.8, the quality level is evaluated as qualified; when Q>0.8, the quality level is evaluated as overshoot.

[0058] Module 4, the Decision Interaction Module, displays real-time inspection results, stitches together the original image data to form a complete image of the scanned area, and annotates rework areas. This stitching uses an improved SIFT feature matching algorithm combined with the RAN SAC algorithm, with a matching point threshold of ≥50 pairs and a homography matrix reprojection error of ≤1.5 pixels.

[0059] In summary, the concrete roughening quality visual inspection and grading system based on deep learning in the present invention has the characteristics of high detection efficiency and accurate evaluation in the field of concrete roughening technology.

[0060] It should be understood that the above embodiments are one or more embodiments of the present invention, and there are many other embodiments and variations thereof based on the present invention; the variations and modifications made by ordinary technicians in this industry through the present invention without making groundbreaking innovations all fall within the scope of protection of the present invention.

Claims

1. A deep learning-based visual inspection and grading system for concrete roughening quality, characterized by: include: The data acquisition module uses an autofocus industrial camera with a ring light source to collect raw image data of the rough surface of the concrete construction joint after roughening; Dynamic preprocessing module for optimizing the quality of raw image data in real time; A deep learning model recognition module uses a concrete roughening quality visual inspection model to identify the quality level of the rough surface of concrete construction joints after roughening; The decision interaction module is used to display real-time detection results, stitch the original image data to form an overall image of the scanned area, and mark the rework area.

2. The deep learning-based concrete roughening quality visual inspection and grading system according to claim 1 is characterized by: The autofocus industrial camera achieves autofocus based on a deep learning-based defocus distance prediction algorithm. The specific implementation process includes model architecture design, data set construction, model optimization training, and focus strategy implementation. The model architecture is designed by using a lightweight CNN network ShuffleNetV2 to extract image features and adding a three-layer MLP to map the features output by ShuffleNetV2 to defocus distance. The dataset is constructed by using a liquid focusing lens. The voltage value is adjusted by a fixed step size to change the refractive index of the liquid. Image data at different defocus distances are collected. The image dataset collected at each data collection point covers the range from severe defocus to optimal focus. The Brenner clarity evaluation function is used to determine the optimal focus position, and the defocus distance of the optimal focus position is marked as 0. The defocus distance l of other images is calculated using formula (1). l=|V i -V b | (1) Where V i -The voltage of the liquid focusing lens corresponding to the i-th shooting, V b -The voltage corresponding to the liquid focus lens when shooting with optimal focus; All image datasets collected at different data collection points are divided into training set, validation set and test set according to the ratio of 7:2:1; Model optimization training uses Huber loss function, L2 regularization term, Adam optimizer to train the defocus distance prediction model, with an initial learning rate of 0.0001; the mean absolute error MAE and determination coefficient R 2 Measure the predictive performance of the model; Focus strategy is implemented based on the predicted defocus distance l p Divided into three levels, l p ∈(10,+∞) is a large defocus, and a large step size is used to quickly approach the optimal focus position. p ∈(1,10] is medium defocus, using a medium step size to balance speed and accuracy, l p ∈(0,1] is a small defocus, and micro-step length is used for fine adjustment; once the image is acquired, the model predicts the defocus distance l p Then, the camera dynamically matches the step size and moves directly to the target area. The focus quality is verified by combining the Brenner function. If the quality does not meet the standard, a secondary fine-tuning of the defocus is triggered.

3. The deep learning-based visual inspection and grading system for concrete roughening quality according to claim 1 is characterized by: The real-time optimization includes decomposing the incident light and reflected light components using a multi-scale Retinex algorithm, and eliminating uneven illumination through multi-scale spatial convolution of a Gaussian filter kernel.

4. The deep learning-based visual inspection and grading system for concrete roughening quality according to claim 1 is characterized by: The training process of the concrete roughening quality visual inspection model includes collecting training sample data sets, data preprocessing, model architecture design and model training; A training sample dataset was collected. The shooting distance between the camera and the rough surface of the concrete construction joint was fixed to ensure that the shooting range was consistent with the single scanning range of the 3D line laser scanner. Each scan corresponded to capturing an image of the rough surface of the concrete construction joint. The 3D line laser scanner was used to collect the 3D spatial coordinate information of the rough surface of the concrete construction joint after the concrete structure was roughened. The quality level of the corresponding image was evaluated based on the 3D spatial coordinate information. Data preprocessing uses a multi-scale Retinex algorithm to decompose the incident and reflected light components, multi-scale spatial convolution with a Gaussian filter kernel to eliminate uneven illumination, and rotation, flipping, and brightness / contrast jittering to enhance data diversity. A hybrid strategy of oversampling and undersampling is used to generate synthetic samples based on feature space interpolation, and the K-nearest neighbor rule is applied to screen boundary samples for interpolation. A dynamic weight allocation mechanism is introduced to adjust the regional weights of generated samples based on category density, prioritizing the addition of synthetic samples to classification boundary areas. A generative adversarial network is combined to perform data augmentation in the latent space, and the discriminator-generator game is used to improve sample authenticity. Noise filtering is performed simultaneously, and anomalies in overlapping areas are removed using the edited nearest neighbor algorithm. In the model architecture design, the feature extraction layer uses an improved ResNet-34 as the backbone network, embedding the ECA attention module to enhance feature extraction; a multi-scale fusion layer is constructed, introducing bidirectional cross-layer connections and deformable convolution modules; the hierarchical decision layer integrates a spatial pyramid pooling layer to distinguish between under-brush, qualified, and over-brush quality of concrete; For model training, the backbone network was initialized based on the ImageNet pre-trained weights, the parameters of the first three convolutional layers were frozen, and the high-level modules were fine-tuned using a cosine annealing learning rate. The parameters of the first three convolutional layers were then unfrozen, and Focal Loss was introduced to balance the distribution of undershoot / overshoot samples. γ was set to 2.0 to suppress the weights of easily classified samples. The AdamW optimizer was used for training, and the learning rate was decayed in stages. Early stopping on the validation set was performed every 10 epochs, and a model rollback was triggered when the validation loss did not decrease for three consecutive times.

5. The deep learning-based visual inspection and grading system for concrete roughening quality according to claim 4 is characterized by: The quality level evaluation is: Q=w1H1+w2S1+w3N1 (2) Where Q is the concrete roughening quality evaluation index, w1, w2, w3 are weight coefficients, H1 is the normalized maximum vertical height difference between adjacent peaks and troughs, S1 is the normalized slope between the maximum peak and the maximum trough, and N1 is the normalized number of peaks per unit area. The normalization is performed based on the maximum and minimum reference values ​​of each index. H=max[z p,i -min(z v,i )] (6) Where H is the maximum vertical height difference between adjacent peaks and troughs in a single scan area, mm; S is the slope between the maximum peak and the maximum trough in a single scan area; and N is the number of peaks per unit area, z. p,i -The height of the i-th peak, mm, z v,i -The height of the valley adjacent to the i-th peak, mm, (x mp ,y mp , z mp )-coordinates of the maximum peak, (x mv ,y mv , z mv )-coordinates of the maximum trough, n-number of peaks in a single scan area, s-area of ​​a single scan area; the maximum and minimum reference values ​​are calculated based on historical quality inspection results; when Q≤0.2, the quality level is evaluated as undershoot; when 0.2<Q≤0.8, the quality level is evaluated as qualified; when Q>0.8, the quality level is evaluated as overshoot.

6. The deep learning-based visual inspection and grading system for concrete roughening quality according to claim 1 is characterized by: The spliced ​​original image data adopts the improved SIFT feature matching combined with the RANSAC algorithm, sets the matching point threshold to ≥50 pairs, and the homography matrix reprojection error to ≤1.5 pixels.

Images