Ground penetrating radar steel bar positioning and parameter measuring method based on dual detection network

Abstract

The invention discloses a ground penetrating radar reinforcing steel bar positioning and parameter measuring method based on a dual detection network. The method comprises the following steps: step 1, preprocessing collected reinforcing steel bar data; step 2, intercepting an image with a reinforcing steel bar; step 3, marking a reinforcing steel bar area; step 4, intercepting the marked area; step 5, performing enhancement; step 6, carrying out yolov8 training to obtain two data sets; and step 7, obtaining steel bar target positioning and parameter measurement. According to the method disclosed by the invention, the efficiency of automatic processing is improved by 80% compared with that of manual interpretation, the method is suitable for large-scale engineering detection, the precision is optimized, noise interference is reduced by adopting dynamic region screening, and the accuracy of a dual-detection network in a complex scene is improved to 95% or above; the method expands the application range, supports double-layer reinforcing mesh recognition, and fills the blank in the prior art.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of ground penetrating radar, and particularly relates to a ground penetrating radar steel bar target automatic positioning and parameter measurement method based on a deep learning network, which is suitable for non-destructive detection of the number, spacing and protective layer thickness of steel bars in a concrete structure. BACKGROUND

[0002] In civil engineering detection, ground penetrating radar (GPR) has become the core technology for steel bar positioning due to its non-destructive and high-resolution characteristics. However, the existing ground penetrating radar steel bar positioning technology has the following defects:

[0003] (1) Low efficiency and accuracy: The traditional interpretation of detection results relies on manual experience, and the processing efficiency of massive data is low, and the subjective error is significant.

[0004] (2) Poor adaptability to complex scenes: Steel bar reflection signals are easily disturbed by internal cavities, cracks and electromagnetic noise in concrete, and existing algorithms have insufficient robustness and high misjudgment rate.

[0005] (3) Strong limitations in parameter measurement: The measurement accuracy of protective layer thickness and horizontal spacing is limited by signal attenuation and offset homing algorithms, and the identification of double-layer steel bar mesh is easily disturbed. SUMMARY

[0006] To solve the problems of noise interference, low computing efficiency and poor adaptability to complex scenes, the present application provides a ground penetrating radar steel bar positioning and parameter measurement method based on a double detection network. When there are double-layer steel bar meshes or single steel bars, the automatic extraction of steel bar parameters is easily disturbed, so the present application first detects the steel bar area and then identifies the steel bars in the steel bar area. Through the double target detection network, the steel bar area is first detected, and then the steel bar is further detected in the steel bar area, realizing automatic, high-precision positioning and parameter measurement of the steel bar.

[0007] The present application adopts the following technical solutions:

[0008] A ground penetrating radar steel bar positioning and parameter measurement method based on a double detection network, which is improved in that it comprises the following steps:

[0009] Step 1: Data preprocessing is performed on the steel bar data collected by the ground penetrating radar;

[0010] Step 2: The image with steel bars in the data image is intercepted;

[0011] Step 3: The steel bar area of each image is labeled, and all images with steel bar areas are enhanced to obtain the final steel bar area data set and label;

[0012] Step 4: Cut out the area marked with steel bars in Step 3;

[0013] Step 5: Label the reinforcing bars in the reinforcing bar area extracted in Step 4, and enhance all reinforcing bar areas to obtain the final reinforcing bar dataset and labels.

[0014] Step 6: Train the two datasets obtained in steps 3 and 5 using YOLOv8 to obtain two detection networks;

[0015] Step 7: Detect ground-penetrating radar data using the region selection model, select the rebar region, detect the rebar in the selected rebar region, and obtain the rebar target location and parameter measurement.

[0016] Furthermore, the detection equipment used in step 1 is a ground-penetrating radar main unit and a 900MHz antenna.

[0017] Furthermore, the data preprocessing method in step 1 is automatic gain control, which relies on multiplying the radar data by an automatic gain weighting function that varies with time.

[0018]

[0019] In the above formula, Let P(t) represent the radar data after automatic gain control, P(t) represent the automatic gain weighting function, and y(t) represent the radar data before automatic gain control.

[0020] The number of time windows is determined by the number of parameter control points. When calculating the average amplitude, half a time window should overlap between every two adjacent time windows. The following parameters are calculated based on the number of control points: time window length, start time of time window, and end time of time window.

[0021] The average amplitude A within the s-th time window s The calculation formula is:

[0022]

[0023] In the above formula, q(t) represents the single-channel data acquired by the radar to be processed; T s-1 T represents the start time of the s-th time window; s+1 B represents the end time of the s-th time window; B represents the number of samples in the s-th time window.

[0024] Finally, the gain parameter corresponding to the center of each time window as the control point is stored.

[0025] The weighting function is:

[0026]

[0027] In the above formula, p s Let F represent the weighting factor corresponding to the center of the s-th time window, and let F represent the balance coefficient used to adjust the effective amplitude after processing. For the weighting factors of points other than the center of the time window, they are obtained by linear interpolation using the weighting factors of the centers of two adjacent time windows.

[0028] Furthermore, in step 2, 6000 images are captured, with a data size of 1024*512.

[0029] Furthermore, step 3 specifically includes the following steps:

[0030] Step 31: Use the open-source image annotation tool labelme to annotate the steel reinforcement area targets in the radar image;

[0031] Step 32: Combine the labeled radar images and labeling information to establish a radar image database of underground steel reinforcement areas;

[0032] Step 33: Expand the data volume using data augmentation methods, including horizontally flipping the image and scaling the image;

[0033] Step 34: Randomly allocate the data according to a ratio of 7:2:1 to obtain the final training set, validation set, and test set.

[0034] Furthermore, step 5 specifically includes the following steps:

[0035] Step 51: Use the open-source image annotation tool labelme to annotate the rebar targets in the rebar area;

[0036] Step 52: Combine the marked rebar areas with the marked information to establish an underground rebar radar image database;

[0037] Step 53: Expand the data volume using data augmentation methods, including horizontally flipping the image and scaling the image;

[0038] Step 54: Randomly allocate the data according to a ratio of 7:2:1 to obtain the final training set, validation set, and test set.

[0039] Furthermore, the two detection networks obtained in step 6 are a region selection model and a rebar detection model; the region selection model is based on a YOLOv8-trained region filtering network to locate the rebar region; the rebar detection model performs secondary detection of the rebar within the rebar region and outputs the coordinates of the center point of the upper edge of the rebar detection box.

[0040] Furthermore, step 7 specifically includes the following steps:

[0041] Step 71: First, perform target detection of the steel reinforcement area on the ground penetrating radar image. After detecting the steel reinforcement area, perform steel reinforcement detection within the steel reinforcement area.

[0042] Step 72: Mark the coordinates of the center point of the detected rebar detection frame with a point and return it to the coordinates of the original image. The y-coordinate of the center point is the thickness of the rebar protective layer.

[0043] Step 73: The number of detected rebars is automatically counted based on the number of marking points in Step 72;

[0044] Step 74: Calculate the rebar spacing using the spacing between lanes and the pixel interval between rebars.

[0045] S = L × N

[0046] In the above formula, S is the spacing between steel bars, L is the spacing between lanes, and N is the pixel spacing.

[0047] The beneficial effects of this invention are:

[0048] The method disclosed in this invention firstly improves efficiency, with automated processing increasing efficiency by 80% compared to manual interpretation, making it suitable for large-scale engineering inspections. Secondly, it optimizes accuracy by employing dynamic region filtering to reduce noise interference, and the dual-detection network achieves an accuracy rate of over 95% in complex scenarios. Finally, it expands the application scope by supporting the identification of double-layer rebar mesh, filling a gap in existing technologies. Attached Figure Description

[0049] Figure 1 This is a flowchart illustrating the method of the present invention;

[0050] Figure 2 This is a sample diagram showing the annotations for the rebar area;

[0051] Figure 3 This is a diagram of the extracted rebar area;

[0052] Figure 4 This is a sample diagram showing the markings for reinforcing bars;

[0053] Figure 5 This is a diagram of the test results during the actual testing process. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0055] Example 1 discloses a ground-penetrating radar (GPR) method for locating and measuring the parameters of reinforcing bars based on a dual-detection network. It uses measured data collected by GPR to create a dataset, train a model, and complete target detection of the reinforcing bar area and the reinforcing bars themselves, providing the rebar location and parameter calculation. Figure 1 As shown, it includes the following steps:

[0056] Step 1: Perform data preprocessing and enhancement on the rebar data collected by ground-penetrating radar to obtain data with clear target features:

[0057] The detection equipment used was a ground-penetrating radar main unit and a 900MHz antenna.

[0058] The data preprocessing method is automatic gain control, which relies on multiplying the radar data by a time-varying automatic gain weighting function.

[0059]

[0060] In the above formula, Let P(t) represent the radar data after automatic gain control, P(t) represent the automatic gain weighting function, and y(t) represent the radar data before automatic gain control.

[0061] The number of time windows is determined by the number of parameter control points. When calculating the average amplitude, half a time window should overlap between every two adjacent time windows. The following parameters are calculated based on the number of control points: time window length, start time of time window, and end time of time window.

[0062] The average amplitude A within the s-th time window s The calculation formula is:

[0063]

[0064] In the above formula, q(t) represents the single-channel data acquired by the radar to be processed; T s-1 T represents the start time of the s-th time window; s+1 B represents the end time of the s-th time window; B represents the number of samples in the s-th time window.

[0065] Finally, the gain parameter corresponding to the center of each time window as the control point is stored.

[0066] The weighting function is:

[0067]

[0068] In the above formula, p sLet F represent the weighting factor corresponding to the center of the s-th time window, and let F represent the balance coefficient used to adjust the effective amplitude after processing. For the weighting factors of points other than the center of the time window, they are obtained by linear interpolation using the weighting factors of the centers of two adjacent time windows.

[0069] Step 2: Extract images containing reinforcing bars from the data image. The number of extracted images is 6000, and the data size is 1024*512.

[0070] Step 3: Label the rebar regions in each image, and enhance all images containing rebar regions to obtain the final rebar region dataset and labels. Annotation examples are shown below. Figure 2 As shown;

[0071] Step 31: Use the open-source image annotation tool labelme to annotate the steel reinforcement area targets in the radar image;

[0072] Step 32: Combine the labeled radar images and labeling information to establish a radar image database of underground steel reinforcement areas;

[0073] Step 33: Use data augmentation methods to expand the data volume and increase sample diversity. The data augmentation methods include horizontally flipping images and scaling images.

[0074] Step 34: Randomly allocate the data according to a ratio of 7:2:1 to obtain the final training set, validation set, and test set.

[0075] Step 4: Extract the rebar area marked in Step 3, and take a sample as shown. Figure 3 As shown;

[0076] Step 5: Label the reinforcing bars in the areas extracted in Step 4, and enhance all areas containing reinforcing bars to obtain the final reinforcing bar dataset and labels. Annotation examples are shown below. Figure 4 As shown;

[0077] Step 51: Use the open-source image annotation tool labelme to annotate the rebar targets in the rebar area;

[0078] Step 52: Combine the marked rebar areas with the marked information to establish an underground rebar radar image database;

[0079] Step 53: Use data augmentation methods to expand the data volume and increase sample diversity. The data augmentation methods include horizontally flipping images and scaling images.

[0080] Step 54: Randomly allocate the data according to a ratio of 7:2:1 to obtain the final training set, validation set, and test set.

[0081] Step 6: Train the two datasets obtained in steps 3 and 5 using YOLOv8 to obtain two detection networks, namely the region selection model and the rebar detection model. The region selection model is based on YOLOv8 to train the region filtering network and locate the rebar region. The rebar detection model performs secondary detection of the rebar within the rebar region and outputs the coordinates of the center point of the upper edge of the rebar detection box.

[0082] Step 7: Using a region selection model to detect ground-penetrating radar data, select rebar areas, and perform rebar detection on the selected areas to obtain rebar target location and parameter measurements.

[0083] Step 71: First, perform target detection of the steel reinforcement area on the ground penetrating radar image. After detecting the steel reinforcement area, perform steel reinforcement detection within the steel reinforcement area.

[0084] Step 72: Mark the coordinates of the center point of the detected rebar detection frame with a point and return it to the coordinates of the original image. The detection result is as follows: Figure 5 As shown. The y-coordinate of the center point at the top represents the thickness of the concrete cover for the reinforcing steel.

[0085] Step 73: The number of detected rebars is automatically counted based on the number of marking points in Step 72;

[0086] Step 74: Calculate the rebar spacing using the spacing between lanes and the pixel interval between rebars.

[0087] S = L × N

[0088] In the above formula, S is the spacing between steel bars, L is the spacing between lanes, and N is the pixel spacing.

Claims

1. A method for locating rebar and measuring parameters using ground-penetrating radar based on a dual-detection network, characterized in that, Includes the following steps: Step 1: Perform data preprocessing on the rebar data collected by ground penetrating radar; Step 2: Extract the image containing the reinforcing bars from the data image; Step 3: Label the rebar areas in each image, and enhance all images with rebar areas to obtain the final rebar area dataset and labels. Step 4: Cut out the area marked with steel bars in Step 3; Step 5: Label the reinforcing bars in the reinforcing bar area extracted in Step 4, and enhance all reinforcing bar areas to obtain the final reinforcing bar dataset and labels. Step 6: Train the two datasets obtained in steps 3 and 5 using YOLOv8 to obtain two detection networks; Step 7: Detect ground-penetrating radar data using the region selection model, select the rebar region, detect the rebar in the selected rebar region, and obtain the rebar target location and parameter measurement.

2. The ground-penetrating radar rebar positioning and parameter measurement method based on a dual-detection network according to claim 1, characterized in that: The detection equipment used in step 1 is a ground-penetrating radar main unit and a 900MHz antenna.

3. The ground-penetrating radar rebar positioning and parameter measurement method based on a dual detection network according to claim 1, characterized in that: The data preprocessing method in step 1 is automatic gain control, which is achieved by multiplying the radar data by an automatic gain weighting function that varies with time. In the above formula, Let P(t) represent the radar data after automatic gain control, P(t) represent the automatic gain weighting function, and y(t) represent the radar data before automatic gain control. The number of time windows is determined by the number of parameter control points. When calculating the average amplitude, half a time window should overlap between every two adjacent time windows. The following parameters are calculated based on the number of control points: time window length, start time of time window, and end time of time window. The average amplitude A within the s-th time window s The calculation formula is: In the above formula, q(t) represents the single-channel data acquired by the radar to be processed; T s-1 This represents the start time of the s-th time window; T s+1 B represents the end time of the s-th time window; B represents the number of samples in the s-th time window. Finally, the gain parameter corresponding to the center of each time window as the control point is stored. The weighting function is: In the above formula, p s Let F represent the weighting factor corresponding to the center of the s-th time window, and let F represent the balance coefficient used to adjust the effective amplitude after processing. For the weighting factors of points other than the center of the time window, they are obtained by linear interpolation using the weighting factors of the centers of two adjacent time windows.

4. The ground-penetrating radar rebar positioning and parameter measurement method based on a dual detection network according to claim 1, characterized in that: The number of images captured in step 2 is 6000, and the data size is 1024*512.

5. The ground-penetrating radar rebar positioning and parameter measurement method based on a dual detection network according to claim 1, characterized in that: Step 3 specifically includes the following steps: Step 31: Use the open-source image annotation tool labelme to annotate the steel reinforcement area targets in the radar image; Step 32: Combine the labeled radar images and labeling information to establish a radar image database of underground steel reinforcement areas; Step 33: Expand the data volume using data augmentation methods, including horizontally flipping the image and scaling the image; Step 34: Randomly allocate the data according to a ratio of 7:2:1 to obtain the final training set, validation set, and test set.

6. The ground-penetrating radar rebar positioning and parameter measurement method based on a dual detection network according to claim 1, characterized in that: Step 5 specifically includes the following steps: Step 51: Use the open-source image annotation tool labelme to annotate the rebar targets in the rebar area; Step 52: Combine the marked rebar areas with the marked information to establish an underground rebar radar image database; Step 53: Expand the data volume using data augmentation methods, including horizontally flipping the image and scaling the image; Step 54: Randomly allocate the data according to a ratio of 7:2:1 to obtain the final training set, validation set, and test set.

7. The ground-penetrating radar rebar positioning and parameter measurement method based on a dual-detection network according to claim 1, characterized in that: The two detection networks obtained in step 6 are the region selection model and the rebar detection model. The region selection model is based on the YOLOv8 trained region filtering network to locate the rebar region. The rebar detection model performs secondary detection of the rebar within the rebar region and outputs the coordinates of the center point of the upper edge of the rebar detection box.

8. The ground-penetrating radar rebar positioning and parameter measurement method based on a dual detection network according to claim 7, characterized in that: Step 7 specifically includes the following steps: Step 71: First, perform target detection of the steel reinforcement area on the ground penetrating radar image. After detecting the steel reinforcement area, perform steel reinforcement detection within the steel reinforcement area. Step 72: Mark the coordinates of the center point of the detected rebar detection frame with a point and return it to the coordinates of the original image. The y-coordinate of the center point is the thickness of the rebar protective layer. Step 73: The number of detected rebars is automatically counted based on the number of marking points in Step 72; Step 74: Calculate the rebar spacing using the spacing between lanes and the pixel interval between rebars. S = L × N In the above formula, S is the spacing between steel bars, L is the spacing between lanes, and N is the pixel spacing.

Images