Microwave nondestructive testing device and method for aging of polyethylene pipe and appendages

Abstract

The invention discloses a microwave nondestructive testing device and method for aging of a polyethylene pipe and appendages. The microwave nondestructive testing device comprises a microwave antenna, a multi-view camera, a control module and a display module. The microwave antenna transmits and receives signals to obtain internal structure information, and the multi-view camera captures high-resolution images for auxiliary analysis. The control module realizes multi-degree-of-freedom movement of the device, performs preprocessing, feature extraction and intelligent analysis on acquired signals, identifies aging features such as cracks, bubbles and layering through a deep learning algorithm, and remotely transmits data. And the display module displays the detection result and the aging evaluation report in real time. The device combines microwave imaging and camera images, realizes accurate defect identification, real-time monitoring and data comparison, supports automatic obstacle avoidance and pipeline safety evaluation, improves the detection precision and efficiency, and is suitable for aging detection of various types of pipelines and accessories.

Description

Technical Field

[0001] The invention belongs to the technical field of pipeline detection, and in particular relates to a microwave nondestructive detection device and method for aging of polyethylene pipes and accessories. Background Art

[0002] With the rapid development of the economy, pipeline transportation technology plays a vital role in various industries. Especially in the fields of oil, natural gas and water conservancy, pipelines are the main mode of transportation of fluid media, and their safety and reliability directly affect production and life. However, during the long-term use of pipelines, affected by environmental, medium and physical factors, they often experience problems such as aging, corrosion and blockage, which not only reduces the transportation efficiency of the pipeline, but also may cause safety hazards. Therefore, regular pipeline inspection and maintenance is particularly important.

[0003] In the existing technology, polyethylene (PE) pipes are widely used in various transportation systems due to their excellent corrosion resistance and economy. However, with the extension of service time, the aging problem of polyethylene pipes and their accessories has become more and more prominent. Traditional detection methods, such as visual inspection and liquid penetration testing, are often difficult to accurately assess the degree of aging inside the pipeline, and are highly dependent on the environment, which can easily lead to missed detection and false detection.

[0004] In recent years, microwave nondestructive testing technology has gradually become a research hotspot in the field of pipeline inspection. This technology uses the penetration and reflection characteristics of microwaves on materials to achieve rapid detection of internal defects in pipelines. In practical applications, microwave detection equipment has the advantages of non-contact and real-time monitoring, and can effectively identify problems such as pipeline aging and cracking. However, there are still some shortcomings in the relevant microwave detection devices and methods on the market, such as high equipment cost, complex operation, lack of automation, and poor adaptability to the environment.

[0005] Therefore, in view of the shortcomings of the existing technology, a device and method for detecting the aging of polyethylene pipes and accessories based on microwave non-destructive testing has been developed, which can efficiently and accurately identify the aging state inside the pipeline, realize real-time monitoring and data analysis, improve the accuracy and efficiency of detection, and ensure the safe operation and maintenance of the pipeline. This will have important practical significance and broad application prospects. The device should have functions such as real-time monitoring, automatic identification and data processing. The device has the characteristics of non-contact detection, rapid response and automated operation, can work stably in harsh environments, and significantly improve the accuracy and reliability of detection. At the same time, this technology reduces the reliance on manual intervention, reduces the cost of detection, and improves the safety and efficiency of pipeline maintenance, thereby providing strong support for ensuring the long-term safe operation of the pipeline. This will then promote the further development of pipeline detection technology.

[0006] Technical solution of prior art 1

[0007] The existing ultrasonic testing technology is a mature non-destructive testing method that uses the propagation characteristics of ultrasonic waves in materials to identify internal defects. In the aging detection of polyethylene pipes and their accessories, ultrasonic waves can effectively penetrate the pipe material and return reflected waves to analyze the internal structural changes. Polyethylene materials have good transmission properties for ultrasonic waves, which enables this detection method to accurately evaluate the degree of aging of the pipeline.

[0008] Disadvantages of prior art 1:

[0009] 1. Limited identification of aging: Ultrasonic testing may be difficult to provide quantitative analysis when evaluating the aging of polyethylene pipes and accessories. Although ultrasound can identify some internal defects, it cannot accurately measure the microscopic aging characteristics of the material (such as polymer chain breakage and changes in physical properties), resulting in an inability to fully evaluate the overall performance degradation of the material.

[0010] 2. Sensitive to environmental conditions: Ultrasonic testing is greatly affected by environmental factors. Under adverse conditions such as high humidity or low temperature, the efficiency of sound wave propagation may be reduced, affecting the accuracy of detection. In addition, the complexity of the on-site environment, such as debris or vibration around the pipeline, may also interfere with the propagation of sound waves, resulting in misjudgment.

[0011] 3. Difficulty in data interpretation: The data generated by ultrasonic testing often requires professional interpretation. Due to the complexity of ultrasonic signals, identifying and analyzing reflected waveforms may require a high level of expertise and experience, increasing the complexity and time cost of the testing process. In addition, the lack of standardized analysis methods may lead to inconsistent results between different technicians.

[0012] 4. Slow detection speed: Ultrasonic testing is usually time-consuming due to the need for segment-by-segment detection and data analysis, especially in large-scale pipeline systems, and cannot meet the needs of real-time monitoring. This limits its application in daily maintenance and monitoring, which may lead to delayed discovery of aging problems.

[0013] 5. High equipment cost: High-performance ultrasonic testing equipment is often expensive, and its maintenance and calibration requirements are high, which increases the overall testing cost. This may become a factor hindering its widespread application for some projects with limited budgets.

[0014] The solution of the second prior art:

[0015] The existing infrared thermal imaging method is an effective non-destructive testing technology that identifies potential aging problems by measuring the temperature distribution on the surface of the material. In the aging inspection of polyethylene pipes and their accessories, infrared thermal imaging can capture the temperature changes caused by material aging, thereby evaluating the health status of the pipeline.

[0016] Disadvantages of the second prior art:

[0017] 1. Temperature distribution is affected by many factors: Infrared thermal imaging relies on the temperature distribution on the surface of the material to assess aging, but this temperature distribution may be affected by many factors, such as ambient temperature, sunlight exposure, wind speed, etc. These external conditions may lead to inaccurate test results, making it difficult to accurately judge the actual aging state of polyethylene pipes.

[0018] 2. High requirements for surface conditions: Infrared thermal imaging has certain requirements for the cleanliness and material properties of the pipeline surface. If there is dirt, grease or other coverings on the pipeline surface, it may affect the emission and reception of thermal radiation, resulting in distortion of thermal imaging results and reducing the effectiveness of detection.

[0019] 3. Unable to detect deep defects: Infrared thermal imaging mainly detects surface temperature changes and cannot directly identify defects or aging conditions inside the pipeline. If aging or damage occurs inside the material, infrared thermal imaging may not be able to detect it in time, resulting in the failure to effectively assess potential risks.

[0020] 4. Detection accuracy depends on equipment performance: The detection accuracy of infrared thermal imaging is highly dependent on the resolution and sensitivity of the equipment used. Low-quality thermal imagers may not be able to accurately capture small temperature changes, thus affecting the accuracy of aging assessment.

[0021] 5. Complex data analysis: The data generated by infrared thermal imaging needs to be professionally analyzed and interpreted, especially in complex piping systems. Identifying and determining the cause of temperature changes may require high expertise and experience, which increases the complexity of the detection process.

[0022] The solution of prior art three:

[0023] The existing liquid penetrant testing method is a commonly used non-destructive testing technology that uses the penetration characteristics of liquid penetrants between the surface and internal defects of materials to identify potential aging problems. In the aging inspection of polyethylene pipes and their accessories, the liquid penetrant testing method can effectively identify cracks, pores or other defects caused by aging.

[0024] Disadvantages of the prior art three:

[0025] 1. Limitations of penetrant selection: The effectiveness of liquid penetrant testing depends on the properties of the selected penetrant. For some polyethylene materials, there may be unsuitable penetrants, resulting in poor penetration and failure to fully identify aging or internal defects. In addition, different types of penetrants have different penetration capabilities for different defects and may not be able to detect all types of defects.

[0026] 2. High requirements for surface cleanliness: Liquid penetrant testing has high requirements for the cleanliness of the pipeline surface. Surface dirt, grease or oxides may hinder the effective penetration of the penetrant, resulting in inaccurate test results. In actual applications, the complex environment of the pipeline surface often affects the penetration effect of the penetrant and reduces the reliability of the test.

[0027] 3. Long drying time: Liquid penetrant testing requires a certain amount of time for the penetrant to fully penetrate the defect, which may make the overall testing process more time-consuming, especially under field conditions, and the need for rapid testing cannot be met.

[0028] 4. Safety hazards during the testing process: Liquid penetrants usually contain chemical ingredients that may have an impact on the environment and the health of operators. Safety regulations need to be strictly followed during handling and use to avoid potential chemical leaks or exposure.

[0029] 5. Interpretation of results depends on experience: The results of liquid penetrant testing often need to be interpreted by professionals, especially when complex defects or multiple defects exist. Analysis and judgment may require rich experience and expertise, which increases the uncertainty of detection.

[0030] 6. Limited applicability: Liquid penetrant testing is mainly applicable to surface defects, and has poor detection effect on deep or internal defects, and cannot fully evaluate the overall aging status of polyethylene pipes and accessories.

[0031] Application number CN202322805607.4, the name of the utility model is A microwave system for detecting defects in polyethylene pipe hot-melt joints, which discloses a clamp, an encoder, a single-chip microcomputer, an open waveguide, a microwave analyzer and a computer. The clamp includes a clamp block and a roller, a notch is provided in the middle of the clamp block, the open waveguide is inserted in the notch, the open waveguide is connected to the microwave analyzer through a coaxial line, and the microwave analyzer is connected to the computer through a network cable; the roller is arranged below the clamp block, the encoder is arranged on one side of the roller, the encoder is connected to the single-chip microcomputer, and the single-chip microcomputer is connected to the computer through a USB cable.

[0032] The defects are:

[0033] 1. The equipment installation is complex and the cost is high: the existing equipment of the microwave open waveguide method requires precise control of the lift-off distance, and the detection process is highly dependent on the scanning platform. The installation is complex and the cost is high, which makes it difficult to meet the needs of on-site detection.

[0034] 2. False detection and missed detection problems: Changes in the lift-off distance will cause signal errors, which may mask the signal changes caused by defects and lead to missed detections. At the same time, the signal changes caused by the lift-off distance may be misjudged as defects, leading to false detections.

[0035] CN201910519821.9, Invention Name A pipeline surface corrosion defect detection method and device based on microwave nondestructive testing, the method includes collecting microwave signals reflected from the surface of the tested pipeline, using microwave imaging technology to perform imaging processing on the reflection coefficient phase value, performing edge detection processing on the imaging result, obtaining defect position and width information, building a defect detection model, and calculating defect depth information. The invention uses the microwave reflection coefficient phase value at different detection positions as imaging pixel points, and identifies the boundaries at the same time, so as to obtain the defect position and width information. At the same time, according to the relationship model between the reflection coefficient phase difference and the defect size, the quantitative information of the defect depth is inversely solved, so as to realize the quantitative detection of the position, width and depth of defects on the pipeline.

[0036] The defects are:

[0037] 1. Insufficient imaging resolution and accuracy: The existing technology uses the Canny operator for edge detection. Although it can determine the location and width of defects, it still has the problem of insufficient accuracy in detecting tiny defects. It is also sensitive to noise, which may cause blurred boundaries.

[0038] 2. Poor depth detection accuracy: The existing technology that estimates the defect depth through a simple linear relationship between the defect depth and the phase difference significantly reduces the accuracy of depth measurement when dealing with complex defect morphology or shallow defects.

[0039] 3. Poor environmental adaptability: The existing technology lacks the ability to automatically compensate for changes in the distance between the probe and the pipe surface, and the detection results are easily affected by probe position deviation and changes in the pipe surface state.

[0040] 4. Insufficient real-time detection capabilities: Existing methods rely on fixed scanning platforms and vector network analyzers. The equipment is bulky and has poor portability, making it difficult to adapt to real-time detection needs under complex pipeline structures. Summary of the invention

[0041] The purpose of the present invention is to solve the defects of the above-mentioned prior art. In view of the problems existing in the existing polyethylene pipe and accessory aging detection technology, namely the deficiencies of traditional methods in detection accuracy, real-time and adaptability, a microwave non-destructive detection device and method for the aging of polyethylene pipes and accessories are designed. The device adopts advanced microwave detection technology, which can effectively penetrate polyethylene materials, accurately identify aging defects inside the pipeline, and monitor them in real time. Through the automated data acquisition and analysis system, the present invention realizes the rapid assessment of the aging status and significantly improves the detection efficiency. In addition, the device has strong environmental adaptability and can work stably under complex field conditions.

[0042] The present invention adopts the following technical solution:

[0043] A microwave nondestructive testing device for aging of polyethylene pipes and their accessories comprises a travel module, a signal module, a control module and a microwave antenna module.

[0044] The walking module includes a fixed part and a moving part.

[0045] The fixing components are a plurality of pipe clamps which are wrapped around the polyethylene pipe, and any two pipe clamps are connected by clamp connection fittings.

[0046] The moving parts include a rotary motion mechanism, an axial motion mechanism and a height adjustment mechanism.

[0047] The rotary motion mechanism includes a rotary servo motor, the output shaft of the rotary servo motor is connected to the rotary motor transmission shaft, a rotary motion magnetic wheel is installed on the rotary motor transmission shaft, the rotary motion magnetic wheel engages the clamp, and the magnetic wheel contacts the clamp through magnetic adsorption force to realize the circumferential rotation of the device.

[0048] The axial motion mechanism includes an axial guide rail, an axial bracket, an axial servo motor and an axial slider. The axial servo motor drives the axial slider to drive the height bracket to move axially through the axial guide rail installed in the axial bracket. The axial slider is connected to the height adjustment mechanism to realize the axial motion adjustment of the device.

[0049] The height adjustment mechanism includes a height guide rail, a height slider, a height bracket and a height servo motor. The output shaft of the height servo motor is connected to the internal height guide rail. The height guide rail drives the height slider to slide along the height guide rail, so that the microwave antenna and the multi-eye camera are installed on the height adjustment mechanism through the height slider to realize the height adjustment function of the microwave antenna and the multi-eye camera.

[0050] The high-resolution surface images of the polyethylene pipe and its accessories are collected by a multi-camera; the multi-camera is connected to the signal module and the control module, and is integrated with the detection data of the microwave antenna module to improve the accuracy of defect recognition. The control module receives the data of the signal module and controls the operation of the moving parts. At the same time, the microwave detection results are combined with the data of the image acquisition module, and the aging assessment report is generated after processing by the data processing module. The report includes the aging location, degree, hazard level and treatment suggestions.

[0051] The signal module is connected to the base station and server through wireless communication, and the detection data can be remotely transmitted and monitored. The base station is responsible for real-time reception and transmission of signals, and the server is responsible for storage and further data analysis and processing.

[0052] A method for microwave nondestructive testing of aging of polyethylene pipes and accessories, comprising the following steps:

[0053] Step 1: After installing and fixing the microwave nondestructive testing device, perform a full-scale scan of the polyethylene pipe and its accessories, collect microwave signals, and preliminarily filter out noise and interference. Multi-camera assists in locating the inspection area, ensuring that the microwave antenna maintains a constant distance from the pipeline surface, while avoiding interference areas that cannot be detected, such as tee joints and energized parts of electric fusion joints.

[0054] Step 2: After receiving the microwave signal, the control module uses a hybrid denoising algorithm to eliminate environmental interference and combines it with an improved inverse scattering algorithm to optimize signal quality. The processed data is converted into a two-dimensional or three-dimensional microwave image through an image reconstruction algorithm, which clearly displays the aging characteristics and defect distribution of the pipeline and its accessories.

[0055] Step 3: Based on the microwave imaging, the system extracts aging features (such as cracks, degradation, bubbles, delamination, etc.), and uses a multi-camera to capture high-resolution surface images, which are compared and analyzed with the microwave images. Combining machine learning algorithms and historical sample data to classify defects, the system further optimizes recognition accuracy and adapts to a variety of pipeline aging features through deep learning models.

[0056] Step 4: The base station receives the equipment data and transmits it wirelessly to the server. The server completes denoising, feature extraction and defect assessment, and automatically generates an assessment report, including aging location, degree, hazard level and treatment suggestions. The assessment results support pipeline maintenance decisions, and the inspection history archives are stored in the database for long-term monitoring and data comparison and analysis.

[0057] Further: Step 1 installation includes first installing the support frame of the detection device on the polyethylene pipe to be tested, and adapting to pipes of different diameters through adjustable clamps to ensure the stability of the device. And adjusting the contact pressure between the device and the pipe surface by rotating the locking clamp to ensure that the device is stable and does not slip.

[0058] Next, the detection device is fixed on the fixture by magnetic wheel adsorption. The drive device is controlled by a servo motor and rotates along the pipeline in all directions to ensure a 360-degree scanning of the pipeline without blind spots. The system is equipped with radial and height guides for precise adjustment of the position of the microwave antenna to ensure that the antenna is always aimed at the detection area of ​​the pipeline.

[0059] When the microwave antenna is located in the area to be detected, the control device sends a command to start the entire device to perform a full-circle scan. During the scanning process, the control device emits a microwave signal, which penetrates the pipeline and reflects back to the receiver. After being pre-processed by the signal acquisition unit, the microwave signal is transmitted to the control device for further processing. At the same time, the data is transmitted to the remote monitoring center in real time to facilitate remote monitoring and data analysis.

[0060] Furthermore, the scanning path adopts segmented circumferential scanning: First, the microwave antenna starts from the initial position, records the initial coordinates, and then scans along the circumference of the pipeline. After each scan, the platform controls the antenna to move to the right by a preset distance and continues to the next scan. This process continues until the entire area to be inspected is covered. After the scan is completed, the system generates a complete inspection report and can export data or generate a defect distribution map for further analysis.

[0061] During the inspection process, the system can automatically detect and avoid interference areas such as areas that cannot be scanned by tee joints and areas where the energized joints of electric fusion joints are connected. After the multi-eye camera monitors the position of the energized joints in real time, it sends a signal to notify the control module to adjust the operating status of the equipment to ensure that the microwave antenna does not contact the relevant area, while ensuring the continuity and integrity of the microwave signal. When the microwave antenna approaches the relevant area, the system automatically controls the vertical or horizontal position of the microwave antenna, avoids obstacles by adjusting the sliding platform of the guide rail, and ensures the stability of the inspection process.

[0062] Furthermore, it also includes automatic obstacle avoidance function, which includes:

[0063] The current environment state S is generated through image processing technology and the measurement results of multiple cameras. t , which includes information such as pipes, obstacles, joints, and powered joints.

[0064] State space S t It can be expressed as:

[0065] S t = {P probe ,P obstacle ,θ scan ,V probe} (1)

[0066] in:

[0067] -P probe is the current antenna position;

[0068] -P obstacle The location information of obstacles (such as electric fusion joints, energized joints, etc.);

[0069] -θ scan is the scanning angle of the current antenna;

[0070] -V probe is the current antenna speed.

[0071] Reinforcement learning uses Q value to evaluate the quality of taking an action in a specific state. The update formula of Q value function is:

[0072]

[0073] in:

[0074] -Q(s t , a t ) is the state s t Take action a t Q value;

[0075] -r t+1 It is the reward of the present moment;

[0076] -γ is the discount factor, which controls the impact of future rewards;

[0077] -α is the learning rate, which controls the degree of integration of new and old information during the learning process;

[0078] - is the next state s t+1 The maximum Q value under .

[0079] The Q-value function is approximated by a deep neural network, with the input being the state s t , the output is the Q value of each possible action:

[0080]

[0081] Where θ is the parameter of the neural network, which is updated by optimizing the loss function. The training goal of the deep Q network is to minimize the following loss function:

[0082]

[0083] where θ - are the parameters of the target network, which is used to maintain stability and avoid introducing high variance during training.

[0084] In order to further optimize the path planning, when encountering complex obstacles and dynamic environments, the present invention adopts a combination strategy of the A* algorithm and the Dijkstra algorithm. These two algorithms are used to calculate and select the optimal path.

[0085] A* algorithm: It uses a heuristic method to calculate the shortest path from the starting point to the end point, taking into account the estimated distance h(n) to the target and the cost g(n) of the current path. The total cost function is:

[0086] f(n)=g(n)+h(n) (5)

[0087] in:

[0088] -g(n) is the actual cost from the starting point to the current node n;

[0089] -h(n) is the estimated cost from node n to the target node, usually using Euclidean distance or Manhattan distance.

[0090] Dijkstra algorithm: It is suitable for finding the shortest path in a graph. The shortest path cost of each node is considered during the calculation process. Its core formula is:

[0091] d(u)=min(d(u),d(υ)+w(υ,u)) (6)

[0092] in:

[0093] -d(u) is the shortest path cost from the source node to node u;

[0094] -w(υ,u) is the path weight (cost) from node v to node u.

[0095] When the system detects a new obstacle (such as a live joint on an electric fusion joint, a tee area, etc.), the DQN model will t Dynamically adjust the path. The specific path adjustment can be achieved through the following real-time obstacle avoidance control strategies:

[0096] Real-time obstacle avoidance: When the antenna approaches an obstacle, the system current Calculate the shortest path d from the obstacle min , and update the path to avoid obstacles. i is the position of the ith obstacle, then the path adjustment target is:

[0097]

[0098] in:

[0099] -||P(t)-O i || represents the path point P(t) and the obstacle O i distance.

[0100] When the obstacle changes or the path is blocked, the system will re-plan the path through the A* algorithm or Dijkstra algorithm to ensure that the obstacle is avoided and the scanning task can be continued.

[0101] A further step 2 comprises,

[0102] Furthermore, the image reconstruction algorithm is a convolutional neural network (CNN) and a sparse representation microwave imaging algorithm, specifically including:

[0103] The inverse scattering algorithm often uses the following formula to express the reconstruction process of the reflected signal:

[0104]

[0105] in:

[0106] -f(x,y) is the reconstructed image intensity;

[0107] -k is the microwave wave number;

[0108] -r and r′ are the distance from the antenna to the target and from the target to the receiver, respectively;

[0109] - A(x′, y′) is the scattering coefficient of the target;

[0110] -S is the scatterer surface area.

[0111] CNN gradually extracts high-order features of the image through multiple convolutional layers, pooling layers, and fully connected layers, and finally reconstructs the denoised image through the deconvolution layer.

[0112] 1. Input layer: The input image is a noisy microwave imaging image with a size of N×M;

[0113] 2. Convolutional layer: Multi-layer convolutional layers are used to extract local features from images. The convolution kernel size is usually 3×3 or 5×5.

[0114] 3. Pooling layer: The pooling layer reduces the size of the image by downsampling while retaining the main features of the image;

[0115] 4. Fully connected layer: Through the fully connected layer, the network can capture global features and perform high-level feature fusion;

[0116] 5. Output layer: output the denoised image.

[0117] The loss function of convolutional neural network denoising uses mean square error (MSE):

[0118]

[0119] in:

[0120] - It is the image after CNN denoising;

[0121] -I clean，i is the corresponding clean image;

[0122] - N is the number of pixels in the image.

[0123] The sparse representation of the image is solved by optimizing the problem. The mathematical formula of sparse representation is:

[0124] I=DX (10)

[0125] in:

[0126] -I is a noisy image;

[0127] -D is a dictionary matrix consisting of a set of basis functions;

[0128] -X is a sparse coefficient matrix representing the weighting coefficients of each basis function in the image.

[0129] The goal of denoising is to restore the clean part of the image by solving the sparse coefficient X. The optimization problem can be expressed by the following formula:

[0130]

[0131] in:

[0132] - is the image reconstruction error;

[0133] -||X||1 is a sparsity constraint, which promotes sparse representation;

[0134] -λ is a hyperparameter that adjusts the trade-off between sparsity and reconstruction error.

[0135] The denoising algorithm combining CNN and sparse representation can effectively solve the balance problem between noise and detail loss in microwave imaging images and obtain clearer imaging results.

[0136] Furthermore, the machine learning algorithm in step 3 is an improved YOLOv11 microwave image defect recognition algorithm, which specifically includes:

[0137] The improved YOLOv11 model uses a more efficient convolutional neural network to extract features. The microwave imaging image is convolved using the following formula to extract features:

[0138] f out =f in *W+b (12)

[0139] in:

[0140] -f in is the feature map of the input image (microwave image).

[0141] -W is the convolution kernel weight and b is the bias.

[0142] -f out is the output feature map.

[0143] Through multi-scale feature fusion, YOLOv11 fuses feature maps at different levels to enhance the model's ability to detect defects of different sizes and scales. The core formula for feature fusion is as follows:

[0144]

[0145] in:

[0146] -F i are feature maps of different scales.

[0147] -α i The weight of each scale feature map is automatically learned by the network.

[0148] For each grid, the YOLOv11 network outputs a fixed number of bounding box candidates and predicts confidence and class probability for each bounding box:

[0149] Prediction={x,y,w,h,c,p1,p2,..,p C} (14)

[0150] in:

[0151] - (x, y) is the center coordinate of the bounding box;

[0152] - (w, h) is the width and height of the bounding box;

[0153] -c is the confidence of the bounding box;

[0154] -p1,p2,...,p C is the probability that the box belongs to each category.

[0155] A more detailed loss function design is introduced to adapt to the characteristics of microwave images. The loss function consists of the following parts:

[0156] L=L bbox +L obj +L class +L IOU (15)

[0157] in:

[0158] -L bbox It is the bounding box regression loss, which measures the difference between the predicted box and the true box. The mean square error (MSE) is usually used as the calculation method:

[0159]

[0160] Where (x i ,y i , w i ,h i ) is the coordinate of the real frame, are the coordinates of the predicted box.

[0161] -L obj * is the target existence loss, which measures whether the box contains the target.

[0162]

[0163] -L class is the classification loss, which measures the accuracy of the target category prediction:

[0164]

[0165] -L IOU The intersection over union loss is used to improve the overlap between the bounding box and the ground truth box. Calculate the intersection over union (IoU) and optimize:

[0166] L IOU =1-IoU (19)

[0167] To avoid detecting the same defect multiple times, YOLOv11 uses non-maximum suppression (NMS) to filter multiple bounding boxes and only retain the bounding box with the highest prediction confidence. NMS is performed using the following formula:

[0168]

[0169] When the IoU is greater than the set threshold, the bounding boxes with low confidence are removed and only the optimal box is retained.

[0170] Beneficial effects:

[0171] By combining microwave imaging and improved YOLOv11 defect recognition algorithm, the present invention can accurately identify the aging characteristics of polyethylene pipes and accessories, including cracks, bubbles, delamination and material degradation. The penetration of microwave signals and the auxiliary visual ability of multi-eye cameras are used to enhance the accuracy of identifying subtle defects. Through the fully automatic microwave nondestructive testing device and the intelligent defect recognition system, the pipeline can be scanned all around efficiently and in real time, and defects can be automatically identified and classified, which significantly improves the detection efficiency and reduces the error of manual operation. The present invention effectively fuses the microwave imaging and the image data obtained by the multi-eye camera, and conducts comprehensive analysis through deep learning and machine learning algorithms, further improving the comprehensiveness and accuracy of defect detection. The introduction of multi-dimensional perspective solves the problem of details that are difficult to identify in microwave imaging, such as microcracks and surface roughness changes caused by aging. The present invention supports the detection of different types and different aging degrees of pipelines and has a high degree of adaptability. Through the continuously optimized deep learning model, the system can continuously improve the recognition ability of new aging phenomena to meet various application requirements. The system supports remote data transmission and real-time monitoring. The test results can be fed back to the control center in real time, which is convenient for real-time evaluation and decision support of pipeline health status. The generation and storage of test history archives also provide data basis for long-term pipeline maintenance. Due to the high degree of automation and intelligence of the test process, the present invention significantly reduces the need for manual inspection and operation and maintenance costs. Combined with the reports generated by the system and the defect assessment, it provides a scientific basis for the repair and maintenance of the pipeline and prolongs the service life of the pipeline. BRIEF DESCRIPTION OF THE DRAWINGS

[0172] Figure 1 It is an overall structural diagram of a microwave nondestructive testing device for aging of polyethylene pipes and accessories of the present invention;

[0173] Figure 2 A schematic diagram of a microwave nondestructive testing device for aging of polyethylene pipes and accessories of the present invention for testing electric fusion joints;

[0174] Figure 3 A schematic diagram of a microwave nondestructive testing device for aging of polyethylene pipes and accessories of the present invention for testing a three-way joint;

[0175] Figure 4 It is a schematic diagram of the circumferential rotation motion of the present invention;

[0176] Figure 5 It is a schematic diagram of the structure of the axial motion mechanism of the present invention;

[0177] Figure 6 It is a schematic diagram of the structure of the axial height adjustment mechanism of the present invention;

[0178] Figure 7 It is a schematic diagram of long-distance signal transmission of the present invention;

[0179] Figure 8 This is a flow chart of the improved YOLOv11 defect recognition algorithm of the present invention;

[0180] Fig. 9 It is a simulation detection modeling diagram of the present invention;

[0181] Fig.10 It is a simulation detection data diagram of the present invention.

[0182] In the figure: 1-PE pipe, 2-clamp, 3-hot-melt joint, 4-rotation servo motor, 5-rotation motor drive shaft, 6-magnetic wheel, 7-axial servo motor, 8-control module, 9-axial bracket, 10-axial guide rail, 11-height servo motor, 12-height bracket, 13-height guide rail, 14-microwave antenna, 15-multi-eye camera, 16-display module, 17-antenna, 18-electric fusion joint, 19-electric fusion joint shell, 20-electric fusion joint internal resistance wire, 21-electric fusion joint power joint I, 22-electric fusion joint power joint II, 23-tee joint, 24-other PE pipes, 25-detection device, 26-base station, 27-server, 28-fixing screw I, 29-fixing screw II, 30-fixing screw III, 31-axial slider, 32-height slider. DETAILED DESCRIPTION

[0183] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention is described clearly and completely below. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0184] In view of the problems existing in the existing polyethylene pipe and its accessories aging detection technology, namely the deficiencies of the traditional methods in detection accuracy, real-time and adaptability, the present invention designs a microwave nondestructive detection device and method for the aging of polyethylene pipes and their accessories. The device adopts advanced microwave detection technology, which can effectively penetrate polyethylene materials, accurately identify aging defects inside the pipeline, and monitor them in real time. Through the automated data acquisition and analysis system, the present invention realizes the rapid assessment of the aging status and significantly improves the detection efficiency. In addition, the device has strong environmental adaptability and can work stably under complex field conditions.

[0185] like Figure 1-Figure 7 As shown, a microwave nondestructive detection device for PE pipe electric fusion joints of the present invention includes a motion module, a signal module and a control module.

[0186] The motion module includes a rotary servo motor 4, a rotary motor transmission shaft 5, a magnetic wheel 6, an axial servo motor 7, an axial bracket 9, an axial guide rail 10, a height servo motor 11, a height bracket 12, a height guide rail 13, a fixing screw I 28, a fixing screw II 29, a fixing screw III 30, 31-axial slider, 32-height slider.

[0187] In the present invention, the connection of the PE pipe 1 is a PE pipe hot-melt joint 2 or a PE pipe electric fusion joint 18 or a three-way joint 23. The PE pipe electric fusion joint 18 has an internal resistance wire 20 of the PE pipe electric fusion joint. The outside of the PE pipe electric fusion joint 18 is an outer shell 19 of the PE pipe electric fusion joint. The outer shell 19 of the PE pipe electric fusion joint has an energized joint I 21 and an energized joint II 22 of the PE pipe electric fusion joint.

[0188] The pipe clamp 2 is wrapped around the outer periphery of the PE pipe 1, and the pipe clamp 2 is connected to each other by fixing screws I 28, fixing screws II 29, and fixing screws III 30. Of course, other numbers of pipe clamps and fixing screws can be set according to the outer diameter of the PE pipe 1, and there is no limit on the number. The pipe clamp 2 is used to fix the PE pipe and other components to avoid any form of displacement during the detection or connection process. This helps to improve the detection accuracy, provide the necessary stability in hot welding or other operations, ensure the firmness and reliability of the connection, and ensure that the PE pipe 1 will not loosen or fall off during the detection process.

[0189] The rotary motion mechanism includes a rotary servo motor 4, the output shaft of the rotary servo motor is connected to the rotary motor transmission shaft 5, a rotary motion magnetic wheel 6 is installed on the rotary motor transmission shaft 5, the rotary motion magnetic wheel 6 engages the clamp 2, and the magnetic wheel 6 contacts the clamp 2 through magnetic adsorption force to realize the circumferential rotation of the device.

[0190] The axial motion mechanism includes an axial guide rail 10, an axial bracket 9, an axial servo motor 11 and an axial slider 31. The axial servo motor 11 drives the axial slider 32 through the axial guide rail 10 installed in the axial bracket 9 to drive the height bracket 12 to move axially. The axial slider is connected to the height adjustment mechanism to realize the axial motion adjustment of the device.

[0191] The control module 8 sends a control signal to the axial servo motor 7, instructing it to start or adjust the moving speed and direction, and set the specific parameters of the axial movement (such as distance, speed, acceleration, etc.). The axial servo motor 7 starts to rotate, driving the transmission shaft connected to it, and converts the rotational motion into linear motion along the guide rail. This driving force causes the axial slider 31 installed on the axial guide rail 10 to start moving along the preset direction. The axial slider 31 slides smoothly along the axial guide rail 10 to accurately control the axial position of the detection device. The guide rail provides a smooth and stable sliding path to ensure that the detection device can achieve continuous and vibration-free movement along the axis of the pipeline. When the slider reaches the target position, the control module 8 sends a stop signal to the axial servo motor 7, and the axial slider 31 stops at the set position. In this process, the control module 8 also locks the slider in this position to ensure the stability and positioning accuracy of the slider during the detection process. During the movement process, the control module 8 monitors the current position of the slider in real time and feeds back the data. The control module 8 dynamically adjusts according to the difference between the actual position and the preset position to ensure that the slider reaches the precise axial target position.

[0192] The height adjustment mechanism includes a height guide rail 13, a height slider 32, a height bracket 12 and a height servo motor 11. The output shaft of the height servo motor 11 is connected to the internal height guide rail 13. The height guide rail 13 drives the height slider 32 to slide along the height guide rail 13, so that the microwave antenna 14 and the multi-eye camera 15 are installed on the height adjustment mechanism through the height slider 32 to realize the height adjustment function of the microwave antenna 14 and the multi-eye camera 15.

[0193] The control module 8 sends a control signal to the height servo motor 11, instructing it to start and set parameters such as the target height and speed of movement. The height servo motor 11 starts to run, converting the rotational motion into a linear motion in the vertical direction. The driving force of the motor acts on the height slider 32 installed on the height guide rail 13 through the transmission system, causing it to start moving up and down. The height slider 32 slides smoothly on the height guide rail 13 to achieve the height adjustment of the detection device in the vertical direction. The guide rail ensures the stability and precise positioning of the slider 32 during the height adjustment process. During the movement of the height slider 32, the system monitors the current position of the slider in real time through the sensor, and feeds back the position data to the control module 8. The control module 8 dynamically adjusts the movement of the height servo motor 11 according to the deviation between the current position and the preset target position to ensure that the slider reaches the target height accurately. When the height slider 32 reaches the specified height, the control module 8 sends a stop signal to the height servo motor 11, and the height slider 32 immediately stops at the height position. At this time, the control module 8 also locks the slider at the height position to ensure that the height of the device remains stable during the detection process.

[0194] The laser locator 6 is fixed on the microwave antenna 7, the control signal positioning module 13 is installed on the laser locator 6, the control module 120 is installed in the control signal positioning module 13, the laser locator 6 is connected to the control module 120 in the control signal positioning module 13, and the microwave antenna 7 is connected to the signal module 110. The signal module 110 is connected to the control module 120.

[0195] Rotary servo motor 4: used to provide rotational power to control the rotational movement of the relevant parts of the detection device, so that the detection system can be evaluated at multiple angles to fully understand the status of the pipeline.

[0196] Rotating motor transmission shaft 5: used to effectively transmit the power generated by the rotating servo motor to the detection device, ensure stable rotation movement, improve the accuracy of detection, and avoid errors caused by unstable motion.

[0197] Magnetic wheel 6: used to fix and guide the detection device in the equipment, and provide the necessary stability during rotation or movement. Through the action of magnetic force, it ensures that the position of the detection device remains unchanged during the working process, improving the accuracy of the overall operation.

[0198] Axial servo motor 7: used to control the axial movement of the device in the PE pipe, and can accurately adjust the position of the detection head. This allows the detection process to adapt to the size and structure of different pipes, and improves the comprehensiveness and accuracy of the detection.

[0199] Control module 8: used to coordinate and control the work of various components, receive signals from sensors and issue instructions to ensure the synchronous operation of various components, thereby improving the efficiency and response speed of the overall system.

[0200] Axial bracket 9: used to provide necessary support for the axial servo motor, ensuring that the motor and the drive shaft remain stable during operation, and can effectively prevent the equipment from position deviation due to vibration or impact.

[0201] Axial guide rail 10: It is used to provide guidance for the axial movement of the equipment. By reducing friction and improving the smoothness of movement, the guide rail can ensure the accuracy of the equipment during the inspection process, so that each adjustment can reach the expected position.

[0202] Height servo motor 11: used to control the height adjustment of the equipment to ensure that it can adapt to the detection of pipes or joints of different heights. This flexible height adjustment function allows the detection system to be used in a variety of occasions to meet the needs of different users.

[0203] Height bracket 12: used to provide stable support for the height servo motor. It ensures that the motor does not move or shake during operation, thereby ensuring the accuracy and reliability of height adjustment, and is an important part of the stability of the entire system.

[0204] Height guide rail 13: used to provide necessary guidance for height adjustment of the equipment, ensuring the accuracy of movement during the adjustment process. By reducing unnecessary friction and resistance, the guide rail can ensure smooth movement of the equipment between different heights, improving the overall detection efficiency.

[0205] Microwave antenna 14: used to receive and send microwave signals, and is a key component for realizing microwave nondestructive testing. By ensuring good signal transmission, the antenna can obtain necessary data, provide a basis for subsequent analysis, and ensure the accuracy of the test results.

[0206] Multi-view camera 15: It is used to capture images from multiple angles, which can assist in visualization and data recording during the inspection process. By acquiring real-time images from different angles, operators can have a more comprehensive understanding of the pipeline status and identify potential problems in a timely manner.

[0207] Display module 16: used to display the test results, images and other relevant information in real time, providing an intuitive monitoring interface for operators. Real-time feedback helps to improve the transparency of the test process, ensure that operators can respond in time, and improve work efficiency.

[0208] Antenna 17: used to receive and send signals so that the display module can display the detection results and other contents in real time.

[0209] Electric fusion joint 18: The electric fusion joint connects the PE pipe by heating with electric current, and is used to ensure the close connection and air tightness of the pipe.

[0210] The outer shell of the electric fusion joint 19 is used to provide protection for the internal structure to prevent it from being damaged by the external environment. The outer shell also has a certain insulation performance to ensure the safety of the electric fusion joint during use.

[0211] Resistance wire 20 inside the electric fusion joint: The resistance wire inside the electric fusion joint is used to generate heat when power is applied to promote the fusion of the joint.

[0212] Electric fusion joint power connector Ⅰ21, electric fusion joint power connector Ⅱ22: used to connect the power supply to the electric fusion joint, ensure that the current can pass smoothly, perform necessary heating operations, and thus achieve a good connection effect.

[0213] Tee connector 23: used to connect three pipelines, allowing multiple pipelines to converge at one point, facilitating the efficient transportation of gas or fluid. It can provide flexibility in a variety of pipeline configurations to meet the needs of different working conditions.

[0214] Other PE pipes 24: used to expand the pipeline system or connect other pipelines to form a more complex pipeline network to meet various application requirements.

[0215] Detection device 25: Microwave nondestructive detection device for aging of polyethylene pipes and accessories.

[0216] Base station 26: used as a central node for data collection and transmission. It is connected to the detection device through wireless communication and receives real-time data (including microwave signal data, image data, etc.) from the detection device.

[0217] Server 27: used for data processing, storage and analysis. After receiving the microwave signal data and image data transmitted from the base station, the server will perform further processing, including denoising, data filtering, feature extraction, etc., to ensure high-quality data output. Based on the processed data, the server generates a defect identification report and archives the test results. The server can also synchronize data with the remote monitoring center to support remote operation and real-time monitoring. At the same time, it is also responsible for maintaining long-term storage and comparative analysis of data, providing an important basis for subsequent pipeline maintenance and management.

[0218] Fixing screws Ⅰ28, Ⅱ29, Ⅲ30: used to fix the position of the clamp on the polyethylene pipe. By tightening or adjusting the screws on the clamp, the operator can ensure that the device maintains stable contact with the pipe surface to prevent the device from sliding, dislocating or tilting during the inspection process.

[0219] Axial slider 31: used to accurately adjust the position of the detection device along the axial direction of the pipeline. By moving the slider, the detection device can be adjusted forward and backward to ensure that the probe covers the target area of ​​the pipeline during the detection process and achieve all-round detection.

[0220] Height slider 32: used to adjust the vertical position of the detection device or probe. The height slider can accurately control the height of the probe to keep it at an optimal distance from the pipeline surface, thereby improving the strength of the detection signal and the clarity of the imaging.

[0221] A microwave nondestructive detection device and method for aging of polyethylene pipes and accessories is designed to detect the aging of polyethylene pipes and accessories through microwave technology. The specific implementation steps are as follows:

[0222] Step 1: After installing and fixing the microwave nondestructive testing device, perform a full-scale scan of the polyethylene pipe and its accessories to be tested, collect microwave signals and perform preliminary processing to filter out noise and interference in the signal. To ensure the detection accuracy, first use the multi-eye camera of the equipment to accurately align the detection area to ensure a constant distance between the microwave antenna and the pipeline surface, while ensuring that the scanning path of the microwave antenna covers the entire detection area of ​​the pipeline and its accessories. During this process, the system can automatically identify and avoid the interference areas of the corresponding accessories such as the areas that cannot be scanned by the tee joints and the energized joint areas of the electric fusion joints, so as to avoid the interference of the interference areas affecting the detection results;

[0223] Step 2: After the control module receives the microwave signal data, it will perform multi-level processing on the data. First, the signal will undergo a hybrid denoising process using a convolutional neural network (CNN) and sparse representation to remove environmental electromagnetic interference and improve the purity of the signal. Next, the improved inverse scattering algorithm is used to optimize the accuracy of the microwave signal data. The optimized data is then used through an image reconstruction algorithm to generate a two-dimensional or three-dimensional microwave image of the polyethylene pipe and its accessories, clearly presenting the aging condition of the pipe and its accessories and possible defects;

[0224] Step 3: Based on the microwave imaging image, the system automatically extracts the aging characteristics of polyethylene pipes and accessories, such as surface cracks, material degradation, bubbles, delamination, etc. In order to further enhance the accuracy and comprehensiveness of defect identification, the system integrates a multi-camera to provide auxiliary visual analysis functions. The multi-camera can capture high-resolution images of the pipeline surface and accessories, and compare them with the microwave imaging image to provide more intuitive defect identification. By combining microwave imaging and images obtained by the multi-camera, the system can identify more subtle surface changes, such as microcracks caused by aging, changes in surface roughness, and other details that are difficult to identify through a single microwave signal analysis. The multi-camera can provide a multi-dimensional perspective of the pipeline surface through image capture and analysis at different angles, further improving the detection accuracy. The images captured by the camera will be analyzed by an image processing algorithm and feature extraction will be performed together with the microwave signal data. The system uses a machine learning algorithm to classify and compare these features, combining historical defect samples and visual image data to further improve the accuracy of defect identification. In addition, the system supports model optimization based on deep learning, which can better identify new aging phenomena and adapt to the aging characteristics of different types of pipelines and accessories;

[0225] Step 4: The base station is responsible for receiving real-time data from the detection equipment, including microwave signals, image data, etc., and transmitting the data to the server through wireless communication. After receiving the data, the server performs denoising, feature extraction, defect identification and other processing. According to the extracted aging features, the system automatically generates an aging assessment report for pipelines and accessories. The report includes the specific location of aging, the degree of aging, the possible level of hazard and the recommended follow-up treatment plan. The assessment results will be used to determine whether the pipeline needs maintenance or replacement, and provide data support for subsequent pipeline maintenance and safety management. At the same time, the server will also generate a historical inspection archive and store the assessment results in the database for long-term monitoring and data comparison and analysis.

[0226] Further: First, the support frame of the detection device is installed on the polyethylene pipe to be detected, and the adjustable clamp 2 is used to adapt to pipes of different diameters to ensure the overall stability of the device. The contact pressure between the device and the surface of the pipe 1 is adjusted by rotating the locking clamp 2 to ensure that the device is stable and does not slip.

[0227] Next, the detection device is fixed on the fixture 2 by adsorption through the magnetic wheel 6. The control module 8 is controlled by the rotary servo motor 4 to rotate along the pipeline 2 in all directions to ensure a 360-degree scanning of the pipeline without blind spots. The system is equipped with radial and height guides for accurately adjusting the position of the microwave antenna to ensure that the antenna is always aligned with the detection area of ​​the pipeline.

[0228] When the microwave antenna 14 is located in the area to be detected, the control module 8 sends a command to start the entire device to perform a full-circle scan. During the scanning process, the control module emits a microwave signal, which penetrates the pipeline and reflects back to the receiver. After being pre-processed by the signal acquisition unit, the microwave signal is transmitted to the control device for further processing. At the same time, the data is transmitted to the remote monitoring center in real time to facilitate remote monitoring and data analysis.

[0229] The scanning path adopts segmented circumferential scanning: First, the microwave antenna 14 starts from the initial position, records the initial coordinates, and then scans along the circumference of the pipeline. After each scan, the platform controls the antenna 14 to move to the right by a preset distance and continue the next scan. This process continues until the entire area to be inspected is covered. After the scan is completed, the system generates a complete inspection report, and can export data or generate a defect distribution map for further analysis.

[0230] During the detection process, the detection device can automatically detect and avoid interference areas such as areas that cannot be scanned by the three-way joint and the energized joint area of ​​the electric fusion joint. After the multi-eye camera 15 monitors the positions of the energized joints 21 and 22 in real time, it sends a signal to notify the control module to adjust the operating status of the equipment to ensure that the microwave antenna 14 is prevented from contacting the relevant area, while ensuring the continuity and integrity of the microwave signal. When the microwave antenna 14 approaches the relevant area, the system automatically controls the vertical or horizontal position of the microwave antenna, avoids obstacles by adjusting the sliding platform of the guide rail, and ensures the stability of the detection process.

[0231] Based on the extracted aging features, the detection device automatically generates an aging assessment report for the pipeline and its accessories, including the specific location, degree, possible hazard level and subsequent treatment recommendations of the aging. The assessment results are processed and analyzed by the server 27, and defects are identified based on microwave imaging data and multi-camera images combined with deep learning algorithms. The base station 26 is responsible for receiving real-time data transmitted by the detection equipment and sending it to the server 27 for processing. The reports and detection files generated by the server 27 will be used to determine whether the pipeline needs maintenance or replacement, and provide data support for pipeline safety management. Through the synergy of the base station 26 and the server 27, the system realizes efficient data transmission, processing and storage, ensuring the accuracy of the assessment and the feasibility of long-term monitoring.

[0232] Furthermore, step 1 specifically includes a deep reinforcement learning dynamic path adjustment automatic obstacle avoidance algorithm based on real-time distance perception

[0233] During the pipeline scanning process, the detection device first collects environmental information through the multi-camera 14 sensing device and constructs an obstacle model around the pipeline 2. The current environmental state S is generated through image processing technology and the measurement results of the multi-camera. t , which includes information such as pipes, obstacles, joints, and powered joints.

[0234] State space S t It can be expressed as:

[0235] S t = {P probe ,P obstacle , θ scan ,V probe} (twenty one)

[0236] in:

[0237] -P probe is the current antenna position;

[0238] -P obstacle The location information of obstacles (such as electric fusion joints, energized joints, etc.);

[0239] -θ scan is the scanning angle of the current antenna;

[0240] -V probe is the current antenna speed.

[0241] Reinforcement learning uses Q value to evaluate the quality of taking an action in a specific state. The update formula of Q value function is:

[0242]

[0243] in:

[0244] -Q(s t , a t ) is the state s t Take action a t Q value;

[0245] -r t+1 It is the reward of the present moment;

[0246] -γ is the discount factor, which controls the impact of future rewards;

[0247] -α is the learning rate, which controls the degree of integration of new and old information during the learning process;

[0248] - is the next state s t+1 The maximum Q value under .

[0249] The Q-value function is approximated by a deep neural network, with the input being the state s t , the output is the Q value of each possible action:

[0250]

[0251] Where θ is the parameter of the neural network, which is updated by optimizing the loss function. The training goal of the deep Q network is to minimize the following loss function:

[0252]

[0253] where θ - are the parameters of the target network, which is used to maintain stability and avoid introducing high variance during training.

[0254] In order to further optimize the path planning, when encountering complex obstacles and dynamic environments, the present invention adopts a combination strategy of the A* algorithm and the Dijkstra algorithm. These two algorithms are used to calculate and select the optimal path.

[0255] A* algorithm: It uses a heuristic method to calculate the shortest path from the starting point to the end point, taking into account the estimated distance h(n) to the target and the cost g(n) of the current path. The total cost function is:

[0256] f(n)=g(n)+h(n) (25)

[0257] in:

[0258] -g(n) is the actual cost from the starting point to the current node n;

[0259] -h(n) is the estimated cost from node n to the target node, usually using Euclidean distance or Manhattan distance.

[0260] Dijkstra algorithm: It is suitable for finding the shortest path in a graph. The shortest path cost of each node is considered during the calculation process. Its core formula is:

[0261] d(u)=min(d(u),d(υ)+w(υ,u)) (26)

[0262] in:

[0263] -d(u) is the shortest path cost from the source node to node u;

[0264] -w(υ,u) is the path weight (cost) from node v to node u.

[0265] When the detection device detects a new obstacle (such as the energized joint 21, energized joint 22, and tee joint 23 area on the electric fusion joint), the DQN model will calculate the obstacle based on the latest environmental state S t Dynamically adjust the path. The specific path adjustment can be achieved through the following real-time obstacle avoidance control strategies:

[0266] Real-time obstacle avoidance: When the microwave antenna 14 approaches an obstacle, the detection system current Calculate the shortest path d from the obstacle min , and update the path to avoid obstacles. i is the position of the ith obstacle, then the path adjustment target is:

[0267]

[0268] in:

[0269] - ||P(t)-O i || represents the path point P(t) and the obstacle O i distance.

[0270] When the obstacle changes or the path is blocked, the system will re-plan the path through the A* algorithm or Dijkstra algorithm to ensure that the obstacle is avoided and the scanning task can be continued.

[0271] The deep reinforcement learning dynamic path adjustment automatic obstacle avoidance algorithm based on real-time distance perception adopted by the present invention can effectively realize high-precision scanning of polyethylene pipes and their accessories, and automatically avoid obstacles such as the energized joint 21, the energized joint 22, and the three-way joint 23 on the electric fusion joint. Through the dynamic adjustment of deep learning and path planning, this method can ensure the stable operation of the microwave nondestructive testing device in a complex environment, provide more accurate detection results, and significantly improve the intelligence and adaptability of the system.

[0272] Furthermore, step 2 specifically includes a microwave imaging algorithm based on convolutional neural network (CNN) and sparse representation.

[0273] In microwave nondestructive testing of aging polyethylene pipes and accessories, the quality of microwave images directly affects the accuracy of defect detection. In order to improve the clarity of microwave imaging images and effectively remove noise, this paper proposes a hybrid denoising imaging algorithm based on convolutional neural network (CNN) and sparse representation. This method uses the powerful feature extraction ability of deep learning and the image reconstruction advantages of sparse representation to better handle the noise generated during microwave imaging and improve the final imaging quality.

[0274] First, the microwave signal is transmitted through the microwave antenna and propagated in the PE pipe before being reflected back and collected by the receiver. The reflected signal is filtered and amplified in the preprocessing stage, and then the image is reconstructed using the Inverse Scattering algorithm. The Inverse Scattering algorithm often uses the following formula to represent the reconstruction process of the reflected signal:

[0275]

[0276] in:

[0277] -f(x,y) is the reconstructed image intensity;

[0278] -k is the microwave wave number;

[0279] -r and r′ are the distance from the antenna to the target and from the target to the receiver, respectively;

[0280] - A(x′, y′) is the scattering coefficient of the target;

[0281] -S is the scatterer surface area.

[0282] Using this formula, the system can recover a two-dimensional or three-dimensional image of the target from the reflected signal, revealing the internal structure and potential defects of the pipe joint.

[0283] Although the inverse scattering algorithm can reconstruct the image, there may still be unnecessary details in the image due to noise interference, which may interfere with subsequent defect detection. Therefore, in order to improve the imaging quality, the present invention combines the convolutional neural network (CNN) and the sparse representation method to form a hybrid denoising technology.

[0284] Convolutional neural network (CNN) has a powerful image feature learning ability and can effectively extract and restore the main structural information of the image from a complex noisy environment. Aiming at the denoising problem of microwave imaging images, the present invention proposes to use a convolutional neural network model to process the noise of the image. CNN gradually extracts the high-order features of the image through multiple convolutional layers, pooling layers and fully connected layers, and finally reconstructs the denoised image through the deconvolution layer.

[0285] 1. Input layer: The input image is a noisy microwave imaging image with a size of N×M;

[0286] 2. Convolutional layer: Multi-layer convolutional layers are used to extract local features from images. The convolution kernel size is usually 3×3 or 5×5.

[0287] 3. Pooling layer: The pooling layer reduces the size of the image by downsampling while retaining the main features of the image;

[0288] 4. Fully connected layer: Through the fully connected layer, the network can capture global features and perform high-level feature fusion;

[0289] 5. Output layer: output the denoised image.

[0290] The loss function of convolutional neural network denoising uses mean square error (MSE):

[0291]

[0292] in:

[0293] - It is the image after CNN denoising;

[0294] -I clean，i is the corresponding clean image;

[0295] - N is the number of pixels in the image.

[0296] Through this deep learning model, CNN can effectively restore clear images from noisy microwave imaging images and reduce the interference of noise on detection.

[0297] Sparse representation is a denoising method based on the sparsity characteristics of the signal. In image denoising, sparse representation assumes that the image can be represented by a few basis functions (dictionary elements), thereby removing redundant information and achieving denoising effects. The core idea of ​​this method is to solve the sparse representation of the image through optimization problems.

[0298] The mathematical formula for sparse representation is:

[0299] I=DX (30)

[0300] in:

[0301] -I is a noisy image;

[0302] -D is a dictionary matrix consisting of a set of basis functions;

[0303] -X is a sparse coefficient matrix representing the weighting coefficients of each basis function in the image.

[0304] The goal of denoising is to restore the clean part of the image by solving the sparse coefficient X. The optimization problem can be expressed by the following formula:

[0305]

[0306] in:

[0307] - is the image reconstruction error;

[0308] -||X||1 is a sparsity constraint, which promotes sparse representation;

[0309] -λ is a hyperparameter that adjusts the trade-off between sparsity and reconstruction error.

[0310] Through this optimization problem, sparse representation can effectively remove noise from the image and retain the main structural information of the image.

[0311] The denoising algorithm combining CNN and sparse representation can effectively solve the balance problem between noise and detail loss in microwave imaging images and obtain clearer imaging results.

[0312] The hybrid denoising imaging algorithm based on convolutional neural network (CNN) and sparse representation proposed in this paper can effectively remove noise in microwave imaging images and restore image clarity through the combination of deep learning and sparse representation. This method not only improves the quality of microwave images, but also enhances the ability to detect tiny defects, thus providing more accurate and reliable technical support for non-destructive testing of aging of polyethylene pipes and accessories.

[0313] Furthermore, step 3 specifically includes a microwave image defect recognition algorithm based on improved YOLOv11

[0314] In pipeline aging inspection, combining microwave imaging and the improved YOLOv11 model can significantly improve the accuracy and efficiency of defect detection. Although the traditional YOLO model can detect targets in a relatively short time, its accuracy and robustness for specific application areas are often insufficient when processing microwave images. To meet this challenge, this paper proposes an improved defect recognition algorithm based on YOLOv11. By optimizing the model architecture and loss function, it can better process complex microwave imaging data. The specific process is as follows: Figure 8 shown.

[0315] The raw data collected by the detection device from the microwave antenna and multi-eye camera enters the detection process. The data includes microwave imaging and high-resolution visual images. Data transmission is carried out through wireless base stations to ensure that the input data is complete and lossless and complies with the format specifications of the detection system. For different types of detection tasks, the system can accept user-defined input parameters, such as target detection area, priority of specific defect types, etc. The preprocessing module converts the format of the raw data, adjusts the microwave image and visual image to a unified resolution and format, and adapts to the input dimension of the detection model. In the preprocessing stage, data enhancement technology is applied, including random flipping, rotation, illumination adjustment, and random cropping, to improve the adaptability of the model to different scenarios. The microwave imaging image is processed by a denoising algorithm to reduce environmental interference, improve the signal-to-noise ratio of the data, and ensure the accuracy of subsequent feature extraction. The feature extraction module adopts an improved YOLOv11 network backbone structure to extract feature maps of different scales through multi-layer convolution and downsampling operations. In the feature extraction process, the system integrates an attention mechanism to enhance the focus on key feature areas and reduce the impact of irrelevant areas on detection performance. The visual images provided by the multi-camera are processed by another parallel network and fused with microwave features to provide richer feature dimensions for defect detection. The detection network generates candidate boxes based on the extracted feature map and calculates the position, size and confidence of the bounding box. The generation of candidate boxes adopts an improved anchor box mechanism to enhance the capture ability of small defects and reduce the false detection rate. In order to improve the efficiency of detection, the system combines the optimized non-maximum suppression algorithm to filter out redundant bounding boxes and retain the optimal solution. The detection module classifies the defect features in each candidate box, including common aging problems such as surface cracks, material degradation, bubbles, and delamination. By comparing the deep learning model with the historical defect sample library, the classification results are further optimized and the recognition ability of new defects is enhanced. In the post-processing stage, the system integrates the information of microwave imaging and visual images, and fine-tunes the size and position of the bounding box to ensure the accuracy of the detection results. The system automatically generates a detailed inspection report based on the classification results, which includes the specific location, morphological characteristics, area, depth and other information of the defect. The inspection report further evaluates the potential hazard level of the defect to the pipeline safety and provides maintenance recommendations and treatment plans. All test results and reports are automatically stored in the remote server and can be transmitted to the user terminal in real time via the base station, facilitating subsequent historical data comparison and long-term monitoring.

[0316] In defect detection in microwave images, microwave images provide in-depth information about pipeline aging from different angles and levels, including cracks, bubbles, delamination, material degradation, etc. YOLOv11 converts these image data into target detection tasks for automated defect recognition.

[0317] The improved YOLOv11 model uses a more efficient convolutional neural network to extract features. The microwave imaging image is convolved using the following formula to extract features:

[0318] f out =f in *W+b (32)

[0319] in:

[0320] -f in is the feature map of the input image (microwave image).

[0321] -W is the convolution kernel weight and b is the bias.

[0322] -f out is the output feature map.

[0323] Through multi-scale feature fusion, YOLOv11 fuses feature maps at different levels to enhance the model's ability to detect defects of different sizes and scales. The core formula for feature fusion is as follows:

[0324]

[0325] in:

[0326] -F i are feature maps of different scales.

[0327] -α i The weight of each scale feature map is automatically learned by the network.

[0328] YOLOv11 outputs the category probability of each candidate region and the four parameters of the bounding box (center coordinates, width, and height) through the fully connected layer. For each grid, the YOLOv11 network outputs a fixed number of bounding box candidates and predicts confidence and category probability for each bounding box:

[0329] Prediction={x,y,w,h,c,p1,p2,...,p C} (34)

[0330] in:

[0331] - (x, y) is the center coordinate of the bounding box;

[0332] - (w, h) is the width and height of the bounding box;

[0333] -c is the confidence of the bounding box;

[0334] -p1,p2,...,p C is the probability that the box belongs to each category.

[0335] A more detailed loss function design is introduced to adapt to the characteristics of microwave images. The loss function consists of the following parts:

[0336] L=L bbox +L obj +L class +L IOU (35)

[0337] in:

[0338] -L bbox It is the bounding box regression loss, which measures the difference between the predicted box and the true box. The mean square error (MSE) is usually used as the calculation method:

[0339]

[0340] Where (x i ,y i , w i ,h i ) is the coordinate of the real frame, are the coordinates of the predicted box.

[0341] -L obj * is the target existence loss, which measures whether the box contains the target.

[0342]

[0343] -L class is the classification loss, which measures the accuracy of the target category prediction:

[0344]

[0345] -L IOU The intersection over union loss is used to improve the overlap between the bounding box and the ground truth box. Calculate the intersection over union (IoU) and optimize:

[0346] L IOU =1-IoU (39)

[0347] To avoid detecting the same defect multiple times, YOLOv11 uses non-maximum suppression (NMS) to filter multiple bounding boxes and only retain the bounding box with the highest prediction confidence. NMS is performed using the following formula:

[0348]

[0349] When the IoU is greater than the set threshold, the bounding boxes with low confidence are removed and only the optimal box is retained.

[0350] After detection and processing by the YOLOv11 model, the system can efficiently identify and locate pipeline aging defects such as cracks, delamination, bubbles, etc. By combining with microwave imaging, the YOLOv11 model can identify and classify more subtle defects, and further analyze and evaluate defects in combination with historical data.

[0351] The improved YOLOv11 algorithm significantly improves the defect recognition capability in microwave images by introducing multi-scale feature fusion, improved loss function and non-maximum suppression. Through this method, the system can more accurately locate and classify aging defects of polyethylene pipes and accessories, providing more accurate analysis results for pipeline health monitoring.

[0352] Simulation experiment:

[0353] In the electromagnetic simulation software CST, the model established is as follows Fig. 9 As shown. According to the dimensions of the actual PE pipe (such as the inner and outer diameters and length of the pipe), create a cylinder in the modeling interface of CST to represent the main body of the PE pipe. Set the material properties of the cylinder to polyethylene (PE), and define its electromagnetic parameters, such as relative dielectric constant and loss factor, to match the actual electromagnetic properties of the pipeline. Define the relative dielectric constant (approximately 2.25) and loss tangent value of the polyethylene material. On the surface or inside of the pipeline model, create a corresponding defect model based on the actual defect morphology (such as aging, bubbles, cracks, or delamination). Set the material parameters of the defect so that the simulation can detect and distinguish the electromagnetic properties of the defect. Arrange microwave antennas around the pipeline model to transmit microwave signals and receive reflected and transmitted signals after penetrating the pipeline. The simulation results are shown in Fig.10 As shown in the figure, the propagation path of the microwave signal in the PE pipe is clearly visible. The signal intensity distribution diagram shows that microwaves can effectively penetrate the PE material and produce obvious signal attenuation and scattering at the defects. Fig.10 As shown in the figure, the defects on the surface and inside of the pipeline (such as bubbles, delamination, and cracks) are clearly marked in the simulation diagram. After the microwave signal passes through the defective area, the amplitude and phase change significantly, and the reflection intensity of the signal is significantly enhanced in the defective area, forming a specific reflection pattern. The simulation data shows that in the intact area, the signal intensity is uniform; in the defective area, the signal is significantly attenuated or abnormal. This is consistent with the actual detection results, indicating that the simulation results have good credibility.

[0354] The above is an explanation of the method and system of the present invention, and researchers in the field can reproduce the present invention through these explanations. Based on the content of the present invention, other embodiments and results obtained by researchers without creative work should fall within the scope of protection of the present invention.

Claims

1. A microwave nondestructive testing device for aging of polyethylene pipes and accessories, characterized in that: It includes a motion module, a signal module and a control module; the motion module includes a fixed component and a motion component, the fixed component is a plurality of pipe clamps wrapped around the polyethylene pipe, and any two pipe clamps are connected by clamp connection accessories; the motion component includes a rotary motion mechanism, an axial motion mechanism and a height adjustment mechanism; the signal module is connected to the control module, both of which are installed on the motion module, and are used to realize all-round detection of the polyethylene pipe and its accessories.

2. The device according to claim 1, characterized in that The rotary motion mechanism includes a rotary servo motor, the output shaft of the rotary servo motor is connected to the rotary motor transmission shaft, a rotary motion magnetic wheel is installed on the rotary motor transmission shaft, the rotary motion magnetic wheel engages the clamp, and the magnetic wheel contacts the clamp through magnetic adsorption force to realize the circumferential rotation of the device.

3. The device according to claim 1, characterized in that The axial motion mechanism includes an axial guide rail, an axial bracket, an axial servo motor and an axial slider. The axial servo motor drives the axial slider to drive the height bracket to move axially through the axial guide rail installed in the axial bracket. The axial slider is connected to the height adjustment mechanism to realize the axial motion adjustment of the device.

4. The device according to claim 1, characterized in that The height adjustment mechanism includes a height guide rail, a height slider, a height bracket and a height servo motor. The output shaft of the height servo motor is connected to the internal height guide rail. The height guide rail drives the height slider to slide along the height guide rail, so that the microwave antenna and the multi-eye camera are installed on the height adjustment mechanism through the height slider to realize the height adjustment function of the microwave antenna and the multi-eye camera.

5. The device according to claim 1, characterized in that The high-resolution surface images of the polyethylene pipe and its appendages are collected by a multi-camera; the multi-camera is connected to the signal module and the control module, and is fused with the detection data of the microwave antenna module to improve the defect recognition accuracy; The control module receives the data from the signal module and controls the operation of the moving parts. At the same time, it combines the microwave detection results with the data from the image acquisition module and generates an aging assessment report after processing by the data processing module. The report includes the aging location, degree, hazard level and treatment suggestions.

6. The device according to claim 1, characterized in that The signal module is connected to the base station and server through wireless communication, and the detection data can be remotely transmitted and monitored. The base station is responsible for real-time reception and transmission of signals, and the server is responsible for storage and further data analysis and processing.

7. A method for microwave nondestructive testing of aging of polyethylene pipes and accessories, characterized in that: The steps include: Step 1: After installing and fixing the microwave nondestructive testing device, perform a full-scale scan of the polyethylene pipe and its accessories, collect microwave signals, and preliminarily filter out noise and interference; Multi-camera assists in locating the inspection area, ensuring that the microwave antenna maintains a constant distance from the pipeline surface, while avoiding interference areas that cannot be detected, such as tee joints and energized parts of electric fusion joints; Step 2: After receiving the microwave signal, the control module uses a hybrid denoising algorithm to eliminate environmental interference and combines it with an improved inverse scattering algorithm to optimize signal quality. The processed data is converted into a two-dimensional or three-dimensional microwave imaging image through an image reconstruction algorithm to clearly display the aging characteristics and defect distribution of the pipeline and its accessories. Step 3: Based on the microwave imaging, the system extracts aging features (such as cracks, degradation, bubbles, delamination, etc.), and uses a multi-camera to capture high-resolution surface images, which are compared and analyzed with the microwave images. The system further optimizes the recognition accuracy by combining machine learning algorithms and historical sample data to classify defects, and adapts to a variety of pipeline aging features through deep learning models. Step 4: The base station receives the equipment data and transmits it wirelessly to the server. The server completes denoising, feature extraction and defect assessment, and automatically generates an assessment report, including the aging location, degree, hazard level and treatment suggestions. The assessment results support pipeline maintenance decisions, and the inspection history archives are stored in the database for long-term monitoring and data comparison and analysis.

8. The method according to claim 7, characterized in that Step 1: Full circumferential scanning also includes automatic obstacle avoidance, which includes: The current environment state S is generated through image processing technology and the measurement results of multiple cameras. t , which includes information such as pipelines, obstacles, joints, and energized joints. The state space S t It can be expressed as: S t 6{P robe ,P obstacle ,θ scan ,V probe } (1) in: -P probe is the current antenna position; -P obstacle The location information of obstacles (such as electric fusion joints, energized joints, etc.); -θ scan is the scanning angle of the current antenna; -V probe is the current antenna speed; Reinforcement learning uses Q value to evaluate the quality of taking an action in a specific state. The update formula of the Q value function is: in: -Q(s t , a t ) is the state s t Take action a t Q value; -r t +1 is the reward at the current moment; -γ is the discount factor, which controls the impact of future rewards; -α is the learning rate, which controls the degree of integration of new and old information during the learning process; - is the next state s t+1 The maximum Q value under The Q-value function is approximated by a deep neural network, with the input being the state s t , the output is the Q value of each possible action: Where θ is the parameter of the neural network, which is updated by optimizing the loss function. The training goal of the deep Q network is to minimize the following loss function: where θ - are the parameters of the target network, which is used to maintain stability and avoid introducing high variance during training; In order to further optimize path planning, when encountering complex obstacles and dynamic environments, the present invention adopts a combination strategy of the A* algorithm and the Dijkstra algorithm, which are used to calculate and select the optimal path; A* algorithm: It uses a heuristic method to calculate the shortest path from the starting point to the end point, taking into account the estimated distance h(n) to the target and the cost g(n) of the current path. The total cost function is: f(n)=g(n)+h(n) (5) in: -g(n) is the actual cost from the starting point to the current node n; -h(n) is the estimated cost from node n to the target node, usually using Euclidean distance or Manhattan distance; Dijkstra algorithm: It is suitable for finding the shortest path in a graph. The shortest path cost of each node is considered during the calculation process. Its core formula is: d(u)=min(d(u),d(v)+w(v,u)) (6) in: -d(u) is the shortest path cost from the source node to node u; -w(v,u) is the path weight (cost) from node v to node u; When the system detects a new obstacle (such as a live joint on an electric fusion joint, a tee area, etc.), the DQN model will t Dynamically adjust the path. The specific path adjustment can be achieved through the following real-time obstacle avoidance control strategies: Real-time obstacle avoidance: When the antenna approaches an obstacle, the system current Calculate the shortest path d from the obstacle min , and update the path to avoid obstacles. Assume that O i is the position of the ith obstacle, then the path adjustment target is: in: -||P(t)-O i || represents the path point P(t) and the obstacle O i distance; When the obstacle changes or the path is blocked, the system will re-plan the path through the A* algorithm or Dijkstra algorithm to ensure that the obstacle is avoided and the scanning task can be continued.

9. The method according to claim 7, characterized in that: Step 2 also includes image reconstruction. The image reconstruction algorithm is a convolutional neural network (CNN) and a sparse representation microwave imaging algorithm, which specifically includes: The inverse scattering algorithm often uses the following formula to express the reconstruction process of the reflected signal: in: -f(x,y) is the reconstructed image intensity; -k is the microwave wave number; -r and r′ are the distance from the antenna to the target and from the target to the receiver, respectively; - A(x′, y′) is the scattering coefficient of the target; -S is the scatterer surface area; CNN gradually extracts high-level features of the image through multiple convolutional layers, pooling layers, and fully connected layers, and finally reconstructs the denoised image through the deconvolution layer:

1. Input layer: The input image is a noisy microwave imaging image with a size of N×M; 2. Convolutional layer: Multi-layer convolutional layers are used to extract local features from images. The convolution kernel size is usually 3×3 or 5×5.

3. Pooling layer: The pooling layer reduces the size of the image by downsampling while retaining the main features of the image; 4. Fully connected layer: Through the fully connected layer, the network can capture global features and perform high-level feature fusion; 5. Output layer: output denoised image; The loss function of convolutional neural network denoising uses mean square error (MSE): in: - It is the image after CNN denoising; -I clean，i is the corresponding clean image; - N is the number of pixels in the image; The sparse representation of the image is solved by optimizing the problem. The mathematical formula of sparse representation is: I=DX (10) in: -I is a noisy image; -D is a dictionary matrix consisting of a set of basis functions; -X is a sparse coefficient matrix, representing the weighted coefficients of each basis function in the image; The goal of denoising is to restore the clean part of the image by solving the sparse coefficient X. The optimization problem can be expressed by the following formula: in: - is the image reconstruction error; -||X||1 is a sparsity constraint, which promotes sparse representation; -λ is a hyperparameter that adjusts the trade-off between sparsity and reconstruction error; The denoising algorithm combining CNN and sparse representation can effectively solve the balance problem between noise and detail loss in microwave imaging images and obtain clearer imaging results.

10. The method according to claim 7, characterized in that Step 3 also includes a machine learning algorithm to improve the microwave image defect recognition algorithm of YOLOv11, specifically including: The improved YOLOv11 model uses a more efficient convolutional neural network to extract features, and the microwave imaging image is convolved to extract features using the following formula: f out =f in *W+b (12) in: -f in is the feature map of the input image (microwave image); -W is the convolution kernel weight, b is the bias; -f out is the output feature map; Through multi-scale feature fusion, YOLOv11 fuses feature maps at different levels to enhance the model's ability to detect defects of different sizes and scales. The core formula for feature fusion is as follows: in: -F i are feature maps of different scales; -α i The weight of each scale feature map is automatically learned by the network; For each grid, the YOLOv11 network outputs a fixed number of bounding box candidates and predicts confidence and class probability for each bounding box: Prediction={x,y,w,h,c,p1,p2,...,p C } (14) in: -(x, y) is the center coordinate of the bounding box; - (w, h) is the width and height of the bounding box; -c is the confidence of the bounding box; -p1,p2,...,p C is the probability that the box belongs to each category; A more detailed loss function design is introduced to adapt to the characteristics of microwave images. The loss function consists of the following parts: L=L bbox +L obj +L class +L IOU (15) in: -L bbox It is the bounding box regression loss, which measures the difference between the predicted box and the true box. The mean square error (MSE) is usually used as the calculation method: Where (x i ,y i , w i ,h i ) is the coordinate of the real frame, is the coordinate of the prediction box; -L obj * is the target existence loss, which measures whether the box contains the target; -L class is the classification loss, which measures the accuracy of the target category prediction: -L IOU The intersection-over-union loss is used to improve the overlap between the bounding box and the true box, calculate the intersection-over-union (IoU) and optimize it: L IOU =1-IoU (19) In order to avoid detecting the same defect multiple times, YOLOv11 uses non-maximum suppression (NMS) to filter multiple bounding boxes and only retain the bounding boxes with the highest prediction confidence. NMS is performed using the following formula: When the IoU is greater than the set threshold, the bounding boxes with low confidence are removed and only the optimal box is retained.

Images