Intelligent detection system for lap joint quality of waterproof roll based on deep learning

Abstract

The invention discloses an intelligent detection system for waterproof roll lap joint quality based on deep learning, and relates to the technical field of waterproof roll lap joint quality detection.The intelligent detection system is characterized in that an area acquisition module is arranged, and after detection personnel complete image data shooting on any waterproof roll lap joint area in a target area, a detection result is obtained; the image processing module is used for receiving and collecting the image data, the image auditing module is used for pre-auditing any image data, and images with unqualified shooting heights and inclination angles are screened out in advance in the pre-auditing process, so that low-quality image data with serious perspective distortion are reduced from the source; therefore, a subsequent quality evaluation platform does not need to execute large-scale matrix operation, pixel coordinate mapping calibration and other complex preprocessing operations on a large number of distorted images, the requirement for platform computing power resources is greatly reduced, and the hardware configuration cost and operation energy consumption of a detection system are effectively reduced; and meanwhile, the evaluation result delay caused by complex processing is avoided, and the response efficiency of the detection service is improved.

Description

Technical Field

[0001] This invention relates to the field of waterproof membrane overlap quality inspection technology, specifically to an intelligent inspection system for waterproof membrane overlap quality based on deep learning. Background Technology

[0002] The quality of overlapping of waterproof membrane is the core link in the construction quality control of building waterproofing projects. The sealing, flatness and dimensional compliance of the overlapping area directly determine the seepage prevention effect of the waterproofing system. Once there are defects in the overlapping quality, it is easy to cause leakage risks, which can lead to serious problems such as damage to the building structure and shortened service life. Therefore, efficient and accurate detection of overlapping quality is crucial during the construction and acceptance stages of waterproof membrane. Traditional waterproof membrane overlap quality inspection mainly relies on manual inspection. Inspectors need to carry tools such as tape measures and feeler gauges to the construction site to observe the appearance of the overlap area with the naked eye and manually measure key parameters such as overlap width and gap. Finally, they judge whether the quality is qualified according to the construction specifications. However, manual inspection is affected by the physical strength, experience and sense of responsibility of the inspectors. The inspection of a single area takes a long time and is difficult to adapt to the rapid inspection needs of large-scale construction projects. In addition, manual inspection is highly subjective. The accuracy of the identification of overlap defects depends on the professional ability of the inspectors and is prone to missed judgments and misjudgments. Especially for subtle overlap misalignment or hidden defects, it is difficult to achieve accurate identification, resulting in the inability to guarantee the reliability and consistency of the inspection results. To address the drawbacks of manual inspection, semi-automatic inspection solutions based on image acquisition and platform analysis have gradually emerged in the industry. These solutions involve using mobile devices to photograph the overlapping areas of waterproof membrane rolls, uploading the collected image data to a cloud or local quality assessment platform. The platform's built-in detection algorithms then analyze and process the images to assess and determine the overlap quality. While this approach improves inspection efficiency and reduces manual intervention to some extent, existing image inspection solutions suffer from challenges. The complexity of the on-site shooting environment makes it difficult for inspectors to maintain consistency in shooting height and tilt angle when operating the equipment. This can cause significant stretching of the geometric shape of the overlapping area in the image. Perspective distortion and other issues lead to inconsistent image data quality when uploaded to the platform. Before performing core quality assessment, in addition to basic preprocessing such as grayscale conversion and denoising, the platform must perform complex correction processing for image distortion, including geometric shape restoration based on perspective transformation, pixel coordinate mapping calibration, feature point reconstruction, and a series of other operations. These preprocessing operations involve large-scale matrix operations and pixel-level data processing, requiring the platform to call up extremely high computing resources to ensure processing accuracy. This not only significantly increases the hardware configuration cost and operating energy consumption of the detection system, but also significantly prolongs the image processing cycle, resulting in a long delay in the output of quality assessment results. To address the above problems, this invention proposes a solution. Summary of the Invention

[0003] The purpose of this invention is to provide an intelligent detection system for the overlap quality of waterproof membranes based on deep learning, in order to solve the problems mentioned in the background art.

[0004] This invention provides an intelligent detection system for the overlap quality of waterproof membranes based on deep learning, comprising: The image review module is used to pre-review the image data of any overlapping area within the target area after receiving it. If the pre-review result is passed, the image data is transmitted to the quality assessment platform. The overlapping area refers to the overlapping and bonding area of ​​adjacent waterproof membranes during the laying of waterproof membrane. A quality assessment platform is used to assess the overlap quality of waterproof membrane in a target area. The quality assessment platform includes an image processing unit and a processing and analysis unit. The image processing unit is used to preprocess the image data of any overlapping area within the target area after receiving it. During the preprocessing process, geometric recognition is performed on the image data to obtain the shooting height and tilt angle of the image data. After the preprocessing is completed, the geometric processing time of the image data is recorded. Recorded data of the image data is generated based on the shooting height, tilt angle and geometric processing time, and the recorded data is transmitted to the processing and analysis unit for storage. The processing and analysis unit is used to analyze the recorded data stored therein after it reaches a preset fixed amount to generate review standard data, and then transmit the review standard data to the image review module for storage.

[0005] Furthermore, it also includes an area acquisition module, which is used to receive and collect the image data after the inspection personnel have completed the image data shooting of any overlapping area of ​​the waterproof membrane in the target area, and transmit the image data to the image review module.

[0006] Furthermore, the quality assessment platform also includes an overlap assessment unit, which pre-stores an overlap assessment strategy. Each time the overlap assessment unit receives a pre-processed image data, it retrieves the overlap assessment strategy and performs feature extraction, target matching, and parameter verification on the standard image data to complete the quality assessment of the overlap area.

[0007] Furthermore, if the review fails, a prompt message is generated based on the overlapping area corresponding to the image data, prompting the inspection personnel to re-photograph the overlapping area.

[0008] Furthermore, the analysis steps for the processing and analysis unit to analyze and generate audit standard data after the stored record data reaches a preset fixed amount are as follows: S11: Obtain all recorded data stored in the processing and analysis unit, and label them as A1, A2, ..., Aa, where a≥1; S12: Obtain the shooting height, tilt angle and geometric processing time contained in the recorded data A1, label them as B1, C1 and D1 in sequence, and construct the geometric time correlation equation of the recorded data A1 based on them. The geometric time correlation equation is (B1-PB)×ɑ1+(C1-PC)×ɑ2+(B1-PB)(C1-PC)×ɑ3= D1-P1; In the formula, PB is the preset standard shooting height, PC is the preset standard tilt angle, P1 is the preset standard geometric processing time; a1 is the time consumption coefficient per unit height deviation, a2 is the time consumption coefficient per unit angle deviation, and a3 is the coupling coefficient between shooting height and tilt angle. S13: Following step S12, construct the geometric time-consuming correlation equations for the recorded data A2, A3, ..., Aa in sequence. Combine every three geometric time-consuming correlation equations in the recorded data A1, A2, ..., Aa into a set to generate several sets of ternary simultaneous equations. By performing the solution operation on each set of three simultaneous equations, a solution set corresponding to each set of equations can be obtained. Each solution set contains one solution for a1, a2, and a3 respectively. S14: Extract the values ​​of a1 from all solution sets, process all extracted a1 values ​​using a discrete point filtering algorithm, and calculate the average of all remaining a1 values ​​after data processing. Label this average as the high-skewness coefficient. Similarly, extract the values ​​of a2 and a3 from all solution sets respectively. The discrete point filtering algorithm was used to process all the extracted values ​​of a2 and a3 to obtain the angular deflection coefficient and the high tilt coupling coefficient. Audit standard data is generated based on the high skewness coefficient, angle skewness coefficient, and high tilt coupling coefficient.

[0009] Furthermore, the pre-screening content is as follows: Perform geometric recognition on the image data to obtain the shooting height D1 and tilt angle F1 of the image data; The geometric time scalar G1 for obtaining the image data is calculated using the formula G1=(D1-PB)×Pɑ1+(F1-PC)×Pɑ2+(D1-PB)(F1-PC)×Pɑ3. G1 and G are compared. If G1>G, the pre-review result is output as passed; otherwise, the pre-review result is output as failed. In the formula, PA1, PA2, and PA3 are the high skewness coefficient, angular skewness coefficient, and high tilt coupling coefficient contained in the review standard data stored in the current image review module, respectively, and G is the preset pre-review judgment threshold.

[0010] Compared with existing technologies, it has the following advantages: This invention, through the setting of an area acquisition module, receives and collects image data after the inspector has captured image data of any overlapping area of ​​waterproof membrane within the target area. An image review module is set up to pre-review any image data. During the pre-review process, images with unqualified shooting height and tilt angle are screened out in advance, reducing low-quality image data with severe perspective distortion from the source. This eliminates the need for the subsequent quality assessment platform to perform complex preprocessing operations such as large-scale matrix operations and pixel coordinate mapping calibration on a large number of distorted images, significantly reducing the demand for platform computing resources, effectively reducing the hardware configuration cost and operating energy consumption of the inspection system, and avoiding the delay in assessment results caused by complex processing, thereby improving the response efficiency of the inspection service. This invention preprocesses image data using an image processing unit. During preprocessing, geometric recognition is performed on the image data to obtain the shooting height and tilt angle. After preprocessing, the geometric processing time is recorded. A processing analysis unit analyzes the shooting height, tilt angle, and geometric processing time. By constructing a geometric processing time correlation equation, solving a system of equations, and refining the data using a discrete point filtering algorithm, this invention can accurately capture the quantitative correlation between shooting height deviation, tilt angle deviation, their coupling effect, and geometric processing time. The calibrated high deviation coefficient, angle deviation coefficient, and high tilt coupling coefficient are more consistent with the actual shooting environment and equipment characteristics of specific projects, providing accurate quantitative basis for the pre-review judgment of the image review module and improving the accuracy and credibility of the pre-review results. Attached Figure Description

[0011] Figure 1 This is a system block diagram of the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] Please see Figure 1 This application provides an intelligent detection system for the overlap quality of waterproof membrane based on deep learning, including an area acquisition module, an image review module, and a quality assessment platform; The area acquisition module is used to receive and collect the image data after the inspection personnel have taken an image of any overlapping area of ​​the waterproof membrane in the target area, and transmit the image data to the image review module. The overlapping area refers to the overlapping and bonding area of ​​adjacent waterproof membranes during the laying of the waterproof membrane, and the target area refers to the construction section in the waterproof membrane laying project that is pre-delineated and requires overlapping quality inspection. It should be noted that the region acquisition module will only transmit image data to the image review module if the image review module stores the review standard data; otherwise, the region acquisition module will directly transmit the image data to the quality assessment platform. The image review module is used to pre-review the image data of any overlapping area within the target area after receiving it. If the pre-review result is passed, the image data is transmitted to the quality assessment platform. Otherwise, if the pre-review result is failed, a prompt message is generated according to the overlapping area corresponding to the image data, prompting the inspection personnel to re-photograph the overlapping area. The pre-screening content is as follows: Perform geometric recognition on the image data to obtain the shooting height D1 and tilt angle F1 of the image data; The geometric time scalar G1 for obtaining the image data is calculated using the formula G1=(D1-PB)×Pɑ1+(F1-PC)×Pɑ2+(D1-PB)(F1-PC)×Pɑ3. G1 and G are compared. If G1>G, the pre-review result is output as passed; otherwise, the pre-review result is output as failed. In the formula, PA1, PA2, and PA3 are the high skewness coefficient, angular skewness coefficient, and high tilt coupling coefficient contained in the review standard data stored in the current image review module, respectively. G is the preset pre-review judgment threshold. In this application, the shooting height is defined as the vertical distance from the center of the camera lens to the plane on which the surface of the waterproof membrane is located, with the plane on which the surface of the waterproof membrane is located as the reference plane; the tilt angle is defined as the angle between the optical axis of the camera and the normal direction of the surface of the waterproof membrane, with the tilt angle of the vertical orthogonal shooting being 0°. A quality assessment platform is used to assess the overlap quality of waterproof membrane in a target area. The quality assessment platform includes an image processing unit, an overlap assessment unit, and a processing and analysis unit. After receiving image data of any overlapping area within the target area, the quality assessment platform transmits it to the image processing unit. After receiving the transmitted image data, the image processing unit preprocesses it. During the preprocessing process, geometric recognition is performed on the image data to obtain the shooting height and tilt angle of the image data. After the preprocessing is completed, the geometric processing time of the image data is recorded. The geometric processing time refers to the total time spent correcting the geometric distortion of the image caused by the shooting angle offset and the device displacement. Based on the shooting height, tilt angle and geometric processing time, the image data is generated and recorded data is transmitted to the processing and analysis unit for storage. The purpose of the preprocessing is to remove interference signals such as Gaussian noise and salt-and-pepper noise introduced during the image data capture process, correct image geometric distortion caused by shooting angle shift and device displacement, enhance key features such as the edge and texture of the target object in the overlapping area, and effectively reduce the interference of invalid background information on subsequent image recognition and quality analysis. In this application, the preprocessing methods include, but are not limited to, image grayscale conversion, image denoising, geometric transformation, contrast enhancement, and threshold segmentation. Specifically, image grayscale conversion uses a weighted average method to convert a three-channel color image into a single-channel grayscale image, reducing data dimensionality and computational load. Image denoising employs median filtering or bilateral filtering algorithms to remove noise while preserving the edge features of the target object. Geometric transformations encompass translation, rotation, scaling, and perspective transformations, used to standardize image size and correct distortion. Contrast enhancement is achieved through histogram equalization or gamma correction, increasing the difference between the target region and the background. Threshold segmentation uses the Otsu method to determine a global threshold, achieving initial separation of the target region from the background, laying the foundation for subsequent feature extraction and quality assessment. The preprocessed image data is transmitted to the overlap evaluation unit as a standard image data of the overlap area. The overlap evaluation unit pre-stores an overlap evaluation strategy. The overlap evaluation unit retrieves the overlap evaluation strategy to perform feature extraction, target matching, and parameter verification on the standard image data, thereby completing the quality evaluation of the overlap area. In this application, the overlap evaluation strategy includes a feature matching sub-strategy, a geometric parameter verification sub-strategy, a quality grading sub-strategy, and an anomaly determination sub-strategy. The feature matching sub-strategy pre-stores standard feature parameters and feature matching algorithms for the overlapping area, used to extract and match features of overlapping edges, overlapping widths, and overlapping gaps in the standard image data to obtain feature matching degree. The geometric parameter verification sub-strategy pre-stores a shooting height-tilt angle correction coefficient table and a pixel-actual size conversion formula, used to eliminate the interference of shooting posture on geometric parameter measurement and calculate the actual geometric parameters of the overlapping area. The quality grading sub-strategy presets overlapping quality grade classification standards and corresponding parameter judgment thresholds, used to determine the quality grade of the overlapping area based on the actual geometric parameters and feature matching degree. The anomaly judgment sub-strategy pre-stores a typical defect feature library and defect recognition rules for the overlapping area, used to identify whether there are abnormal defects such as overlapping misalignment, overlapping gaps, and surface damage in the standard image data. The processing and analysis unit analyzes the image data stored within it after it reaches a preset fixed amount. The analysis steps are as follows: S11: Obtain all recorded data stored in the processing and analysis unit, and label them as A1, A2, ..., Aa, where a≥1; S12: Obtain the shooting height, tilt angle and geometric processing time contained in the recorded data A1, label them as B1, C1 and D1 in sequence, and construct the geometric time correlation equation of the recorded data A1 based on them. The geometric time correlation equation is (B1-PB)×ɑ1+(C1-PC)×ɑ2+(B1-PB)(C1-PC)×ɑ3= D1-P1; In the formula, PB is the preset standard shooting height, PC is the preset standard tilt angle, P1 is the preset standard geometric processing time; α1 is the time consumption coefficient per unit height deviation, in s / m, representing the basic time increment corresponding to each 1m deviation from the standard shooting height; α2 is the time consumption coefficient per unit angle deviation, in s / °, representing the basic time increment corresponding to each 1° deviation from the standard tilt angle; α3 is the coupling coefficient between shooting height and tilt angle, in s / (m·°), reflecting the influence strength of the coupling effect between the two. S13: Following step S12, construct the geometric time-consuming correlation equations for the recorded data A2, A3, ..., Aa in sequence. Combine every three geometric time-consuming correlation equations in the recorded data A1, A2, ..., Aa into a set to generate several sets of ternary simultaneous equations. By performing the solution operation on each set of three simultaneous equations, a solution set corresponding to each set of equations can be obtained. Each solution set contains one solution for a1, a2, and a3 respectively. S14: Extract the values ​​of a1 from all solution sets, process all extracted a1 values ​​using a discrete point filtering algorithm, calculate the average value of all remaining a1 values ​​after data processing, and label the average value as a high-skewness coefficient. Extract the values ​​of a2 from all solution sets, process all extracted a2 values ​​using a discrete point filtering algorithm, and calculate the average value of all remaining a2 values ​​after data processing. Then, label the average value as the angular deflection coefficient. Extract the values ​​of α3 from all solution sets, process all extracted α3 ​​values ​​using a discrete point filtering algorithm, and calculate the average value of all remaining α3 values ​​after data processing. Then, define the average value as the high-inclination coupling coefficient. In this application, the discrete point filtering algorithm can be one of the Z-score filtering algorithm, IQR filtering algorithm, and density filtering algorithm; Based on the high skewness coefficient, angular skewness coefficient and high tilt coupling coefficient, the review standard data is generated and then transmitted to the image review module for storage. It should be noted that the processing and analysis unit stores the recorded data of all overlapping areas of several target regions.

[0014] Some of the data in the above formulas are numerical calculations with dimensions removed, and the contents not described in detail in this specification are all prior art known to those skilled in the art.

[0015] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A deep learning-based intelligent detection system for the overlap quality of waterproof membrane, characterized in that, include: The image review module is used to pre-review the image data of any overlapping area within the target area after receiving it. If the pre-review result is passed, the image data is transmitted to the quality assessment platform. The overlapping area refers to the overlapping and bonding area of ​​adjacent waterproof membranes during the laying of waterproof membrane. A quality assessment platform is used to assess the overlap quality of waterproof membrane in a target area. The quality assessment platform includes an image processing unit and a processing and analysis unit. The image processing unit is used to preprocess the image data of any overlapping area within the target area after receiving it. During the preprocessing process, geometric recognition is performed on the image data to obtain the shooting height and tilt angle of the image data. After the preprocessing is completed, the geometric processing time of the image data is recorded. Recorded data of the image data is generated based on the shooting height, tilt angle and geometric processing time, and the recorded data is transmitted to the processing and analysis unit for storage. The processing and analysis unit is used to analyze the recorded data stored therein after it reaches a preset fixed amount to generate review standard data, and then transmit the review standard data to the image review module for storage.

2. The intelligent detection system for waterproof membrane overlap quality based on deep learning according to claim 1, characterized in that, It also includes an area acquisition module, which is used to receive and collect the image data after the inspector has taken an image of any overlapping area of ​​the waterproof membrane within the target area, and transmit the image data to the image review module.

3. The intelligent detection system for waterproof membrane overlap quality based on deep learning according to claim 1, characterized in that, The quality assessment platform also includes an overlap assessment unit, which pre-stores an overlap assessment strategy. Each time the overlap assessment unit receives a pre-processed image data, it retrieves the overlap assessment strategy and performs feature extraction, target matching, and parameter verification on the standard image data to complete the quality assessment of the overlap area.

4. The intelligent detection system for waterproof membrane overlap quality based on deep learning according to claim 1, characterized in that, If the review fails, a prompt message will be generated based on the overlapping area corresponding to the image data, prompting the inspection personnel to retake the image of the overlapping area.

5. The intelligent detection system for the overlap quality of waterproof membrane based on deep learning according to claim 1, characterized in that, The analysis steps for the processing and analysis unit to analyze and generate audit standard data after the stored record data reaches a preset fixed amount are as follows: S11: Obtain all recorded data stored in the processing and analysis unit, and label them as A1, A2, ..., Aa, where a≥1; S12: Obtain the shooting height, tilt angle and geometric processing time contained in the recorded data A1, label them as B1, C1 and D1 in sequence, and construct the geometric time correlation equation of the recorded data A1 based on them. The geometric time correlation equation is (B1-PB)×ɑ1+(C1-PC)×ɑ2+(B1-PB)(C1-PC)×ɑ3= D1-P1; In the formula, PB is the preset standard shooting height, PC is the preset standard tilt angle, P1 is the preset standard geometric processing time; a1 is the time consumption coefficient per unit height deviation, a2 is the time consumption coefficient per unit angle deviation, and a3 is the coupling coefficient between shooting height and tilt angle. S13: Following step S12, construct the geometric time-consuming correlation equations for the recorded data A2, A3, ..., Aa in sequence. Combine every three geometric time-consuming correlation equations in the recorded data A1, A2, ..., Aa into a set to generate several sets of ternary simultaneous equations. By performing the solution operation on each set of three simultaneous equations, a solution set corresponding to each set of equations can be obtained. Each solution set contains one solution for a1, a2, and a3 respectively. S14: Extract the values ​​of a1 from all solution sets, process all extracted a1 values ​​using a discrete point filtering algorithm, and calculate the average of all remaining a1 values ​​after data processing. Label this average as the high-skewness coefficient. Similarly, extract the values ​​of a2 and a3 from all solution sets respectively. The discrete point filtering algorithm was used to process all the extracted values ​​of a2 and a3 to obtain the angular deflection coefficient and the high tilt coupling coefficient. Audit standard data is generated based on the high skewness coefficient, angle skewness coefficient, and high tilt coupling coefficient.

6. The intelligent detection system for waterproof membrane overlap quality based on deep learning according to claim 5, characterized in that, The pre-screening content is as follows: Perform geometric recognition on the image data to obtain the shooting height D1 and tilt angle F1 of the image data; The geometric time scalar G1 for obtaining the image data is calculated using the formula G1=(D1-PB)×Pɑ1+(F1-PC)×Pɑ2+(D1-PB)(F1-PC)×Pɑ3. G1 and G are compared. If G1>G, the pre-review result is output as passed; otherwise, the pre-review result is output as failed. In the formula, PA1, PA2, and PA3 are the high skewness coefficient, angular skewness coefficient, and high tilt coupling coefficient contained in the review standard data stored in the current image review module, respectively, and G is the preset pre-review judgment threshold.

Images