Intelligent visual detection method and system for blockage of drain pipe of condenser

Abstract

The invention provides an intelligent visual detection method and system for pipe blockage of a condenser drainage pipe, and belongs to the technical field of computer vision, and the method comprises the steps: collecting a local image of a to-be-detected condenser drainage device under illumination, and carrying out the image splicing to obtain a panorama; target identification is carried out through a deep learning model, and position information of a blocked tube and a non-blocked tube is obtained; acquiring image data of an area to be subjected to pipe plugging installation operation, and matching the position of the current image data in the panorama; the method comprises the following steps: acquiring image data of a plugged pipe installation operation area in a current state during execution of plugged pipe installation operation in real time, identifying and marking a plugged pipe area and a non-plugged pipe area through a deep learning model, and marking a position with a problem in an installation state if the identified data are inconsistent. In addition, the invention also provides a method for measuring the height of the drainage pipe orifice by using the laser technology. According to the method, high-precision detection, panoramic mapping and pipe orifice height measurement of the blocked pipe installation state of the condenser drain pipe can be realized.

Description

Technical Field

[0001] The invention relates to the technical field of machine vision, and in particular to an intelligent visual detection method for blockage of a condenser drain pipe. Background Art

[0002] Large condensate drainage pipelines in nuclear power plants are important systems used to discharge condensate generated by steam condensation or other processes in nuclear power plants. These pipelines are usually designed to be quite large to cope with the discharge needs of large amounts of condensate in nuclear power plants. Condensate is mainly produced in steam generators, cooling systems, air conditioning units and other equipment in nuclear power plants. This water condenses into liquid during the cooling process. If it is not discharged in time, it may cause equipment corrosion, performance degradation, and even cause safety problems. Therefore, large condensate drainage pipelines play a vital role in nuclear power plants.

[0003] However, in current industrial applications, the identification of condenser drain pipes and the installation of blocked pipes still rely mainly on manual judgment and manual operation. This traditional method is not only time-consuming and labor-intensive, but also difficult to ensure the accuracy and consistency of installation due to the subjectivity of manual operation. In complex working environments, such as insufficient light, small space, or ambient temperature fluctuations, the efficiency of manual inspection and adjustment is significantly reduced. In addition, manual operation requires high experience from the operator, and the slightest carelessness may cause installation deviations, affecting the operational stability and safety of the entire system. Therefore, the existing manual inspection methods have obvious deficiencies in efficiency and accuracy, and cannot meet the needs of modern industry for efficient and precise installation. Summary of the invention

[0004] In order to solve the above problems, the present invention provides an intelligent visual detection method for blockage of a condenser drain pipe. The method can achieve high-precision detection and panoramic mapping of the blockage installation status of the condenser drain pipe through image acquisition and processing and deep learning technology in a complex environment.

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

[0006] An intelligent visual detection method for blockage of a condenser drain pipe comprises the following steps:

[0007] Based on the principle of orthographic slice imaging, local images of the condenser drainage device to be tested under illumination are collected, and the local images are stitched to obtain a panoramic view of the condenser drainage device to be tested;

[0008] Through the trained deep learning model, the blocked and non-blocked pipes are identified in the panorama, the location information of the blocked and non-blocked pipes is obtained, and the blocked and non-blocked pipes are marked in the panorama to construct an initial panorama with the blocked pipe location information.

[0009] Obtain image data of the area where the pipe plugging installation operation is to be performed, compare it with the initial panoramic image, and match the position of the current image data in the panoramic image;

[0010] The image data of the current state of the pipe blocking installation operation area during the execution of the pipe blocking installation operation is obtained in real time. Through the trained deep learning model, the blocked and non-blocked areas in the overall view are identified and marked. If the identified data is inconsistent with the data in the initial panoramic view, the location where the installation status is problematic is marked.

[0011] Preferably, it also includes:

[0012] Determine the nozzle height test area based on the panoramic image, and collect image data of the nozzle height test area under red and green laser irradiation; after adaptive thresholding the image data, extract the preliminary fringe skeleton mask through morphological calculation to obtain the fringe image;

[0013] The normal vector of the laser stripe in the stripe image is obtained, and the sub-pixel positioning of the laser stripe is performed using the direction-adaptive grayscale centroid method based on the normal vector of the laser stripe to obtain the position information of the laser stripe; the bulge height of the drainage pipe is determined according to the position information of the laser stripe.

[0014] Preferably, the collecting of the local image of the condenser drainage device to be tested under illumination based on the principle of front view slice imaging comprises the following steps:

[0015] Positioning the imaging device above the target area of ​​the condenser drain to be tested so that the optical axis is perpendicular to the scene plane;

[0016] Start taking images from one end of the target area to obtain the first partial image, move the imaging device along a fixed direction to the next position to take images, and then get the last partial image on the straight line after reaching the other end of the target area; move down one area to take images, and continue taking images on the straight line along a fixed direction;

[0017] The shooting is performed repeatedly until the imaging area completely covers the entire target area; wherein adjacent partial images have an overlapping area whose overlapping area is not less than 10% of the partition size.

[0018] Preferably, the step of stitching the local images to obtain a panoramic image of the condenser drainage device to be tested comprises the following steps:

[0019] For each local image, a feature detection algorithm is used to extract feature points in the overlapping area, and feature descriptors of the feature points are calculated;

[0020] Based on the feature descriptors of the feature points, matching feature point pairs are obtained, the transformation matrix between adjacent local images is calculated, and the view of a local image is mapped to the view of an adjacent local image through the transformation matrix to perform image alignment;

[0021] According to the geometric transformation relationship, each aligned partial image is transformed to the same plane, the size and position of the partial image are adjusted to align the overlapping areas, and they are gradually stitched together to form a complete panorama;

[0022] The overlapping areas are processed by using image fusion technology, and color correction and brightness adjustment are performed, including color balance and contrast adjustment, to obtain an adjusted panorama;

[0023] Post-process the panorama, including removing lens distortion, noise filtering, and image enhancement.

[0024] Preferably, matching the position of the current image data in the panoramic image comprises the following steps:

[0025] Extract feature points from image data and generate descriptive feature descriptors for each feature point;

[0026] Through the feature point detection algorithm, significant feature points are found in the image data and described;

[0027] Use the matching algorithm to match the significant feature points in the panoramic image, find similar feature point pairs, and perform feature point matching;

[0028] The transformation matrix is ​​calculated according to the feature point matching information, the coordinate system of the image data is mapped to the coordinate system of the panorama according to the transformation matrix, and the image data is transformed so as to be aligned with the panorama in space.

[0029] Preferably, the recognition of the image data of the pipe plugging installation operation area in the current state during the pipe plugging installation operation comprises the following steps:

[0030] The trained deep learning model is used to identify the blocked pipe in real-time images or the panoramic image obtained by stitching all the images collected in real-time during the blocked pipe installation process. The trained deep learning model is used to output the identification result of the blocked pipe position and confidence score in the current image.

[0031] The identification result is compared with the blocked pipe position identified in the previous panoramic view. If the blocked pipe position installed in the panoramic view is not installed under the installation condition, it is marked with a red box; if the blocked pipe position installed in the panoramic view is also installed under the installation condition, it is marked with a green box.

[0032] Preferably, the height of the drainage pipe protrusion is determined according to the position information of the laser stripes, as shown in the following formula:

[0033]

[0034] Where B is the baseline length from the laser to the camera focus, f is the focal length of the camera, α is the angle between the laser line projection surface and the baseline, and x is the image coordinate of the center point on the extracted laser line.

[0035] An intelligent visual detection system for condenser drain pipe blockage, the system comprising:

[0036] An imaging camera, used for collecting a local image of the drainage device of the condenser to be tested under illumination based on the principle of orthographic slice imaging;

[0037] LED light source, used to provide white light illumination for front-view slice imaging;

[0038] A multi-claw support structure, including symmetrically distributed retractable claw-like extension structures, is used to fix and support the imaging camera to ensure that the working distance remains unchanged;

[0039] Embedded platforms, including:

[0040] An image preprocessing module is used to preprocess the original image acquired by the imaging camera;

[0041] The panoramic mapping and blocked pipe identification module is used to stitch local images to obtain a panoramic view of the condenser drainage device to be tested; it is also used to identify blocked and non-blocked pipe targets in the panoramic view through the trained deep learning model, obtain the location information of blocked and non-blocked pipes, mark them in the panoramic view, and construct a panoramic view with blocked pipe location information;

[0042] The panoramic image registration and marking module is used to register and identify the position of the image to be identified with the panoramic image collected in real time during the installation process. At the same time, the previously pre-trained deep learning model is used to locate, identify and mark the pipe blocking installation position of the image to be identified.

[0043] Preferably, the embedded platform further includes:

[0044] Red and green lasers, laser emission structures symmetrically arranged with the camera optical axis as the center, are used to provide red and green dual-color laser irradiation; the imaging camera is also used to collect image data of the nozzle height to be measured area under red and green laser irradiation;

[0045] The light stripe extraction and positioning module is used to adaptively threshold the image data under red and green laser irradiation, and extract the preliminary stripe skeleton mask through morphological calculation to obtain the stripe map; obtain the normal vector of the laser stripe in the stripe map, and use the direction-adaptive grayscale centroid method to perform sub-pixel positioning on the laser stripe based on the normal vector of the laser stripe to obtain the position information of the laser stripe;

[0046] The three-dimensional shape measurement and calculation module is used to determine the height of the drainage pipe protrusion based on the position information of the laser stripes.

[0047] Preferably, the system further comprises:

[0048] User interaction device: The user interaction device is a touch screen device used to provide image display, panoramic view, recognition results and measurement data display and operation control.

[0049] Beneficial effects of the present invention:

[0050] The present invention proposes an intelligent visual detection method and system for condenser drain pipe blockage. The method can obtain consistent image quality under different lighting conditions through high-resolution image acquisition and active lighting technology, thereby ensuring the reliability of the recognition result; the method adopts real-time imaging of the blocked pipe installation area, and uses image processing algorithms to extract and analyze key features in the image, so as to judge the installation status of the blocked pipe, such as feature points, position, angle, alignment, etc. At the same time, the method combines the target detection and recognition algorithms in deep learning, and the system can automatically determine whether the blocked pipe is installed correctly or whether there is a deviation. According to the recognition results, the system can provide real-time feedback to guide the operator to make adjustments, or directly perform auxiliary installation operations through automated devices, thereby ensuring the accuracy and efficiency of the installation. The system provides a panoramic mapping function so that the operator can fully grasp the installation status of the drain pipe blockage, reducing the necessity of manual inspection. In addition, the real-time image registration and marking technology provided by the method ensures that each step of the installation process can obtain instant feedback, reducing the risk of operational errors, and is suitable for complex industrial environments, providing a more efficient and safe solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 1 is an overall architecture diagram of a condenser drain pipe identification and pipe orifice height measurement system based on machine vision according to an embodiment of the present invention;

[0052] Figure 2 Schematic diagram of the working process of the drainage pipe blockage detection system according to an embodiment of the present invention;

[0053] Figure 3 1 is a schematic diagram of the working process of the drainage pipe orifice height measurement system according to an embodiment of the present invention;

[0054] Figure 4 It is a structural schematic diagram of an intelligent visual detection system for blockage of a condenser drain pipe according to an embodiment of the present invention.

[0055] In the figure, 1. mode toggle switch; 2. interface; 3. green laser; 4. multi-claw support structure; 5. red laser; 6. LED light source; 7. RGB camera. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0057] Example 1

[0058] The intelligent visual detection method and system for condenser drain pipe blockage of the present invention mainly have three working modes: panoramic mapping mode, real-time recognition mode for blocked pipe installation status and pipe orifice height measurement mode, wherein the panoramic mapping mode and real-time recognition mode for blocked pipe installation status belong to the drain pipe blockage detection system, and the pipe orifice height measurement mode belongs to the drain pipe orifice height measurement system. The working mode of the device can be switched by a mode toggle switch 1 integrated on the embedded platform. The overall architecture of the condenser drain pipe recognition and pipe orifice height measurement system based on machine vision is shown in FIG. Figure 1 shown.

[0059] Real-time identification process of pipe blocking installation status, such as Figure 2 As shown, the following steps are included:

[0060] S1: Based on the principle of orthographic slice imaging, local images of the condenser drainage device to be tested under illumination are collected, and the local images are stitched to obtain a panoramic image of the condenser drainage device to be tested.

[0061] S2: Through the trained deep learning model, the blocked and non-blocked pipes are identified in the panoramic image, the location information of the blocked and non-blocked pipes is obtained, and they are marked in the panoramic image to construct an initial panoramic image with the blocked pipe location information.

[0062] S3: Obtain image data of the area where the pipe plugging installation operation is to be performed, compare it with the initial panoramic image, and match the position of the current image data in the panoramic image.

[0063] S4: Through the trained deep learning model, the blocked and non-blocked areas in the overall view are identified and marked. If the identified data is inconsistent with the data in the initial panoramic view, the location with problematic installation status is marked.

[0064] Drain pipe outlet height measurement, such as Figure 3As shown, the following steps are included:

[0065] S1: Determine the area to be measured for the pipe mouth height according to the panoramic image, and collect image data of the area to be measured for the pipe mouth height under the irradiation of red and green lasers.

[0066] S2: After adaptive thresholding of the image data, a preliminary fringe skeleton mask is extracted through morphological calculation to obtain a fringe map;

[0067] S3: Obtain the normal vector of the laser stripe in the stripe image, perform sub-pixel positioning of the laser stripe using a direction-adaptive grayscale centroid method based on the normal vector of the laser stripe, and obtain the position information of the laser stripe; determine the bulge height of the drain pipe according to the position information of the laser stripe.

[0068] S4: Drain pipe blockage detection system and drain pipe orifice height measurement system. The drain pipe blockage detection system can realize the functions of panoramic mapping of the condenser plane and identifying and recording the blockage position of the drain pipe; the drain pipe orifice height measurement system calculates the height of the protruding part of the drain pipe orifice, thereby realizing the measurement of the protruding height of the orifice.

[0069] like Figure 2 and Figure 3 As shown, the machine vision-based condenser drain pipe recognition and pipe mouth height measurement subsystem includes but is not limited to eleven function implementation modules: imaging camera, LED light source, red and green laser, embedded platform, image preprocessing module, panoramic mapping and pipe blockage recognition module, panoramic image registration and marking module, laser stripe extraction and positioning module, three-dimensional computing module, high-speed storage module and user interaction device, i.e., touch screen.

[0070] As the core imaging device, the imaging camera is responsible for acquiring high-resolution images. These images provide the basis for the subsequent processing and analysis of the system, and can accurately identify the blocked pipe installation and pipe opening height of the condenser drain pipe.

[0071] LED light sources provide a uniform and stable lighting environment, ensuring that the imaging camera can capture clear, shadow-free images under various lighting conditions. High-quality light sources are critical to image quality and directly affect the accuracy of subsequent image processing.

[0072] The red and green lasers are responsible for generating laser marks for measurement by illuminating the surface of the object in the drainage pipe height measurement system. These laser marks are captured by the imaging camera, and the corresponding laser coordinate information is extracted. Through the precise analysis of these laser coordinates, the system can calculate the height information of the drainage pipe relative to the plane of the device.

[0073] The embedded platform is the computing core of the system. It uses a single-board computer with a high-performance SoC processor, including but not limited to Raspberry Pi 5, NVIDIA Orin, Luban Cat 5 and other embedded core boards running Linux system and computing chips such as FPGA. This type of embedded platform not only contains a powerful CPU, but also integrates a GPU, which can provide excellent computing performance and graphics processing capabilities. Some models focus on strengthening AI computing power and significantly improve their deep learning reasoning capabilities. It not only performs reasoning and three-dimensional shape measurement calculations of deep learning models, but also controls the coordination between various modules, providing the real-time processing capabilities required by the system.

[0074] The image preprocessing module focuses on a series of processing on the raw images from the imaging camera, such as denoising, contrast enhancement, and geometric correction. Through these steps, the quality and clarity of the image are significantly improved, laying a solid foundation for subsequent analysis.

[0075] The panoramic mapping and pipe blockage identification module is one of the core modules of the drainage pipe blockage detection system. The panoramic image of the condenser drainage pipe is generated by image stitching technology, and the deep learning model is applied on this basis to identify and store the locations of blocked and non-blocked pipes. This module enables the system to fully grasp the overall condition of the drainage pipe and accurately locate the location of the blocked pipe, which improves the comprehensiveness and accuracy of the detection.

[0076] The panoramic image registration and marking module is the key module for the drainage pipe blockage detection system to identify the real-time blockage status. By accurately aligning the real-time acquired image with the existing panoramic image, and using a deep learning model to dynamically monitor and mark the blockage location. This process ensures real-time comparison between the panoramic image and the real-time image, provides accurate and timely feedback information, and facilitates operators to make immediate judgments and processing during the installation of the blockage.

[0077] The laser stripe extraction and positioning module is a key module for the drainage pipe mouth height measurement system to realize the height measurement function. Due to the diffuse reflection of the object surface, the laser stripe image obtained by the camera often occupies several pixels. Therefore, it is necessary to extract the laser stripe image and locate the center point, and then calculate the three-dimensional data in combination with the system calibration information. Its basic workflow is: first perform OTSU adaptive thresholding, and then perform morphological calculation to obtain a preliminary stripe skeleton mask. Next, the Gaussian kernel function is used to convolve along the extracted initial stripes, and then the second-order partial derivatives at each pixel are calculated to obtain the second-order structure matrix. The maximum eigenvalue of the structure matrix is ​​the normal vector of the calculation point. Finally, the direction-adaptive grayscale centroid method is used for positioning. The window direction is determined by the normal vector, and finally the stripe center line with sub-pixel accuracy is obtained.

[0078] The three-dimensional calculation module is the key module to realize the drainage pipe orifice height measurement system to finally obtain the orifice height information. The height calculation formula of the current measurement point is as follows:

[0079]

[0080] In the formula, B is the baseline length between the laser and the camera focus, f is the focal length of the camera, α is the angle between the laser line projection surface and the baseline, and x is the image coordinate of the center point on the extracted laser line. By combining the camera and laser parameters with the extracted laser stripe center line position coordinates for calculation, nanometer-level height measurement accuracy is achieved to ensure that the data of each raised part is accurate. At the same time, a non-contact measurement method is realized, which reduces the difficulty of operation and avoids any damage to the measurement surface.

[0081] The high-speed storage module is responsible for storing the panoramic images and identification information of blocked and non-blocked pipes generated by the panoramic mapping and blocked pipe identification modules. The module uses high-performance storage media and has fast data writing and reading capabilities to ensure timely archiving and retrieval of image data and identification information. Through the high-speed storage module, the system can achieve efficient and stable data processing, ensuring the accuracy and reliability of the entire detection process.

[0082] Finally, the touch screen, as the interactive interface between the user and the system, provides an intuitive operating experience. Through the touch screen, users can view the panoramic view, identify results and measurement data, and perform related operational controls. This greatly enhances the ease of use of the system, enabling it to be quickly deployed and applied in complex industrial environments.

[0083] The working process of the drainage pipe blockage detection system is as follows Figure 2 The working process of the drainage pipe height measurement system is as follows: Figure 3 shown.

[0084] The working process of the condenser drain pipe identification and pipe orifice height measurement system based on machine vision proposed by the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0085] like Figure 4As shown, in this example, the LED light source 6 is an LED ring white light source, which is installed around the RGB camera 7 and is responsible for supplementing the imaging plane. The multi-claw support structure is a multi-claw support structure 4, which is erected on the plane to be measured and fixed to the camera. Specifically, the multi-claw support structure 4 is a retractable claw-shaped outward extension structure symmetrically distributed under the device. When extended, it contacts the drainage pipe structure plane to be photographed, and can achieve the optimal working distance of the imaging device. The design of this multi-claw support structure 4 has many advantages. A three-claw structure is adopted in this embodiment. First, due to the retractability of the claw-shaped structure, the device can be stably supported on drainage pipes of different heights and positions, while ensuring that the imaging device is always at a standard working distance, so as to obtain a clear and accurate image. In addition, the non-contact operation of the structure reduces direct interference with the drainage pipe structure, reduces the potential risk of damage to the equipment or the object to be photographed, and makes the device more suitable for application in various industrial environments. The red and green lasers are divided into a green laser 3 and a red laser 5, which are distributed on both sides of the camera with the camera optical axis as the symmetry axis. The RGB camera 7 is connected to the interface 2 of the embedded computing platform via a USB or network cable, and the embedded platform is responsible for powering the camera and receiving image data from the RGB camera 7. The embedded platform is also connected to the touch screen via a cable and supplies power to it.

[0086] (1) Panoramic Mapping Mode:

[0087] The execution steps of the panoramic mapping mode are as follows: Figure 2 As shown, the system will automatically turn off the laser in panoramic mapping mode.

[0088] S1: The imaging device needs to be set above the target to be measured, and the optical axis needs to be perpendicular to the scene plane. The working distance of the camera needs to be fixed by a multi-claw support structure. This step ensures the verticality and accuracy of the image and prevents image distortion caused by angle and working distance issues.

[0089] S2: Install LED light sources around the RGB cameras to fill in the light for the imaging plane. Users need to complete the imaging of the entire plane according to the requirements of the orthographic slice imaging described above. The image data captured by these RGB cameras will be transmitted to the embedded platform in turn.

[0090] S3: After receiving these images, the image preprocessing module in the embedded platform will perform preliminary processing on them, including basic image optimization operations such as denoising, enhancement and information fusion. The purpose of these processing steps is to improve the quality of the image, remove noise, enhance the contrast and details of the image, make the image clearer and suitable for subsequent processing steps.

[0091] S4: The pre-processed images will then enter the panoramic mapping and pipe blockage identification module. In this stage, the first thing to do is to extract and match the features of the overlapping parts of the slices. The system extracts key feature points in the image and matches these feature points to find the overlapping areas between the images. Based on the results of these feature matching, the system will stitch multiple images into a complete panoramic image. The effect of this step is to generate a continuous, seamless panoramic view that can fully display the scene.

[0092] S5: Use the trained deep learning model to identify blocked and non-blocked pipes in the stitched panorama. The deep learning model can accurately identify the location and features of blocked pipes in the image by learning a large amount of image data, extract specific blocked pipe information from the panorama, and obtain corresponding data information.

[0093] S6: This data will be transferred to the touch screen for users to check the recognition results and the final effect of the panorama. The touch screen provides users with an intuitive interactive interface, through which users can view the recognition results in detail, including the panorama and the specific location and recognition results of the blockage. Users can choose to save these recognition contents. If the results are not ideal, they can also choose to discard the poor results and regenerate or recognize them. This step ensures that the final saved data is accurate and high-quality.

[0094] S7: After identification and confirmation, the system will save the panoramic image and the blocked pipe location information to a high-speed storage device that is responsible for storing the panoramic image and the blocked pipe location information. This high-speed storage device can be regarded as a database that is specifically used to store processed and analyzed image data and related location information.

[0095] S8: Users can view and operate the stored information through the touch screen display and operation module. The design of the touch screen ensures that users can easily view the saved panoramic images and pipe blockage location information, and perform data management and query operations. The final effect of this process is to achieve a complete closed loop of image data from acquisition, processing to storage and user interaction, ensuring the accuracy, integrity and ease of use of the data.

[0096] Furthermore, the algorithm implementation of the panoramic image stitching technology includes the following steps:

[0097] S1: For each captured local image, a feature detection algorithm (such as SIFT, SURF, ORB, etc.) is used to extract feature points in the overlapping area and calculate their feature descriptors.

[0098] S2: Calculate the transformation relationship between the images so that the images can be aligned. According to the matched feature point pairs, calculate the transformation matrix (such as the homography matrix), which is used to map the coordinate system of one imaging area to the coordinate system of the adjacent imaging area.

[0099] S3: After all local images are aligned, all images are projected into the same reference frame to complete image registration. According to the geometric transformation relationship, each image is transformed to the same plane, and the size and position of the images are adjusted to align their overlapping areas, and they are gradually stitched together to form a complete panoramic image.

[0100] S4: In order to eliminate the transition effect at the seam, image fusion technology is used to process the overlapping area. Commonly used methods include weighted averaging, multi-resolution fusion, pyramid decomposition, and Laplacian pyramid fusion.

[0101] S5: In order to ensure the consistency of the panoramic image, color correction and brightness adjustment are required, including color balance and contrast adjustment to improve the visual effect.

[0102] S6: Post-processing steps can improve the quality of panoramic images, including removing lens distortion, noise filtering, and image enhancement to improve image details.

[0103] (2) Real-time identification mode of pipe blocking installation status:

[0104] The real-time identification process of pipe blocking installation status is as follows Figure 2 As shown, the system will automatically turn off the laser in the real-time identification mode of pipe blocking installation status.

[0105] S1: The user needs to set the imaging device above the area where the pipe plugging installation is currently being performed, and ensure that the optical axis is perpendicular to the scene plane, and fix the working distance of the camera through the multi-claw support structure. The imaging camera will capture the current image data and transmit the captured image to the embedded platform for processing through the image sensor.

[0106] S1: In the image preprocessing module, these raw images will go through a series of processing steps such as noise reduction and enhancement to prepare for the subsequent panorama registration and labeling module. This step ensures that the image is clear and rich in details, providing high-quality data for subsequent processing.

[0107] S2: The processed image enters the panorama registration and marking module, where the pre-processed image is first compared with the previously saved initial panorama image data. The registration technology ensures that the current image is consistent with the initial panorama and accurately matches the position of the current image in the panorama. Next, the system calls the trained deep learning model to identify and mark the blocked and non-blocked areas in the current image. If the identified data is inconsistent with the data in the initial panorama, the system will highlight the location where the installation status is problematic and display the results on the touch screen. In this way, users can immediately find and locate the problem area for timely processing and correction.

[0108] Users can also choose to generate a new panoramic image from the images taken in the pipe blocking installation auxiliary mode. The imaging of the entire plane is completed according to the aforementioned orthographic slice imaging workflow, and the generated panoramic image will be compared with the initial panoramic image. If there is a problem with the identified installation status, the system will highlight the problem location and display the result on the touch screen. In this way, users can intuitively observe and judge the current installation status in a panoramic view, which is convenient for comprehensive inspection and evaluation.

[0109] The panoramic image registration and marking module realizes the registration and position recognition of the image to be identified and the panoramic image collected in real time during the installation process, and uses the pre-trained deep learning model to locate the pipe blocking installation position and identify and mark the state of the image to be identified. The algorithm specifically includes the following steps:

[0110] S1: Extract meaningful feature points from the image and generate descriptive feature descriptors for each feature point. Through feature point detection algorithms, such as SIFT or SURF introduced earlier, it is possible to find significant feature points in the image and describe them. Next, the feature points in the current image are matched in the panorama through a matching algorithm to find similar feature point pairs. The ultimate goal of this stage is to find feature point pairs that accurately describe the correspondence between images, laying the foundation for subsequent image alignment.

[0111] S2: After the feature points are matched, the next task is to calculate the transformation matrix and apply it to the images to achieve image registration. In this stage, the information obtained from the matching pairs is used to calculate a transformation matrix that maps the coordinate system of one image to the coordinate system of the other image. After this, the image needs to be actually transformed to align it with the panoramic image in space.

[0112] S3: After completing the panoramic image registration, the blocked pipe positions of the two states are in an overlapping state, and the blocked pipe detected in the second step is also converted to the panoramic coordinate system, realizing the blocked pipe positioning. Next, the previously trained deep learning model is used to identify the blocked pipe for the pictures taken during the blocked pipe installation condition or the images collected in real time. The deep learning model will output the position and confidence score of the blocked pipe in the current picture. These recognition results are compared with the blocked pipe positions that have been identified in the previous panoramic image. If the blocked pipe position that has been installed in the panoramic image is not installed under the installation condition (that is, it is identified as a position to be installed), it is marked with a red box; if the blocked pipe position that has been installed in the panoramic image is also installed under the installation condition (that is, it is identified as an installed position), it is marked with a green box.

[0113] (2) Nozzle height measurement mode:

[0114] The process of measuring nozzle height is as follows: Figure 3 As shown in the figure, the system will automatically turn on the laser in the nozzle height measurement mode.

[0115] S1: The user moves the imaging device above the plane to be measured and fixes the working distance of the camera through the multi-claw support structure. The red laser and the green laser will illuminate the surface of the drainage device. At this time, it is necessary to note that the imaging camera should collect lasers of different wavelengths reflected from the surface of the object at a normal viewing angle. Ensure that the imaging position of the laser in the imaging camera is accurate to ensure the accuracy of the subsequent measurement result calculation.

[0116] S2: The image captured by the imaging camera is transmitted to the embedded platform through a high-speed interface. The image preprocessing module of the embedded platform first processes the captured image, including denoising and OTSU adaptive thresholding. Then, morphological calculation is performed to obtain a preliminary stripe skeleton mask to complete the preliminary extraction of the laser stripes. This step improves the image quality and lays a solid foundation for further extraction and positioning of the laser stripes.

[0117] S3: The preprocessed image enters the laser stripe extraction and positioning module. After calculating the normal vector of the stripe image, the module uses the direction-adaptive grayscale centroid method to perform sub-pixel positioning of the laser stripes, thereby obtaining accurate stripe position information. This step ensures the accuracy of the stripe position through a high-precision positioning algorithm, providing reliable data support for subsequent three-dimensional calculations.

[0118] S4: The 3D calculation module performs complex 3D calculations based on the extracted and located information. This step involves a lot of data processing and calculations to ensure that the system can generate accurate 3D models and positioning information. This 3D information can be used to measure key parameters such as the height of the drainage pipe protrusion, which is of great significance for actual operation and maintenance. Through accurate 3D measurement, the system can provide detailed morphological information to help users make accurate judgments and decisions.

[0119] S5: All processing results and three-dimensional information are presented to the user through the touch screen display and operation module. The touch screen serves as a human-computer interaction interface, and the user can view the image processing results in real time, adjust the camera and light source parameters, monitor the system operation status, and perform other operations on the touch screen.

[0120] This embodiment is based on the integration of multiple advanced technologies, mainly including high-resolution industrial cameras, active lighting sources, red and green lasers, and embedded computing platforms. These devices work together to achieve accurate detection of the blockage status and pipe mouth height of the condenser drain pipe through image acquisition, processing, and analysis. After the image acquired by the imaging camera is processed by the image preprocessing module, the blockage position is identified and marked through the deep learning model. At the same time, using red and green lasers and three-dimensional morphology measurement modules, the system can accurately measure the height information of the drain pipe mouth, and ensure that the imaging device is always at the optimal working distance through a multi-claw support structure. In addition, the embedded computing platform provides powerful computing power, supports real-time image processing and panoramic image generation, enabling the system to perform precise detection and mapping tasks in complex industrial environments.

[0121] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An intelligent visual detection method for condenser drain pipe blockage, characterized in that: The following steps are involved: Based on the principle of orthographic slice imaging, local images of the condenser drainage device to be tested under illumination are collected, and the local images are stitched to obtain a panoramic view of the condenser drainage device to be tested; Through the trained deep learning model, the blocked and non-blocked pipes are identified in the panorama, the location information of the blocked and non-blocked pipes is obtained, and the blocked and non-blocked pipes are marked in the panorama to construct an initial panorama with the blocked pipe location information. Obtain image data of the area where the pipe plugging installation operation is to be performed, compare it with the initial panoramic image, and match the position of the current image data in the panoramic image; The image data of the current state of the pipe blocking installation operation area during the execution of the pipe blocking installation operation is obtained in real time. The blocked pipe and non-blocked pipe areas are identified and marked through the trained deep learning model. If the identified data is inconsistent with the data in the initial panoramic image, the location where the installation status is problematic is marked.

2. The intelligent visual detection method for condenser drain pipe blockage according to claim 1 is characterized in that: Also includes: Determine the nozzle height test area based on the panoramic image, and collect image data of the nozzle height test area under red and green laser irradiation; after adaptive thresholding the image data, extract the preliminary fringe skeleton mask through morphological calculation to obtain the fringe image; The normal vector of the laser stripe in the stripe image is obtained, and the sub-pixel positioning of the laser stripe is performed using the direction-adaptive grayscale centroid method based on the normal vector of the laser stripe to obtain the position information of the laser stripe; the bulge height of the drainage pipe is determined according to the position information of the laser stripe.

3. The intelligent visual detection method for condenser drain pipe blockage according to claim 1 is characterized in that: The method of collecting a local image of the condenser drainage device to be tested under illumination based on the principle of frontal slice imaging includes the following steps: Positioning the imaging device above the target area of ​​the condenser drain to be tested so that the optical axis is perpendicular to the scene plane; Start taking images from one end of the target area to obtain the first partial image, move the imaging device along a fixed direction to the next position to take images, and then get the last partial image on the straight line after reaching the other end of the target area; move down one area to take images, and continue taking images on the straight line along a fixed direction; The shooting is performed repeatedly until the imaging area completely covers the entire target area; wherein adjacent partial images have an overlapping area whose overlapping area is not less than 10% of the partition size.

4. The intelligent visual detection method for condenser drain pipe blockage according to claim 1 is characterized in that: The method of performing image stitching on the local images to obtain a panoramic image of the condenser drainage device to be tested includes the following steps: For each local image, a feature detection algorithm is used to extract feature points in the overlapping area, and feature descriptors of the feature points are calculated; Based on the feature descriptors of the feature points, matching feature point pairs are obtained, the transformation matrix between adjacent local images is calculated, and the view of a local image is mapped to the view of an adjacent local image through the transformation matrix to perform image alignment; According to the geometric transformation relationship, each aligned partial image is transformed to the same plane, the size and position of the partial image are adjusted to align the overlapping areas, and they are gradually stitched together to form a complete panorama; The overlapping areas are processed by using image fusion technology, and color correction and brightness adjustment are performed, including color balance and contrast adjustment, to obtain an adjusted panorama; Post-process the panorama, including removing lens distortion, noise filtering, and image enhancement.

5. The intelligent visual detection method for condenser drain pipe blockage according to claim 1 is characterized in that: The step of matching the position of the current image data in the panoramic image comprises the following steps: Extract feature points from image data and generate descriptive feature descriptors for each feature point; Through the feature point detection algorithm, significant feature points are found in the image data and described; Use the matching algorithm to match the significant feature points in the panoramic image, find similar feature point pairs, and perform feature point matching; The transformation matrix is ​​calculated according to the feature point matching information, the coordinate system of the image data is mapped to the coordinate system of the panorama according to the transformation matrix, and the image data is transformed so as to be aligned with the panorama in space.

6. The intelligent visual detection method for condenser drain pipe blockage according to claim 5 is characterized in that: The recognition of image data of the current state of the pipe plugging installation operation area during the pipe plugging installation operation includes the following steps: The trained deep learning model is used to identify the blocked pipe in real-time images or the panoramic image obtained by stitching all the images collected in real-time during the blocked pipe installation process. The trained deep learning model is used to output the identification result of the blocked pipe position and confidence score in the current image. The identification result is compared with the blocked pipe position identified in the previous panoramic view. If the blocked pipe position installed in the panoramic view is not installed under the installation condition, it is marked with a red box; if the blocked pipe position installed in the panoramic view is also installed under the installation condition, it is marked with a green box.

7. The intelligent visual detection method for condenser drain pipe blockage according to claim 2 is characterized in that: The height of the drainage pipe protrusion is determined according to the position information of the laser stripes, as shown in the following formula: Where B is the baseline length from the laser to the camera focus, f is the focal length of the camera, α is the angle between the laser line projection surface and the baseline, and x is the image coordinate of the center point on the extracted laser line.

8. An intelligent visual detection system for condenser drain pipe blockage, characterized in that: The system comprises: An imaging camera, used for collecting a local image of the drainage device of the condenser to be tested under illumination based on the principle of orthographic slice imaging; LED light source, used to provide white light illumination for front-view slice imaging; A multi-claw support structure, including symmetrically distributed retractable claw-like extension structures, is used to fix and support the imaging camera ， Ensure that the working distance remains unchanged; Embedded platforms, including: An image preprocessing module is used to preprocess the original image acquired by the imaging camera; The panoramic mapping and blocked pipe identification module is used to stitch local images to obtain a panoramic view of the condenser drainage device to be tested; it is also used to identify blocked and non-blocked pipe targets in the panoramic view through the trained deep learning model, obtain the location information of blocked and non-blocked pipes, mark them in the panoramic view, and construct a panoramic view with blocked pipe location information; The panoramic image registration and marking module is used to register and identify the position of the image to be identified with the panoramic image collected in real time during the installation process. At the same time, the previously pre-trained deep learning model is used to locate, identify and mark the pipe blocking installation position of the image to be identified.

9. The intelligent visual detection system for condenser drain pipe blockage according to claim 8, characterized in that: The embedded platform also includes: Red and green lasers, laser emission structures symmetrically arranged with the camera optical axis as the center, are used to provide red and green dual-color laser irradiation; the imaging camera is also used to collect image data of the nozzle height to be measured area under red and green laser irradiation; The light stripe extraction and positioning module is used to adaptively threshold the image data under red and green laser irradiation, and extract the preliminary stripe skeleton mask through morphological calculation to obtain the stripe map; obtain the normal vector of the laser stripe in the stripe map, and use the direction-adaptive grayscale centroid method to perform sub-pixel positioning on the laser stripe based on the normal vector of the laser stripe to obtain the position information of the laser stripe; The three-dimensional shape measurement and calculation module is used to determine the height of the drainage pipe protrusion based on the position information of the laser stripes.

10. The intelligent visual detection system for condenser drain pipe blockage according to claim 8, characterized in that: The system further comprises: User interaction device: The user interaction device is a touch screen device used to provide image display, panoramic view, recognition results and measurement data display and operation control.

Images