Image acquisition method and device for gas pipeline

Abstract

The application provides a kind of image acquisition method and device for gas pipeline, relating to gas pipeline security and video monitoring technical field.The application includes explosion-proof housing, black light full-color camera module, window self-cleaning structure, flexible press-fit sealing structure and inner cavity humidity balance structure, through micro-angle ring-shaped eaves and hydrophobic flow guide groove to form Venturi effect to realize window active cleaning, use floating flexible press-fit structure to offset temperature stress deformation, and through independent desiccant bin to realize inner cavity humidity regulation, to guarantee the imaging clarity under micro-light and extreme weather from the hardware level, in addition, on this basis, through night blur correction algorithm, the collected image is enhanced and error corrected at pixel level, target feature extraction and classification identification are carried out for mobile excavator and tool-holding damage personnel, and double-mode verification is carried out by synchronously fusing voiceprint and vibration sensing data to eliminate identification error.The application can at least improve pipeline detection efficiency.

Description

Technical Field

[0001] This invention relates to data processing technology, and more particularly to an image acquisition method and apparatus for gas pipelines. Background Technology

[0002] As a crucial infrastructure for urban energy transmission, the stability and safety of gas pipelines have always been a major concern. It's worth noting that during long-term natural gas transmission, trace amounts of heavy oil and moisture in the gas, along with the pipeline's own microscopic oxidation, cause various deposits to accumulate on the inner walls of gas pipelines over time, forming deposits such as tar clumps, rust flakes, or sludge. These deposits not only reduce the effective flow cross-sectional area of ​​the pipeline and affect normal gas transmission efficiency but may also lead to localized corrosion and perforation risks. Therefore, regular image inspection and cleaning of the interior of gas pipelines are essential.

[0003] Currently, the industry's traditional approach to inspecting and cleaning deposits inside gas pipelines mainly relies on manual collection and testing. Specifically, it usually requires excavation or opening of inspection ports at specific pipeline nodes, where inspectors use handheld image acquisition devices such as endoscopes to insert into the pipeline for inspection. Alternatively, in some large-diameter pipelines, manual inspection can be carried out directly, followed by manual cleaning using high-pressure water guns or long-handled scrapers.

[0004] However, the inventors discovered in their research that this manual collection and cleaning method suffers from technical flaws in practical applications, including vague target positioning, indiscriminate cleaning, and a tendency to leave residues. During manual operations, limited by the short viewing distance of the detection equipment and the narrow, dimly lit environment inside the pipes, operators typically only obtain a two-dimensional image from a limited perspective. This makes it difficult to accurately perceive the three-dimensional shape, thickness, and edge adhesion of the dirt on the pipe wall surface. Operators cannot accurately locate the weakest point where the deposits are bonded to the pipe wall for targeted removal. They often rely on experience to use indiscriminate fluid flushing or blind mechanical scraping, making it difficult to completely remove stubborn dirt in one go. This easily leaves patchy residues on the pipe wall, and the indiscriminate mechanical scraping can cause unnecessary physical wear on the gas pipe wall itself, making it difficult to effectively guarantee the overall descaling efficiency and pipeline maintenance quality.

[0005] Therefore, there is an urgent need to provide an image acquisition method and device for gas pipelines that can improve pipeline inspection efficiency. Summary of the Invention

[0006] In view of the above problems, the present invention is proposed to provide an image acquisition method and apparatus for gas pipelines that overcomes or at least partially solves the above problems.

[0007] According to one aspect of the present invention, an image acquisition method for gas pipelines is provided, comprising the following steps: Acquire a first three-dimensional image of the target pipeline area corresponding to the gas pipeline, and identify the target attachments existing in the target pipeline area by recognizing the first three-dimensional image. When the target attachment is determined to have cleaning properties based on the type of attachment, the edge features of the corresponding target attachment are obtained based on the first three-dimensional image, and the peeling and cleaning points of the corresponding target attachment are determined based on the edge features; The control cleaning actuator performs cleaning operations on the target deposits based on the stripping and cleaning points, and responds to complete the cleaning operation to obtain a second three-dimensional image of the corresponding target pipeline area; Based on the second three-dimensional image, the residual attachments of the corresponding target attachments are determined, and the scraping actuator is controlled to scrape the residual attachments.

[0008] Optionally, in the method according to the invention, identifying the first three-dimensional image to determine the target attachment existing in the target pipe region includes: Determine the inherent construction structures and their corresponding locations within the target pipeline area based on the as-built drawings; When the target pipeline area is determined to have at least a portion of inherent construction structure based on the structural location, the construction structure area indicating the at least portion is determined based on the first three-dimensional image, and the remaining portion is determined as the pipe wall structure area indicating the pipe wall structure. The elevation values ​​of each point corresponding to each image pixel in the construction structure area are compared with the three-dimensional point cloud data of the corresponding inherent construction structure obtained based on the as-built drawings to obtain the elevation deviation values. If any elevation deviation value exceeds a preset tolerance threshold, the image pixel corresponding to that elevation deviation value will be identified as an abnormal pixel. Based on the as-built drawings, the reference pixel value of the corresponding pipe wall structure is determined, and if the pixel value of any image pixel in the area that makes up the pipe wall structure is different from the reference pixel value, the image pixel is identified as an abnormal pixel. Based on the first 3D image, all adjacent abnormal pixels are connected to determine the target attachments existing in the target pipeline region based on the obtained abnormal sub-regions.

[0009] Optionally, in the method according to the present invention, determining an image pixel as an abnormal pixel in response to a pixel value difference between any image pixel value constituting the pipe wall structure region and a reference pixel value includes: Based on the as-built drawings, the curvature distribution of the corresponding pipe wall structure is determined, and the pipe wall structure is divided based on the curvature distribution to obtain sub-regions corresponding to different curvature intervals. The pixel compensation coefficients corresponding to each curvature interval are retrieved and the base pixel values ​​are adjusted to obtain the dynamic pixel ranges corresponding to different sub-regions. If the pixel value of an image pixel located in any sub-region falls within the dynamic pixel range corresponding to that sub-region, the image pixel is identified as a pipe wall pixel; otherwise, it is identified as an abnormal pixel.

[0010] Optionally, in the method according to the present invention, when it is determined that the target attachment has cleaning properties based on the attachment type, the edge features of the corresponding target attachment are obtained based on the first three-dimensional image, and the peeling and cleaning points of the corresponding target attachment are determined according to the edge features, including: The region outline of the corresponding abnormal sub-region is divided to obtain the array points; Starting from the center point of the corresponding abnormal sub-region, generate shape representation lines connecting each array point, and determine the shape variance value of the corresponding abnormal sub-region based on the representation length obtained from each shape representation line. Based on all image pixels that make up the abnormal sub-region, determine the maximum and minimum pixel values ​​of the corresponding pixels, and determine the pixel distribution interval formed by the maximum and minimum pixel values; Based on the shape variance value and pixel distribution range of the corresponding abnormal sub-region, the preset type table is traversed to determine the attachment type of the corresponding target attachment. When the target attachment is determined to have cleaning properties based on the attachment type, the image pixels that make up the region contour are determined as contour pixels, and each shape characterization line is controlled to extend to the image pixels located outside the abnormal sub-region and adjacent to the contour pixels, so as to obtain each edge extension line. The edge features of the corresponding target attachments are determined based on the elevation change trend obtained from each edge extension line, and the peeling and cleaning points of the corresponding target attachments are determined based on the edge features.

[0011] Optionally, in the method according to the present invention, determining the attachment type of the corresponding target attachment by traversing a preset type table based on the shape variance value and pixel distribution interval of the corresponding abnormal sub-region includes: Retrieve a preset type table, wherein the preset type table includes standard mapping groups corresponding to different standard types, and each standard mapping group includes different standard variance values ​​and standard distribution intervals; If the response shape variance, pixel distribution interval and the standard variance of any standard type, the variance similarity between the standard distribution interval and the standard type are all greater than the preset similarity threshold, then the standard type is determined as the candidate type. There is only one candidate response type, and that candidate type is determined as the attachment type of the corresponding target attachment. If there are multiple candidate types and at least one candidate type has a cleaning attribute, then any candidate type with a cleaning attribute is determined as the attachment type of the corresponding target attachment.

[0012] Optionally, in the method according to the present invention, determining the edge features of the corresponding target attachment based on the elevation change trend obtained from each edge extension line, and determining the peeling and cleaning points of the corresponding target attachment according to the edge features, includes: The difference in elevation values ​​of adjacent image pixels arranged along the same edge extension line is calculated to obtain an elevation sequence composed of the elevation extension differences. The elevation extension difference at the end of the high program column is determined as the edge extension difference, and at least two edge extension differences are less than the preset fitting threshold. Based on the edge extension line corresponding to the edge extension difference, the abnormal sub-region is divided to obtain each cleaning and stripping partition. The fitting evaluation value for each edge extension line corresponding to the same cleaning and stripping zone is determined based on the edge extension difference, and the slope evaluation value is determined based on the height difference between the maximum and minimum elevation values ​​corresponding to the edge extension line. Based on the high program sequence, all image pixels whose elevation extension difference shows a continuous increase are identified as increased pixels, and the number of interval pixels corresponding to all image pixels located between adjacent increased pixels are determined, so as to determine the change evaluation value based on the number of increased pixels and the number of interval pixels corresponding to all increased pixels. The fitting evaluation value, slope evaluation value and change evaluation value corresponding to the same edge extension line are weighted and summed to obtain the comprehensive evaluation value of each contour pixel. The contour pixel with the largest comprehensive evaluation value for each cleaning and stripping partition is determined as the stripping and cleaning point.

[0013] Optionally, in the method according to the invention, controlling the cleaning actuator to perform cleaning operations on the target deposits based on the stripping and cleaning points includes: Based on the 3D point cloud data of the corresponding pipe wall structure obtained from the as-built drawings, the center of the pipe wall section of the corresponding target pipe area is determined, and the reference axis of the corresponding target pipe area is constructed based on the gravity direction and the center of the pipe wall section. Determine the spatial line connecting each stripping and cleaning point to the center of the pipe wall cross-section, and determine the circumferential deflection angle of each spatial line corresponding to the reference axis. All stripping and cleaning points are arranged in ascending order according to the circumferential deflection angle from smallest to largest. Based on the obtained stripping execution sequence, the cleaning execution mechanism is controlled to perform cleaning operations on each stripping and cleaning point in sequence with a preset rinsing force.

[0014] Optionally, in the method according to the present invention, the cleaning actuator is controlled to perform cleaning operations on each stripping and cleaning point sequentially with a preset rinsing force based on the obtained stripping execution sequence, including: Based on the stripping execution sequence, each stripping and cleaning point is determined as the current cleaning point, and other stripping and cleaning points located after and adjacent to the current cleaning point are determined as future cleaning points. The reference incident angle of the corresponding edge extension line is determined based on the point elevation value of the current cleaning point. Based on the current location of the cleaning point, the regional curvature distribution of the corresponding pipe wall structure, and the direction of gravity, a runoff prediction surface is constructed, and the interference distance of the runoff prediction surface corresponding to the future cleaning point is obtained. If the response interference distance is greater than the preset safety threshold, the control cleaning actuator will perform cleaning operation on the current stripping and cleaning point based on the reference incident angle; otherwise, the runoff diversion vector will be determined based on the interference distance and the future position of the corresponding future cleaning point. The reference incident angle is corrected based on the runoff guide vector, and the cleaning actuator is controlled to perform cleaning operation on the current stripping and cleaning point with a preset flushing force based on the obtained corrected incident angle.

[0015] Optionally, in the method according to the present invention, constructing a runoff prediction surface based on the current location of the current cleaning point, the regional curvature distribution of the corresponding pipe wall structure, and the gravity direction includes: Starting from the current position corresponding to the current cleaning point, a gravity guidance vector is generated along the gravity direction; Based on the regional curvature distribution of the corresponding pipe wall structure, the gravity guidance vector is fitted to the surface projection of the pipe wall structure region to obtain the central runoff curve that fits the pipe wall structure. The waste liquid overflow range is determined by the water jet flow rate parameters of the corresponding cleaning actuator, and a flow prediction surface corresponding to the waste liquid overflow range is formed along the pipe wall structure with the central flow curve as the reference axis.

[0016] Optionally, in the method according to the present invention, determining the runoff diversion vector based on the interference distance and the future location of the corresponding future cleaning point includes: Based on the future location of the corresponding future cleaning point, the interference direction of the central runoff curve corresponding to the future cleaning point is determined, and the spatial direction that is opposite to the interference direction is determined as the guiding direction. Determine the longitudinal span distance between the future cleaning point and the current cleaning point in the direction of gravity, and calculate the lateral avoidance margin based on the difference between the interference distance corresponding to the preset safety threshold. Determine the tangent characterization angle of the longitudinal span distance corresponding to the lateral clearance margin, and determine the runoff guiding vector corresponding to the tangent characterization angle along the guiding direction.

[0017] According to another aspect of the present invention, an image acquisition device for a gas pipeline is provided, comprising: The attachment identification module is configured to acquire a first three-dimensional image of the target pipeline area corresponding to the gas pipeline, and to identify the target attachment existing in the target pipeline area by identifying the first three-dimensional image. The cleaning determination module is configured to, when it is determined that the target attachment has cleaning attributes based on the attachment type, acquire the edge features of the corresponding target attachment based on the first three-dimensional image, and determine the peeling and cleaning points of the corresponding target attachment based on the edge features; The cleaning execution module is configured to control the cleaning execution mechanism to perform cleaning operations on the target attachments based on the stripping cleaning points, and respond to complete the cleaning operation to obtain a second three-dimensional image of the corresponding target pipeline area; The scraping execution module is configured to determine the residual attachments of the corresponding target attachments based on the second three-dimensional image, and control the scraping execution mechanism to perform scraping operations on the residual attachments.

[0018] According to the solution of the present invention, the present invention has the following beneficial effects: First, by acquiring a three-dimensional image of the target pipeline area, this invention makes up for the lack of depth information in the two-dimensional images of traditional manual endoscopes. At the same time, by combining the pipeline as-built drawings and dividing the region curvature and adjusting the pixels according to the curvature of the pipe wall, it can better distinguish the original construction structure such as flanges and supports from external impurities, eliminate the light and shadow interference in narrow and dim environments, and improve the accuracy of target attachment identification and positioning. Secondly, by analyzing the undulating features of the outer contour of the attachment, this invention can find the relatively weak points where the attachment is bonded to the pipe wall and use them as the peeling and cleaning points. In addition, controlling the cleaning mechanism to spray concentratedly along the point and at the set angle helps to peel off stubborn dirt. This not only makes reasonable use of the water flow impact force but also improves the actual cleaning effect. Next, after the fluid cleaning is completed, the present invention will scan again to obtain a second three-dimensional image for review, so as to detect the stubborn residues that remain after flushing, and control the scraping mechanism to perform targeted treatment. Using local mechanical scraping as a supplement to fluid flushing helps to restore the smoothness of the inner wall of the gas pipeline and avoids physical wear caused to the pipeline body by large-scale blind scraping. Finally, regarding the high-pressure wastewater generated during cleaning, this invention can combine parameters such as gravity direction, pipe wall curvature, and water jet volume to pre-determine the flow surface of the wastewater in the model. By calculating the interference distance between the subsequent cleaning points and the flow surface, the nozzle incident angle can be finely adjusted in real time to guide the wastewater to flow in a directional manner. This reduces the risk of sewage overflow covering and cross-contamination of uncleaned pipe sections, and facilitates the smooth progress of overall automated inspection and operation. Attached Figure Description

[0019] Figure 1 A flowchart of an image acquisition method for gas pipelines according to an embodiment of the present invention is shown; Figure 2 A schematic diagram of the construction of the corresponding shape characterization lines in this embodiment is shown; Figure 3 A structural block diagram of an image acquisition device for gas pipelines according to another embodiment of the present invention is shown. Detailed Implementation

[0020] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0021] To address the problems existing in the prior art, the inventors proposed the solution of this invention. One embodiment of this invention provides an image acquisition method for gas pipelines, which can be executed in a computing device, wherein the computing device can be understood as a terminal with data processing capabilities, such as a computer.

[0022] Figure 1 A flowchart of an image acquisition method for gas pipelines according to an embodiment of the present invention is shown, such as... Figure 1 As shown, the method begins with step S1, which includes the following: A first three-dimensional image of the target pipeline area corresponding to the gas pipeline is acquired, and the first three-dimensional image is identified to determine the target attachments existing in the target pipeline area.

[0023] For example, in this embodiment, in order to gain a clearer and more intuitive understanding of the actual conditions inside the gas pipeline, this embodiment can use a server to control a pipeline inspection device traveling inside the gas pipeline to perform preliminary image acquisition tasks. It can be noted that the pipeline inspection device is typically equipped with acquisition sensors such as a 3D laser scanner or a high-precision depth camera. When the inspection device slowly moves along the inner wall of the pipeline and reaches a pre-set target pipeline area, the pipeline inspection device can continuously emit scanning beams towards the pipe wall and receive reflected signals to capture the geometric undulations and surface 3D point cloud data of that section of the pipeline inner wall in as much detail as possible. This data is then used to generate and send a first 3D image of the target pipeline area to the server in real time.

[0024] After acquiring the first 3D image containing pipe wall depth information, the server invokes a preset 3D image analysis and / or point cloud recognition algorithm to process the image. Specifically, the server analyzes the spatial topological features of the pipe wall surface in the 3D image. Under normal and clean operating conditions, the 3D image features of the gas pipeline's inner wall should present a relatively smooth and continuous curved surface structure. During the recognition and calculation process, if the server detects a 3D point cloud cluster that clearly protrudes towards the center of the pipe, or an irregular elevation change area in the image, and these abrupt changes exceed the normal manufacturing and deformation tolerances of the pipe material, the server can determine that foreign substances have indeed been adsorbed on that local pipe wall.

[0025] Next, the server will clearly mark and identify this abnormal three-dimensional structure tightly attached to the pipe wall as a target deposit in the target pipeline area. For example, these target deposits identified by the server may be tar condensation hardened lumps formed by the long-term deposition of trace heavy oil impurities in the gas, or rust protrusions caused by localized erosion of the pipe wall by condensed water vapor. This embodiment, based on the identification method of three-dimensional depth information, can eliminate as much as possible the conventional visual interference caused by the dim environment inside the gas pipeline and the complex gas, thereby pinpointing the spatial location and three-dimensional shape of the foreign object as accurately as possible.

[0026] Furthermore, in this embodiment, the aforementioned "identifying the first three-dimensional image and determining the target attachment existing in the target pipeline area" may further include the following steps: Determine the inherent construction structures and their corresponding locations within the target pipeline area based on the as-built drawings; When the target pipeline area is determined to have at least a portion of inherent construction structure based on the structural location, the construction structure area indicating the at least portion is determined based on the first three-dimensional image, and the remaining portion is determined as the pipe wall structure area indicating the pipe wall structure. The elevation values ​​of each point corresponding to each image pixel in the construction structure area are compared with the three-dimensional point cloud data of the corresponding inherent construction structure obtained based on the as-built drawings to obtain the elevation deviation values. If any elevation deviation value exceeds a preset tolerance threshold, the image pixel corresponding to that elevation deviation value will be identified as an abnormal pixel. Based on the as-built drawings, the reference pixel value of the corresponding pipe wall structure is determined, and if the pixel value of any image pixel in the area that makes up the pipe wall structure is different from the reference pixel value, the image pixel is identified as an abnormal pixel. Based on the first 3D image, all adjacent abnormal pixels are connected to determine the target attachments existing in the target pipeline region based on the obtained abnormal sub-regions.

[0027] For example, in this embodiment, when the server performs recognition processing on the first three-dimensional image to determine whether there is a target attachment, it first calls and constructs a spatial topological benchmark of the target pipeline area in the server based on the as-built drawings of the gas pipeline in the early stage of construction. Specifically, the server will accurately parse the inherent construction structures that should exist in the current detection area based on the as-built drawings, such as flanges at pipeline connections, junctions of branch pipelines, or support brackets for reinforcement, and clarify the theoretical spatial coordinates of these structures. Next, the server compares the object outlines displayed in the first 3D image with the inherent structures in the as-built drawings. When the server determines that these inherent construction structures do exist in the current detection area based on the spatial relationship of the structures, the server uses, for example, image segmentation algorithms to accurately delineate the areas indicating these inherent structures in the first 3D image and identify them as construction structure areas. The remaining parts in the first 3D image, which should be smooth inner wall surfaces, are clearly defined as pipe wall structure areas indicating pipe wall structures. Subsequently, in order to quantitatively identify the attachments, the server can compare the actual 3D point cloud elevation value corresponding to each image pixel in the construction structure area with the standard 3D point cloud data of the corresponding inherent construction structure based on the digital reconstruction of the as-built drawings, thereby calculating the elevation deviation value at each sampling point. If the server responds during the comparison process and finds that the elevation deviation value corresponding to any image pixel significantly exceeds the preset tolerance threshold allowed for the manufacturing and installation of gas pipelines, it means that there may be non-pre-set material accumulation at this point. Here, the preset tolerance threshold can be set based on actual needs, and this embodiment does not limit it. At this point, the server can mark the image pixels corresponding to the elevation deviation value as part of the target attachment. It can be explained that by combining the design base map benchmark with the identification logic of real-time three-dimensional deviation analysis, it is possible to distinguish the normal structural features of the inner wall of the pipe from externally illegally intruded impurities as much as possible, which greatly improves the accuracy of target attachment identification and provides reliable spatial positioning parameters for subsequent cleaning operations.

[0028] Furthermore, in this embodiment, the aforementioned step of "responding to a pixel value difference between any image pixel in the region constituting the pipe wall structure and a reference pixel value, and determining the image pixel as an abnormal pixel" may further include the following steps: Based on the as-built drawings, the curvature distribution of the corresponding pipe wall structure is determined, and the pipe wall structure is divided based on the curvature distribution to obtain sub-regions corresponding to different curvature intervals. The pixel compensation coefficients corresponding to each curvature interval are retrieved and the base pixel values ​​are adjusted to obtain the dynamic pixel ranges corresponding to different sub-regions. If the pixel value of an image pixel located in any sub-region falls within the dynamic pixel range corresponding to that sub-region, the image pixel is identified as a pipe wall pixel; otherwise, it is identified as an abnormal pixel.

[0029] For example, in this embodiment, since gas pipelines often have bends or undulations in their actual three-dimensional spatial routing, the curvature of the pipe wall itself will cause natural changes in the brightness or darkness of the light reflection signal or the pixel features of the generated image when the detection device is scanning. Therefore, in order to avoid misjudging such visual changes caused by normal bending as attachments as much as possible, the server will consult and clarify the curvature distribution of the originally designed area of ​​the pipeline based on the as-built drawings in advance. For example, the server can know that a certain section of the pipe wall is straight, while another section of the pipe wall next to it has a 30-degree bend. Next, the server will divide the pipe wall structure into grid-like blocks based on these curvature characteristics. For example, the server can divide the smooth pipe wall with a curvature between 0 and 5° into an independent sub-region, divide the slightly curved pipe wall with a curvature between 5° and 15° into another sub-region, and so on, so as to obtain the sub-regions corresponding to different curvature ranges. Subsequently, the server can retrieve the pixel compensation coefficient corresponding to the curvature range of each of the previously divided sub-regions. For example, for a corner sub-region with a large curvature, the reflection signal is attenuated due to the skewed angle of light illumination, and its corresponding pixel compensation coefficient may be set to 1.2; while for a straight sub-region, the pixel compensation coefficient may be maintained at the baseline of 1. The pixel compensation coefficient can be set based on actual needs. In this embodiment, no line segment is used for this. Next, the server uses these specific pixel compensation coefficients to adjust the baseline pixel value that originally represented the pipe wall without dirt. Assuming that the baseline pixel value of the original flat inner wall is 120, after being amplified and adjusted by a compensation coefficient of 1.2, plus the allowable fluctuation space of environmental noise, the server can generate a dynamic pixel range of, for example, 130 to 150 for that bent sub-region, so as to tailor the most suitable dynamic pixel range for each pipe wall region with a specific curvature. After setting this localized baseline range, the server can check the actual captured image data point by point. When the server checks an image pixel located in a specific sub-region, it will extract the actual pixel value of that point and compare whether the value falls exactly within the dynamic pixel range. If the value is within this range, it means that the surface features at this location may conform to the normal physical behavior of the pipe wall under this curvature, and the server will determine and identify the image pixel as a normal pipe wall pixel. Conversely, if the pixel value exceeds the reasonable dynamic pixel range, such as being abnormally dark or exhibiting unnaturally high reflectivity, it indicates that the location is likely to be covered with oil or has undergone rust and discoloration. In this case, the server will decisively mark the image pixel and identify it as an abnormal pixel, thereby filtering out the interference of light and shadow errors caused by the natural bending of the pipe structure as much as possible, and greatly ensuring the rigor and accuracy of identifying minute abnormal points.

[0030] Step S2 includes the following: When the target attachment is determined to have cleaning properties based on the attachment type, the edge features of the corresponding target attachment are obtained based on the first three-dimensional image, and the peeling and cleaning points of the corresponding target attachment are determined according to the edge features.

[0031] For example, in this embodiment, after identifying the target deposit, the server will further comprehensively evaluate the relevant features of the deposit in the first three-dimensional image to determine its specific deposit type. For instance, if the evaluation result indicates that the deposit is loose sludge formed by a mixture of condensate and pipe dust, or soft tar accumulated over many years, the server will determine that such substances do not require strong scraping operations, but rather have cleaning properties that can be removed by high-pressure fluid flushing.

[0032] Here, once it is determined that the target attachment has cleaning properties, the server will perform edge detection on the target attachment based on the first 3D image to obtain the edge features of the corresponding target attachment. For example, the server can calculate and determine the surface bulge of the target attachment at a certain edge corner, showing an obvious edge protrusion or an edge feature that is not tightly attached to the pipe wall.

[0033] Next, the server uses the extracted edge features to find the entry point where the cleaning fluid can most easily penetrate and exert the greatest force to overturn and destroy the overall dirt, and then identifies this point as the peeling and cleaning point for the corresponding target deposits. It can be explained that this design of finding the optimal peeling angle based on three-dimensional morphological edge features provides accurate targeting guidance for subsequent fluid injection actions, greatly optimizing the consistency of the mechanical cleaning and scraping execution logic and the actual decontamination effect inside the gas pipeline.

[0034] Furthermore, in this embodiment, the aforementioned "when it is determined that the target attachment has cleaning properties based on the attachment type, the edge features of the corresponding target attachment are obtained based on the first three-dimensional image, and the peeling and cleaning points of the corresponding target attachment are determined according to the edge features" may further include the following steps: The region outline of the corresponding abnormal sub-region is divided to obtain the array points; Starting from the center point of the corresponding abnormal sub-region, generate shape representation lines connecting each array point, and determine the shape variance value of the corresponding abnormal sub-region based on the representation length obtained from each shape representation line. Based on all image pixels that make up the abnormal sub-region, determine the maximum and minimum pixel values ​​of the corresponding pixels, and determine the pixel distribution interval formed by the maximum and minimum pixel values; Based on the shape variance value and pixel distribution range of the corresponding abnormal sub-region, the preset type table is traversed to determine the attachment type of the corresponding target attachment. When the target attachment is determined to have cleaning properties based on the attachment type, the image pixels that make up the region contour are determined as contour pixels, and each shape representation line is controlled to extend to the image pixels located outside the abnormal sub-region and adjacent to the contour pixels to obtain each edge extension line. The edge features of the corresponding target attachments are determined based on the elevation change trend obtained from each edge extension line, and the peeling and cleaning points of the corresponding target attachments are determined based on the edge features.

[0035] For example, in this embodiment, after the server initially determines the abnormal sub-region representing the attachment, in order to determine its material type as accurately as possible and find a breakthrough in cleaning, it will first conduct an in-depth analysis of the external shape and internal pixel characteristics of the region. Specifically, the server can first discretize the outer closed edge of the corresponding abnormal sub-region, that is, the region outline, at equal intervals. For example, it can set to extract a node every two millimeters, so as to obtain a series of array points evenly on the entire outline perimeter. Next, the server will use a geometric algorithm to calculate the geometric centroid of the corresponding abnormal sub-region and use it as the region center point. Furthermore, with this region center point as the starting point for divergence, the server will radiate outwards to generate shape representation lines connecting each array point that was just divided. Subsequently, the server can determine the actual spatial physical length of each shape representation line, that is, the length of each representation. It can be explained that, since the attachments inside the pipes often grow randomly and have irregular shapes, the obtained representation lengths may vary. Based on this, the server can use statistical algorithms to calculate the dispersion of these representation lengths, and then determine the shape variance value that can objectively reflect the irregularity of the corresponding abnormal sub-region. For example, if the calculated shape variance value is large, it means that the edge of the attachment is like a starfish with many tentacles and an uneven shape; conversely, if the shape variance value is small, it means that its overall shape is closer to a regular circle with smooth edges. After quantizing the external shape, the server can traverse all image pixels that make up the abnormal sub-region and determine the maximum and minimum pixel values ​​of the corresponding pixels. For example, the maximum pixel value is 240 and the minimum pixel value is 110. Then, the server will use these two extreme values ​​as the upper and lower limits of the interval to determine the pixel distribution interval formed by the maximum and minimum pixel values. It can be said that the width of this interval can intuitively reflect the uniformity of the surface of the attached object. Next, the server will retrieve a pre-configured preset type table in the system. This table can record typical morphological parameters of various common pipe foulings such as tar, rust, and silt. The server will use the shape variance value and pixel distribution range obtained for the same abnormal sub-region as joint search conditions to traverse and compare this preset type table in order to find the standard record with the most matching parameter characteristics and determine the specific type of the target attachment as accurately as possible. When the server determines, based on the type of attachment determined above, that the target attachment is of a relatively loose texture and can be removed by fluid flushing, that is, when it is determined that it has cleaning properties, the server will clearly define each image pixel that makes up the outline of the outer region of the attachment as the outline pixel. Then, the server will control each shape representation line previously generated from the center point to continue to penetrate and extend outward along the original direction until these line segments cross the boundary of the attachment and extend to those image pixels that represent the normal pipe wall that are located outside the abnormal sub-region and closely adjacent to the outline pixel. Through the cross-boundary extension operation, the corresponding edge extension lines can be obtained. Finally, the server will extract and analyze the elevation undulation data of each edge extension line point by point to obtain the specific elevation change trend when crossing the edge of the attachment. For example, if it is found that the elevation data of a certain edge extension line drops sharply from a raised four millimeters to zero millimeters representing the clean pipe wall at the moment of crossing the boundary, presenting a steep micro-fault, the server will determine this specific three-dimensional shape drop as the edge feature of the corresponding target attachment.

[0036] It can be explained that, based on the acquired edge features, the server can identify as clearly as possible the location where the adhesion of the attachment edge is weakest and most easily overturned by high-pressure water flow, and then accurately determine the peeling and cleaning point of the corresponding target attachment based on the edge features.

[0037] Figure 2 The diagram illustrates the construction of the corresponding shape representation line in this embodiment. After identifying the abnormal sub-region formed by connecting adjacent abnormal pixels, the server extracts the irregular outer edge of this abnormal sub-region as the region contour. To quantify the geometry of this contour, the region contour is first discretized, as shown below. Figure 2 As shown, a series of array points, such as A, B, C, D, E, F, G, and H, are extracted along the closed contour edge. Next, the server calculates the center point of the corresponding abnormal sub-region using a geometric algorithm. Figure 2 The intersection point of all arrow line segments in the figure is used as the starting point for divergence, and radial lines are generated to each array point on the contour, thereby obtaining the shape representation lines with arrows in the figure. Understandably, each shape representation line has a corresponding actual spatial span, i.e., representation length. Since the deposits inside gas pipelines usually grow naturally and randomly, their outer edges often exhibit irregular and uneven shapes. This results in natural differences in the representation lengths of the shape representation lines pointing to different array points in the diagram. The server obtains the representation length of each shape representation line and uses statistical algorithms to calculate the dispersion of these length data, thereby determining the shape variance value of the corresponding abnormal sub-region. Generally speaking, a larger calculated shape variance value indicates that the outer contour of the attachment has strong undulations and an irregular shape. For example... Figure 2 The middle finger indicates that some points have longer line segments and others have shorter ones; conversely, if the shape variance value is small, it indicates that the characteristic lengths in each direction are relatively consistent and the shape of the attached object is relatively rounded and smooth.

[0038] Furthermore, in this embodiment, the aforementioned "traversing the retrieved preset type table based on the shape variance value and pixel distribution interval of the corresponding abnormal sub-region to determine the attachment type of the corresponding target attachment" may also include the following steps: Retrieve a preset type table, wherein the preset type table includes standard mapping groups corresponding to different standard types, and each standard mapping group includes different standard variance values ​​and standard distribution intervals; If the response shape variance, pixel distribution interval and the standard variance of any standard type, the variance similarity between the standard distribution interval and the standard type are all greater than the preset similarity threshold, then the standard type is determined as the candidate type. There is only one candidate response type, and that candidate type is determined as the attachment type of the corresponding target attachment. If there are multiple candidate types and at least one candidate type has a cleaning attribute, then any candidate type with a cleaning attribute is determined as the attachment type of the corresponding target attachment.

[0039] For example, in this embodiment, after the server obtains the shape variance value reflecting the shape characteristics of the deposit and the pixel distribution range reflecting its surface reflection or material texture, it needs to use an experience database to determine its specific composition. Based on this, the server can retrieve a preset type table pre-configured inside the server. It can be explained that the preset type table can be understood as a digital feature map of the corresponding pipe dirt, which can record various standard mapping groups corresponding to different standard types such as tar blocks, scale, and loose silt. Here, for each type of dirt, the corresponding standard mapping group provides detailed reference values ​​under normal conditions, which includes different standard variance values ​​and standard distribution ranges. For example, for the standard type of tar blocks, the corresponding standard variance value may be set to 1.8, and the standard distribution range may be a numerical range of 80 to 130. Next, the server will take the shape variance value and pixel distribution range calculated based on the first 3D image and compare them one by one with the standard data recorded in the preset type table. It can be explained that, through the correlation calculation formula, the server can calculate the degree of fit between the actual value and the standard value, that is, to obtain the variance similarity and distribution similarity. Furthermore, the server has a preset similarity threshold for controlling the matching accuracy, for example, set to 85%. When the server responds and finds that the variance similarity and distribution similarity obtained when comparing the actual data with the standard data of a certain standard type are both greater than the preset similarity threshold of 85%, it indicates that the attachment is as close as possible to the standard type in terms of 3D shape and pixel performance. At this time, the server will include the standard type in the consideration scope and determine it as a candidate type. It should be noted that the preset similarity threshold can be set based on actual needs, and this embodiment does not limit it in this way. After the traversal is complete, if only one candidate type is obtained after filtering, such as only tar block meeting the similarity requirement of more than 85%, it indicates that the recognition result is relatively clear. The server then determines the unique candidate type as the attachment type of the corresponding target attachment. However, since impurities in gas pipelines are often mixed and sometimes exhibit complex characteristics, multiple candidate types are selected. For example, if both soft, loose silt and mixed rust with a certain degree of hardness are selected, and the server finds that at least one of these candidate types has cleaning properties, such as loose silt being able to be removed by fluid flushing, then in order to prioritize non-contact fluid cleaning methods and reduce the risk of scratches to the pipe wall caused by direct mechanical scraping, the server will prioritize and determine any candidate type with cleaning properties as the type of attachment for the corresponding target deposit.

[0040] Furthermore, in this embodiment, the aforementioned "determining the edge features of the corresponding target attachment based on the elevation change trend obtained from each edge extension line, and determining the peeling and cleaning points of the corresponding target attachment according to the edge features" may further include the following steps: The difference in elevation values ​​of adjacent image pixels arranged along the same edge extension line is calculated to obtain an elevation sequence composed of the elevation extension differences. The elevation extension difference at the end of the high program column is determined as the edge extension difference, and at least two edge extension differences are less than the preset fitting threshold. Based on the edge extension line corresponding to the edge extension difference, the abnormal sub-region is divided to obtain each cleaning and stripping partition. The fitting evaluation value for each edge extension line corresponding to the same cleaning and stripping zone is determined based on the edge extension difference, and the slope evaluation value is determined based on the height difference between the maximum and minimum elevation values ​​corresponding to the edge extension line. Based on the high program sequence, all image pixels whose elevation extension difference shows a continuous increase are identified as increased pixels, and the number of interval pixels corresponding to all image pixels located between adjacent increased pixels are determined, so as to determine the change evaluation value based on the number of increased pixels and the number of interval pixels corresponding to all increased pixels. The fitting evaluation value, slope evaluation value and change evaluation value corresponding to the same edge extension line are weighted and summed to obtain the comprehensive evaluation value of each contour pixel. The contour pixel with the largest comprehensive evaluation value for each cleaning and stripping partition is determined as the stripping and cleaning point.

[0041] For example, in this embodiment, after the server obtains each edge extension line, it will delve into the microscopic undulations of the attachment edge along these line segments in order to find a breakthrough in cleaning. Specifically, the server can extract the elevation value of each adjacent image pixel along the same edge extension line. Then, the server will calculate the difference between the elevation values ​​of these adjacent points one by one. For example, if the elevation of the previous pixel is 5mm and the elevation of the next pixel is 3mm, then the calculated difference is 2mm. By arranging all the calculated differences in order of position, a continuous sequence of elevation values ​​reflecting the local edge undulations can be obtained. Subsequently, the server can define the elevation difference at the junction of the outermost end of the extension line across the boundary of the attachment and the pipe wall as the edge extension difference. In order to judge the tightness of the adhesion between the edge of the attachment and the pipe wall, the server can preset a bonding threshold, such as 0.5mm. When the server is traversing and calculating, if it responds and finds that there are at least two different directions where the edge extension difference is less than the preset bonding threshold of 0.5mm, it means that in these specific edge directions, the attachment has smoothly transitioned to the pipe wall and the two are tightly attached, making it difficult to be lifted by the water flow. At this time, the server will directly use these tightly attached edge extension lines as virtual dividing boundaries to divide the entire abnormal sub-region, thereby successfully obtaining each independent cleaning and stripping zone. It can be noted that the preset bonding threshold can be set based on actual needs, and this embodiment does not limit it. After dividing these independent zones, the server can quantitatively assess the cleaning difficulty of each edge extension line within the same cleaning and stripping zone. On the one hand, the server can calculate and determine a bonding assessment value representing the edge bonding strength based on the edge extension difference extracted earlier. On the other hand, the server can find the highest and lowest elevation values ​​along the entire edge extension line and calculate the height difference between the two. The larger the height difference, the steeper the terrain and the larger the water-facing section when impacted by water flow. Therefore, the server can convert and determine a slope assessment value that reflects the local morphology based on the height difference. Meanwhile, in order to further determine the material porosity or surface roughness at this location, the server can search for image pixels whose elevation extension difference shows a continuous increasing trend based on the high program sequence, and further unify them as increasing pixels. Then, the server will calculate the number of interval pixels between two adjacent increasing pixels. Subsequently, the server can summarize the total number of increasing pixels and the number of interval pixels determined above, and perform comprehensive calculations through a preset roughness evaluation model to determine a change evaluation value that can characterize the degree of change of pits on the surface of the attachment. It can be noted that the establishment and training process of the roughness evaluation model can be implemented based on existing technology, and this embodiment does not limit it. Finally, the server can deeply integrate the fit evaluation value, slope evaluation value, and variation evaluation value obtained for the same edge extension line. For example, the fit evaluation value is given a weight ratio of 30%, the slope evaluation value a weight ratio of 40%, and the variation evaluation value a weight ratio of 30%. Then, these weights are used to perform a weighted summation calculation to obtain a comprehensive evaluation value. The higher the comprehensive evaluation value, the steeper and rougher the edge is, and the less tightly it fits the pipe wall. Naturally, it is the weakest link that is most easily overturned by high-pressure water flow. Based on this, the server can perform numerical comparison in each cleaning and stripping section and determine the contour pixel point with the largest corresponding comprehensive evaluation value as the stripping and cleaning point of that section. This allows the server to find the optimal fluid impact coordinate point for complex pipe fouling as scientifically as possible, which greatly improves the operating efficiency of the subsequent cleaning actuator.

[0042] Step S3 includes the following: The control cleaning actuator performs cleaning operations on the target deposits based on the stripping and cleaning points, and responds to complete the cleaning operation to obtain a second three-dimensional image of the corresponding target pipeline area.

[0043] For example, in this embodiment, after the server locates the peeling and cleaning point of the target deposit, it issues a cleaning execution command to the pipeline inspection equipment deployed inside the gas pipeline. Upon receiving the command, the cleaning execution mechanism on the pipeline inspection equipment enters a working preparation state. It can be noted that the cleaning execution mechanism typically features a high-pressure micro-nozzle with high-pressure fluid directional injection capability and a matching multi-axis mechanical rocker arm assembly. To efficiently break the stubborn bond between the target deposit and the pipe wall, the cleaning execution mechanism can adjust the spatial orientation of the mechanical rocker arm, aiming the injection nozzle at the peeling and cleaning point where the edge is raised or where it is most vulnerable.

[0044] Next, the cleaning actuator will spray high-frequency pulsed water or gas-liquid mixture with the set impact parameters, such as a working pressure of 5MPa to 8MPa, to perform a fixed and continuous flushing and wedging operation on the peeling and cleaning point, so as to concentrate the kinetic energy of the water flow as much as possible at the weakest adhesion breakthrough point and peel the dirt off the pipe wall surface.

[0045] After the cleaning actuator continuously sprays fluid for a specific duration according to the preset operating procedure, such as continuous high-pressure rinsing for 45 seconds or one minute, the server determines that the current single cleaning process has been completed, that is, the cleaning operation has been completed. At this time, in order to objectively and accurately verify the actual decontamination effect of the cleaning action, the server sends a control signal to the pipeline detection device, instructing the cleaning actuator to stop fluid spraying, and then uses the pipeline detection device to collect data on the target pipeline area again, obtaining a second-dimensional image of the corresponding target pipeline area. It can be noted that the second-dimensional image can reflect the cleanliness of the pipe wall as realistically as possible after high-pressure fluid flushing, thus providing an intuitive data basis for the server to subsequently judge whether there are still stubborn stains and whether a mechanical scraping process needs to be initiated.

[0046] Furthermore, in this embodiment, the aforementioned "controlling the cleaning actuator to perform cleaning operations on the target attachments based on the stripping and cleaning points" may further include the following steps: Based on the 3D point cloud data of the corresponding pipe wall structure obtained from the as-built drawings, the center of the pipe wall section of the corresponding target pipe area is determined, and the reference axis of the corresponding target pipe area is constructed based on the gravity direction and the center of the pipe wall section. Determine the spatial line connecting each stripping and cleaning point to the center of the pipe wall cross-section, and determine the circumferential deflection angle of each spatial line corresponding to the reference axis. All stripping and cleaning points are arranged in ascending order according to the circumferential deflection angle from smallest to largest. Based on the obtained stripping execution sequence, the cleaning execution mechanism is controlled to perform cleaning operations on each stripping and cleaning point in sequence with a preset rinsing force.

[0047] For example, in this embodiment, in order to plan the cleaning operation trajectory as efficiently and coherently as possible for the cleaning actuator, the server first constructs a unified orientation coordinate reference in three-dimensional space. Specifically, the server determines the three-dimensional point cloud data of the original design shape of the pipeline based on the aforementioned as-built drawings. By analyzing this part of the standard pipe wall point cloud data, the server determines the geometric center position of the current target pipeline area on the cross section, that is, obtains the center of the pipe wall cross section. Subsequently, the server can introduce the vertically downward gravity direction as a natural reference, and combine the gravity direction with the just calculated center of the pipe wall cross section to construct a reference axis that runs through the target pipeline area. It can be explained that the reference axis can be understood as a vertical scale line on a clock face, thereby providing a stable corresponding reference for the subsequent determination of the orientation of each dirt point. Next, the server can generate virtual spatial lines in the virtual 3D model, starting from the center of the pipe wall cross-section and pointing to each specific stripping and cleaning point. Then, the server will measure and determine the angle between each virtual spatial line and the previously established reference axis, thereby obtaining the circumferential deflection angle corresponding to each point. For example, if a cleaning point is located at the bottom of the pipe directly below it, its corresponding circumferential deflection angle may be 0°; while a point located on the right side of the pipe wall may have a deflection angle of 90° or 120°. Furthermore, after obtaining the angle data of all cleaning points, the server can arrange all stripping and cleaning points in ascending order according to the circumferential deflection angle from small to large, that is, along the direction of the pipe wall circumference. It can be explained that through this angular sorting operation, the server actually connects the originally scattered dirt penetration points into a stripping execution sequence that conforms to the rotation operation law. Finally, the server can issue control commands to control the cleaning actuators on the inspection robot to start the cleaning operation in the order of the stripping sequence. The cleaning nozzles will rinse with a preset rinsing force, such as a moderate water pressure of 6MPa, and perform rinsing and stripping operations on each stripping and cleaning point in the sequence one by one.

[0048] It can be explained that, in this embodiment, the planning strategy of sorting and washing based on the gravity reference axis and circumferential angle can guide the nozzles to rotate in an orderly manner in one direction. This not only minimizes the ineffective back-and-forth swinging of the robotic arm, but also allows the water flow generated during cleaning and the detached dirt to flow more smoothly along the pipe wall, which largely avoids the cross-splashing of sewage into the cleaned area and effectively improves the overall cleaning efficiency.

[0049] Furthermore, in this embodiment, the aforementioned "controlling the cleaning actuator to perform cleaning operations on each stripping and cleaning point sequentially with a preset rinsing force based on the obtained stripping execution sequence" may further include the following steps: Based on the stripping execution sequence, each stripping and cleaning point is determined as the current cleaning point, and other stripping and cleaning points located after and adjacent to the current cleaning point are determined as future cleaning points. The reference incident angle of the corresponding edge extension line is determined based on the point elevation value of the current cleaning point. Based on the current location of the cleaning point, the regional curvature distribution of the corresponding pipe wall structure, and the direction of gravity, a runoff prediction surface is constructed, and the interference distance of the runoff prediction surface corresponding to the future cleaning point is obtained. If the response interference distance is greater than the preset safety threshold, the control cleaning actuator will perform cleaning operation on the current stripping and cleaning point based on the reference incident angle; otherwise, the runoff diversion vector will be determined based on the interference distance and the future position of the corresponding future cleaning point. The reference incident angle is corrected based on the runoff guide vector, and the cleaning actuator is controlled to perform cleaning operation on the current stripping and cleaning point with a preset flushing force based on the obtained corrected incident angle.

[0050] For example, in this embodiment, after the server obtains the arranged stripping execution sequence, it can sequentially determine the first stripping and cleaning point or the one that is currently being processed as the current cleaning point, and define and determine the other unprocessed points that follow it as future cleaning points. Then, in order to allow the high-pressure water flow to cut into the bottom of the dirt as smoothly as possible, the server can extract the elevation value of the current cleaning point. For example, based on the elevation value, it is known that the dirt at this point is raised by 4mm. Then, based on the elevation value and the edge trend obtained from the previous analysis, the reference incident angle can be determined, for example, set to an oblique angle of 30° with the pipe wall surface. Subsequently, in order to minimize the risk of wastewater or detached dirt and debris from the flushing operation flowing downstream and causing secondary pollution or covering of unwashed areas, the server can simulate the wastewater flow in virtual space. That is, the server can combine the current location of the cleaning point in the pipe, the curvature distribution characteristics of the pipe wall in that local area, and the natural direction of gravity to deduce and construct a runoff prediction surface that simulates the range of wastewater flow. Then, based on the runoff prediction surface, the server can measure and obtain the shortest spatial straight-line distance between each future cleaning point and the boundary of this wastewater prediction surface, thereby obtaining the interference distance. It can be explained that the server can pre-set a preset safety threshold for assessing the risk of pollution spread, such as 5cm. When the server responds and finds that the calculated interference distance is greater than this preset safety threshold of 5cm, it means that the sewage flow generated by the current flushing will flow away beyond the safety distance and will not affect the subsequent work area. In this case of non-interference, the server can issue a command to control the cleaning execution mechanism to keep the high-pressure nozzles at the originally calculated reference incident angle and perform normal cleaning operations on the current stripping and cleaning point. Conversely, if the interference distance does not meet the safety requirements of the preset safety threshold, it means that sewage will likely flow and cover the future cleaning point. At this time, the server will calculate a spatial mechanical vector based on the specific value of the interference distance and the spatial future location of the corresponding future cleaning point. This vector is designed to use the reaction force of the water flow to directionally divert the sewage. It can be noted that the preset safety threshold can be set based on actual needs, and this embodiment does not limit it. Next, the server can adjust the original spray angle based on the runoff diversion vector. For example, it can slightly deflect the original 30° reference incident angle by 5° on the horizontal plane to obtain a corrected incident angle that can guide the sewage to avoid the area to be cleaned. Finally, the server can control the robotic arm of the cleaning actuator to carry out cleaning operations on the current stripping and cleaning point according to the adjusted incident angle and the preset flushing force, thereby ensuring the clarity and cleanliness of the continuous cleaning operation surface as much as possible and effectively improving the smoothness of the overall pipeline descaling operation.

[0051] Furthermore, in this embodiment, the aforementioned "constructing a runoff prediction surface based on the current location of the current cleaning point, the regional curvature distribution of the corresponding pipe wall structure, and the gravity direction" may further include the following steps: Starting from the current position corresponding to the current cleaning point, a gravity guidance vector is generated along the gravity direction; Based on the regional curvature distribution of the corresponding pipe wall structure, the gravity guidance vector is fitted to the curved surface projection of the pipe wall structure region to obtain the central runoff curve that fits the pipe wall structure. The waste liquid overflow range is determined by the water jet flow rate parameters of the corresponding cleaning actuator, and a flow prediction surface corresponding to the waste liquid overflow range is formed along the pipe wall structure with the central flow curve as the reference axis.

[0052] For example, in this embodiment, in order to simulate the flow trajectory of high-pressure wastewater generated during cleaning as realistically as possible inside the pipe and to prevent the sewage from spreading randomly and affecting subsequent cleaning work, the server can perform preliminary calibration of the mechanical direction. Specifically, the server will use the specific three-dimensional spatial coordinates of the cleaning point that is currently being cleaned, i.e. the current point position, as the starting source point, and generate a gravity guidance vector in a virtual space, for example, in the direction of the Earth's gravity vertically downward, to indicate the natural downward trend of the liquid. For example, it can be understood that at the moment when the attached material is broken by the high-pressure water flow, the server can pre-calibrate the original downward direction of the sewage under force. Next, since the inner wall of the gas pipeline is not a flat plane, but a spatial curved surface with a specific curvature, sewage generally does not fall straight down after it is actually generated, but flows along the pipe wall. Therefore, the server can retrieve the regional curvature distribution data and gravity guidance vector of the corresponding pipe wall structure, and perform fitting calculations on the curved surface projection towards the pipe wall structure region to obtain a meandering central runoff curve attached to the inner wall of the pipe, representing the main convergence path of the water flow. Subsequently, the server also needs to further consider the actual water consumption during flushing, because the larger the flushing flow rate, the wider the area of ​​sewage spreading outward on the pipe wall. Based on this, the server can read and obtain the water jet flow rate parameters of the corresponding cleaning actuator. For example, the current set cleaning nozzle flow rate is 5L / min. Based on this specific water jet flow rate parameter, combined with the physical properties of liquid diffusion in the fluid dynamics model, the server can determine the waste liquid diffusion range that reflects the width of the sewage lateral flow. For example, it can be calculated that this sewage will spread 8cm to both sides on the pipe wall. Finally, the server uses the calculated central runoff curve as the reference center of the water flow, i.e., the reference axis, and spreads it out to the left and right along the surface of the pipe wall structure. This forms a runoff prediction surface on the pipe wall model that corresponds to the range of waste liquid overflow. This allows the server to predict the area affected by the flow of cleaning wastewater as accurately as possible, and provides a valuable spatial basis for subsequent calculations on whether the cleaning point will be contaminated.

[0053] Furthermore, in this embodiment, the aforementioned "determining the runoff diversion vector based on the interference distance and the future location of the corresponding future cleaning point" may also include the following steps: Based on the future location of the corresponding future cleaning point, the interference direction of the central runoff curve corresponding to the future cleaning point is determined, and the spatial direction that is opposite to the interference direction is determined as the guiding direction. Determine the longitudinal span distance between the future cleaning point and the current cleaning point in the direction of gravity, and calculate the lateral avoidance margin based on the difference between the interference distance corresponding to the preset safety threshold. Determine the tangent characterization angle of the longitudinal span distance corresponding to the lateral clearance margin, and determine the runoff guiding vector corresponding to the tangent characterization angle along the guiding direction.

[0054] For example, in this embodiment, when the server determines that sewage needs to be guided away from areas that have not yet been cleaned, it will first determine the specific interference direction of the sewage flow. That is, the server can analyze which side the future cleaning point is on relative to the previously deduced central runoff curve based on the future point position of the corresponding future cleaning point in three-dimensional space, thereby determining the specific interference direction. For example, if it is found that the future cleaning point is located at the lower left of the central runoff curve, then the trend of water flow spreading to the lower left is the corresponding interference direction. In order to divert the sewage away from this easily affected area as much as possible, the server will determine the direction that is geometrically opposite to the interference direction, that is, the position towards the upper right, as the guiding direction for sewage discharge. Next, in order to calculate the specific deflection range that the cleaning water flow needs to change, the server will determine the longitudinal span distance between the future cleaning point and the current cleaning point in the direction of gravity vertically downward. For example, the actual physical height difference between the two points is calculated to be 8cm. At the same time, the server will extract the previously set preset safety threshold, such as a safety distance of 5cm, and subtract the actual calculated interference distance from this preset safety threshold, such as subtracting 3cm, to obtain the lateral avoidance margin that represents the additional avoidance distance required. For example, the lateral avoidance margin calculated here is 2cm, which represents the physical distance that the water flow needs to be pushed outward in the lateral direction at least. Finally, the server transforms the abstract vector calculations into an intuitive spatial geometric diagram. That is, the server combines the longitudinal span distance, which represents the vertical drop, and the lateral clearance, which represents the lateral displacement distance, to construct a spatial right triangle. It then calculates the tangent angle corresponding to the ratio of these two side lengths, thus successfully determining a scientific tangent representation angle. For example, if the calculated deflection angle is 15°, the server will deflect the reference water jet angle by 15° along the previously determined upward and rightward flow direction. This ultimately determines the runoff guiding vector that can guide the wastewater to flow safely away, thereby controlling the direction of the wastewater generated by high-pressure washing as accurately and intuitively as possible. This largely avoids the cross-contamination problem caused by the wastewater spreading everywhere, making the scheduling of automated cleaning operations clearer and more reasonable.

[0055] Step S4 includes the following: Based on the second three-dimensional image, the residual attachments of the corresponding target attachments are determined, and the scraping actuator is controlled to scrape the residual attachments.

[0056] For example, in this embodiment, after the initial cleaning steps, to ensure the cleanliness of the gas pipeline's inner wall as much as possible, the server performs a secondary comparative analysis on the newly acquired second-dimensional image. Specifically, based on the point cloud data of the second-dimensional image, the server checks whether there are still stubborn clumps in the area where the target deposit is located that have not been completely flushed away by the fluid. For instance, if the point cloud comparison process finds that there are still irregularly patchy raised areas of point cloud on the pipe wall in that area, the server will clearly identify and determine these stubborn residues as the residues of the corresponding target deposit.

[0057] It can be explained that, when faced with stubborn residues that have withstood initial flushing, relying solely on fluid dynamics may often be ineffective. Therefore, the server can switch cleaning modes and send electrical control commands to activate the scraping actuator mounted on the pipeline inspection equipment. Here, the scraping actuator is typically a physical cutting module composed of a carbide scraper and a robotic arm capable of adjusting the output force. Upon receiving coordinate commands, the robotic arm adjusts the spatial orientation of the carbide scraper based on the three-dimensional position of the residual deposits sent by the server, ensuring that its cutting edge adheres to the pipe wall surface at a preset tilt angle.

[0058] For example, the server controls a scraper to cut into the bottom edge of the residual dirt at a 35° angle. Then, a robotic arm applies a constant wall-mounted thrust, such as 15N, to smoothly and reciprocally scrape the residual deposits along the pipe axis. This direct contact mechanical cutting and peeling action aims to remove as thoroughly as possible the deeply attached or partially calcified stubborn dirt.

[0059] It can be explained that through this mechanical scraping supplementary treatment, the smooth and flat state of the inner wall of the gas pipeline can be restored as much as possible, thereby largely eliminating the obstruction factors on the inner wall and providing a reliable channel environment guarantee for the efficient and safe transportation of gas.

[0060] Another embodiment of the present invention provides an image acquisition device for gas pipelines. Figure 2 Its corresponding device block diagram includes: The attachment identification module is configured to acquire a first three-dimensional image of the target pipeline area corresponding to the gas pipeline, and to identify the target attachment existing in the target pipeline area by identifying the first three-dimensional image. The cleaning determination module is configured to, when it is determined that the target attachment has cleaning attributes based on the attachment type, acquire the edge features of the corresponding target attachment based on the first three-dimensional image, and determine the peeling and cleaning points of the corresponding target attachment based on the edge features; The cleaning execution module is configured to control the cleaning execution mechanism to perform cleaning operations on the target attachments based on the stripping cleaning points, and respond to complete the cleaning operation to obtain a second three-dimensional image of the corresponding target pipeline area; The scraping execution module is configured to determine the residual attachments of the corresponding target attachments based on the second three-dimensional image, and control the scraping execution mechanism to perform scraping operations on the residual attachments.

[0061] Furthermore, to meet the 24 / 7 security monitoring needs along gas pipelines, improve the accuracy of intrusion event identification, and reduce false alarms and system errors caused by environmental interference, this embodiment, based on the aforementioned image acquisition hardware structure, also includes intelligent recognition and multimodal verification functions. The specific implementation process is as follows: 1. Improved image blur correction and recognition accuracy at night After the black light full-color camera module acquires the initial monitoring image, the image preprocessing link first completes nighttime blur correction and error elimination, specifically including: First, the degree of dirt and fog adhesion on the window surface is evaluated in real time based on the image grayscale variance and gradient distribution. When the evaluation value exceeds the preset cleanliness threshold, the dynamic adjustable keypad of the window self-cleaning structure is linked to adjust the opening. Combined with natural wind and rainwater, a directional scouring vortex is formed to reduce the image blur caused by window contamination from the hardware source. Secondly, to address the issues of lost details and noise in dark areas during low-light nighttime scenes, a multi-scale Retinex algorithm is used to separate the illumination and reflection components of the image. Adaptive gamma correction is then used to dynamically brighten dark areas while suppressing highlights, restoring image detail and color information. To address scattering blur caused by rain and fog, a dark channel prior algorithm is used for defogging to correct image clarity. Furthermore, for motion blur caused by dynamic targets such as mobile excavators and walking personnel, the target motion trajectory is estimated based on the optical flow field between adjacent frames, and a convolutional neural network is used to reconstruct the blurred edges at the pixel level to correct the deformation and trailing of the target contour, ensuring the accuracy of subsequent feature extraction. Finally, an environment-adaptive pixel error calibration model is established, pre-stores standard pixel reference values ​​under different ambient temperatures and light intensities, and calls the corresponding compensation coefficients based on real-time collected temperature and humidity data to dynamically compensate for color difference, dark noise and edge distortion of the image, thereby eliminating recognition reference errors caused by environmental fluctuations.

[0062] 2. Identification of intrusion targets along the pipeline route Based on image enhancement and error correction, intelligent identification and classification of targets in the monitoring screen are performed, with a focus on precise detection of mobile excavators and personnel carrying tools for damage. The specific process is as follows: A sample library of gas pipeline security scenarios is pre-built, covering targets such as mobile excavators, people holding tools, ordinary pedestrians, and civilian vehicles under different lighting, weather, and distance conditions, with target categories and tool subcategories labeled. Model training is completed based on a lightweight target detection network and deployed in the device's built-in edge computing unit. During real-time detection, candidate targets in the image are first extracted, and contour geometric features, motion trajectory features, and texture features are extracted for each candidate target. For personnel targets, fine-grained recognition is performed on the hand area to extract the contour features of destructive tools such as picks, shovels, and drills, distinguishing between ordinary pedestrians and destructive personnel holding tools. For mechanical targets, specific structural features such as excavator arms and tracks are extracted to identify mobile excavators and distinguish their working and driving states. Furthermore, combined with the depth detection data of the black light full-color camera module, the straight-line distance between the target and the gas pipeline body is calculated. When the target category is an excavator or a person holding tools, and the distance is less than a preset warning threshold, it is judged as a high-risk intrusion event, triggering an early warning and linking to subsequent multimodal verification processes.

[0063] 3. Elimination of audioprint-vibration dual-mode fusion error To further reduce the false alarm rate of single-vision recognition and eliminate recognition errors caused by factors such as light and shadow, and occlusion, this embodiment introduces a dual verification mechanism consisting of a voiceprint acquisition unit and a vibration sensing unit. The specific implementation steps are as follows: voiceprint acquisition units and fiber optic vibration sensors are deployed at intervals along the gas pipeline, and each sensing unit is networked synchronously with the image acquisition device; when the image recognition module outputs a high-risk warning, the acquisition of voiceprint and vibration signals in the corresponding area is triggered simultaneously; feature extraction is performed on the acquired voiceprint and vibration signals respectively: for the voiceprint signal, the spectral distribution is obtained through a fast Fourier transform, and features are extracted... The system extracts the dominant frequency, harmonic characteristics, and duration characteristics of vibration signals. For vibration signals, it extracts vibration frequency, amplitude, impact interval, and propagation attenuation characteristics. It then performs a weighted fusion calculation of the target category, confidence level, voiceprint feature matching degree, and vibration feature matching degree of image recognition to obtain a comprehensive verification confidence level. When the comprehensive verification confidence level is higher than a preset confirmation threshold, it is ultimately determined to be a real intrusion event, and an alarm message is output and uploaded to the monitoring center. When the comprehensive verification confidence level is lower than the preset confirmation threshold and only a single mode triggers the alarm, it is determined to be a false alarm caused by environmental interference and is filtered out, thereby eliminating errors and false alarms in single-mode detection.

[0064] For example, when an image identifies a suspected excavator target, if the acoustic signature matching the low-frequency spectrum of a diesel engine and the periodic impact vibration signature of the corresponding mechanical operation are simultaneously collected, the overall confidence level is met, and the excavator intrusion is confirmed. If only image recognition is triggered and there are no corresponding acoustic signatures or vibration signatures, it is determined to be a light and shadow misjudgment, and a filtering warning is issued.

[0065] 4. All-weather extreme weather adaptive operation Combining hardware structure and algorithm optimization, this device can achieve stable operation 24 / 7 in all weather conditions and under various extreme weather conditions. At the hardware level, a self-cleaning structure for the viewing window addresses window contamination caused by rain and dust, a flexible press-fit sealing structure offsets the stress of thermal expansion and contraction under high and low temperatures, preventing optical path distortion caused by optical glass deformation, and an internal humidity balancing structure prevents condensation in the internal cavity under low temperatures, thus ensuring optical imaging stability under extreme conditions from a physical perspective. At the algorithm level, an environmental adaptive scheduling module is set up to collect ambient light, temperature, humidity, and wind speed data in real time, automatically switching the parameter combination of the image enhancement algorithm and the weight branch of the target recognition model. For example, it automatically switches to low-light enhancement mode at night, automatically starts defogging correction mode in rainy and foggy weather, and automatically increases the pixel compensation coefficient in hot weather to ensure optimal recognition accuracy and imaging effect in different environments. With the all-weather monitoring capability of soundprint and vibration sensors, even in scenarios where visual imaging is severely limited by extreme weather, preliminary warnings can still be issued through soundprint and vibration modes, and a final verification can be completed after visual conditions recover, achieving full-time, all-condition safety coverage of gas pipelines.

[0066] In the specification provided herein, the algorithms and displays are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems can also be used with the examples of this invention. The required structure for constructing such systems is apparent from the above description. Furthermore, this invention is not directed to any particular programming language. It should be understood that the contents of the invention described herein can be implemented using various programming languages, and the above description of specific languages ​​is for the purpose of disclosing preferred embodiments of the invention.

[0067] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0068] Similarly, it should be understood that, in order to streamline this disclosure and aid in understanding one or more of the various aspects of the invention, in the above description of exemplary embodiments of the invention, various features of the invention are sometimes grouped together in a single embodiment, figure, or description thereof.

[0069] Those skilled in the art will understand that modules, units, or components of the devices disclosed in the examples herein can be arranged in the devices described in this embodiment, or alternatively, can be located in one or more devices different from the devices in this example. The modules in the foregoing examples can be combined into a single module or, in addition, can be divided into multiple sub-modules.

[0070] Those skilled in the art will understand that the modules in the device of the embodiment can be adaptively changed and placed in one or more devices different from that embodiment. Modules, units, or components in the embodiment can be combined into a single module, unit, or component, and further, they can be divided into multiple sub-modules, sub-units, or sub-components.

[0071] Furthermore, those skilled in the art will understand that although some embodiments described herein include certain features included in other embodiments but not others, combinations of features from different embodiments are meant to be within the scope of the invention and form different embodiments.

[0072] Furthermore, some of the embodiments described herein are methods or combinations of method elements that can be implemented by a processor of a computer system or by other means of performing the functions. Therefore, a processor having the necessary instructions for implementing the methods or method elements forms means for implementing the methods or method elements. Furthermore, the elements described herein in the apparatus embodiments are examples of means for implementing the functions performed by elements for the purposes of carrying out the invention.

[0073] As used herein, unless otherwise specified, the use of ordinal numbers such as “first,” “second,” “third,” etc., to describe ordinary objects merely indicates different instances of similar objects and is not intended to imply that the objects being described must have a given order in time, space, ordering, or any other manner.

[0074] Although the invention has been described with respect to a limited number of embodiments, those skilled in the art will understand from the foregoing description that other embodiments are conceivable within the scope of the invention described herein. Furthermore, it should be noted that the language used in this specification has been chosen primarily for readability and edibility purposes, and not for the purpose of explaining or limiting the subject matter of the invention.

Claims

1. An image acquisition method for gas pipelines, characterized in that, Includes the following steps: Acquire a first three-dimensional image of the target pipeline area located in the gas pipeline, and identify the target attachments existing in the target pipeline area by recognizing the first three-dimensional image. When the target attachment is determined to have cleaning properties based on the type of attachment, the edge features of the corresponding target attachment are obtained based on the first three-dimensional image, and the peeling and cleaning points of the corresponding target attachment are determined based on the edge features; The control cleaning actuator performs cleaning operations on the target deposits based on the stripping and cleaning points, and responds to complete the cleaning operation to obtain a second three-dimensional image of the corresponding target pipeline area; Based on the second three-dimensional image, the residual attachments of the corresponding target attachments are determined, and the scraping actuator is controlled to scrape the residual attachments.

2. The method according to claim 1, characterized in that, The first three-dimensional image is used to identify target attachments existing in the target pipe area, including: Determine the inherent construction structures and their corresponding locations within the target pipeline area based on the as-built drawings; When the target pipeline area is determined to have at least a portion of inherent construction structure based on the structural location, the construction structure area indicating the at least portion is determined based on the first three-dimensional image, and the remaining portion is determined as the pipe wall structure area indicating the pipe wall structure. The elevation values ​​of each point corresponding to each image pixel in the construction structure area are compared with the three-dimensional point cloud data of the corresponding inherent construction structure obtained based on the as-built drawings to obtain the elevation deviation values. If any elevation deviation value exceeds a preset tolerance threshold, the image pixel corresponding to that elevation deviation value will be identified as an abnormal pixel. Based on the as-built drawings, the reference pixel value of the corresponding pipe wall structure is determined, and if the pixel value of any image pixel in the area that makes up the pipe wall structure is different from the reference pixel value, the image pixel is identified as an abnormal pixel. Based on the first 3D image, all adjacent abnormal pixels are connected to determine the target attachments existing in the target pipe region based on the obtained abnormal sub-regions.

3. The method according to claim 2, characterized in that, If the pixel value of any image pixel in the region constituting the pipe wall structure differs from the reference pixel value, that image pixel is identified as an abnormal pixel, including: Based on the as-built drawings, the curvature distribution of the corresponding pipe wall structure is determined, and the pipe wall structure is divided based on the curvature distribution to obtain sub-regions corresponding to different curvature intervals. The pixel compensation coefficients corresponding to each curvature interval are retrieved and the base pixel values ​​are adjusted to obtain the dynamic pixel ranges corresponding to different sub-regions. If the pixel value of an image pixel located in any sub-region falls within the dynamic pixel range corresponding to that sub-region, the image pixel is identified as a pipe wall pixel; otherwise, it is identified as an abnormal pixel.

4. The method according to claim 2, characterized in that, When the target attachment is determined to have cleaning properties based on the attachment type, the edge features of the corresponding target attachment are obtained based on the first 3D image, and the peeling and cleaning points of the corresponding target attachment are determined based on the edge features, including: The region outline of the corresponding abnormal sub-region is divided to obtain the array points; Starting from the center point of the corresponding abnormal sub-region, generate shape representation lines connecting each array point, and determine the shape variance value of the corresponding abnormal sub-region based on the representation length obtained from each shape representation line. Based on all image pixels that make up the abnormal sub-region, determine the maximum and minimum pixel values ​​of the corresponding pixels, and determine the pixel distribution interval formed by the maximum and minimum pixel values; Based on the shape variance value and pixel distribution range of the corresponding abnormal sub-region, the preset type table is traversed to determine the attachment type of the corresponding target attachment. When the target attachment is determined to have cleaning properties based on the attachment type, the image pixels that make up the region contour are determined as contour pixels, and each shape representation line is controlled to extend to the image pixels located outside the abnormal sub-region and adjacent to the contour pixels to obtain each edge extension line. The edge features of the corresponding target attachments are determined based on the elevation change trend obtained from each edge extension line, and the peeling and cleaning points of the corresponding target attachments are determined based on the edge features.

5. The method according to claim 4, characterized in that, Based on the shape variance value and pixel distribution range of the corresponding abnormal sub-region, the retrieved preset type table is traversed to determine the attachment type of the corresponding target attachment, including: Retrieve a preset type table, wherein the preset type table includes standard mapping groups corresponding to different standard types, and each standard mapping group includes different standard variance values ​​and standard distribution intervals; If the response shape variance, pixel distribution interval and the standard variance of any standard type, the variance similarity between the standard distribution interval and the standard type are all greater than the preset similarity threshold, then the standard type is determined as the candidate type. There is only one candidate type for the response; this candidate type is determined as the attachment type of the corresponding target attachment. If there are multiple candidate types and at least one candidate type has a cleaning attribute, then any candidate type with a cleaning attribute is determined as the attachment type of the corresponding target attachment.

6. The method according to claim 4, characterized in that, Based on the elevation change trend obtained from each edge extension line, the edge features of the corresponding target attachment are determined, and the peeling and cleaning points of the corresponding target attachment are determined according to the edge features, including: The difference in elevation values ​​of adjacent image pixels arranged along the same edge extension line is calculated to obtain an elevation sequence composed of the elevation extension differences. The elevation extension difference at the end of the high program column is determined as the edge extension difference, and at least two edge extension differences are less than the preset fitting threshold. Based on the edge extension line corresponding to the edge extension difference, the abnormal sub-region is divided to obtain each cleaning and stripping partition. The fitting evaluation value for each edge extension line corresponding to the same cleaning and stripping zone is determined based on the edge extension difference, and the slope evaluation value is determined based on the height difference between the maximum and minimum elevation values ​​corresponding to the edge extension line. Based on the high program sequence, all image pixels whose elevation extension difference shows a continuous increase are identified as increased pixels, and the number of interval pixels corresponding to all image pixels located between adjacent increased pixels are determined, so as to determine the change evaluation value based on the number of increased pixels and the number of interval pixels corresponding to all increased pixels. The fitting evaluation value, slope evaluation value and change evaluation value corresponding to the same edge extension line are weighted and summed to obtain the comprehensive evaluation value of each contour pixel. The contour pixel with the largest comprehensive evaluation value for each cleaning and stripping partition is determined as the stripping and cleaning point.

7. The method according to claim 2, characterized in that, The cleaning actuator is controlled to perform cleaning operations on the target deposits based on the stripping and cleaning points, including: Based on the 3D point cloud data of the corresponding pipe wall structure obtained from the as-built drawings, the center of the pipe wall section of the corresponding target pipe area is determined, and the reference axis of the corresponding target pipe area is constructed based on the gravity direction and the center of the pipe wall section. Determine the spatial line connecting each stripping and cleaning point to the center of the pipe wall cross-section, and determine the circumferential deflection angle of each spatial line corresponding to the reference axis. All stripping and cleaning points are arranged in ascending order according to the circumferential deflection angle from smallest to largest. Based on the obtained stripping execution sequence, the cleaning execution mechanism is controlled to perform cleaning operations on each stripping and cleaning point in sequence with a preset rinsing force.

8. The method according to claim 7, characterized in that, Based on the obtained stripping execution sequence, the cleaning actuator is controlled to perform cleaning operations on each stripping and cleaning point sequentially with a preset rinsing force, including: Based on the stripping execution sequence, each stripping and cleaning point is determined as the current cleaning point, and other stripping and cleaning points located after and adjacent to the current cleaning point are determined as future cleaning points. The reference incident angle of the corresponding edge extension line is determined based on the point elevation value of the current cleaning point. Based on the current location of the cleaning point, the regional curvature distribution of the corresponding pipe wall structure, and the direction of gravity, a runoff prediction surface is constructed, and the interference distance of the runoff prediction surface corresponding to the future cleaning point is obtained. If the response interference distance is greater than the preset safety threshold, the control cleaning actuator will perform cleaning operation on the current stripping and cleaning point based on the reference incident angle; otherwise, the runoff diversion vector will be determined based on the interference distance and the future position of the corresponding future cleaning point. The reference incident angle is corrected based on the runoff guide vector, and the cleaning actuator is controlled to perform cleaning operation on the current stripping and cleaning point with a preset flushing force based on the obtained corrected incident angle.

9. The method according to claim 8, characterized in that, Based on the current location of the cleaning point, the regional curvature distribution of the corresponding pipe wall structure, and the direction of gravity, a runoff prediction surface is constructed, including: Starting from the current position corresponding to the current cleaning point, a gravity guidance vector is generated along the gravity direction; Based on the regional curvature distribution of the corresponding pipe wall structure, the gravity guidance vector is fitted to the curved surface projection of the pipe wall structure region to obtain the central runoff curve that fits the pipe wall structure. The waste liquid overflow range is determined by the water jet flow rate parameters of the corresponding cleaning actuator, and a flow prediction surface corresponding to the waste liquid overflow range is formed along the pipe wall structure with the central flow curve as the reference axis.

10. An image acquisition device for gas pipelines, characterized in that, include: The attachment identification module is configured to acquire a first three-dimensional image of the target pipeline area corresponding to the gas pipeline, and to identify the target attachment existing in the target pipeline area by identifying the first three-dimensional image. The cleaning determination module is configured to, when it is determined that the target attachment has cleaning attributes based on the attachment type, acquire the edge features of the corresponding target attachment based on the first three-dimensional image, and determine the peeling and cleaning points of the corresponding target attachment based on the edge features; The cleaning execution module is configured to control the cleaning execution mechanism to perform cleaning operations on the target attachments based on the stripping cleaning points, and respond to complete the cleaning operation to obtain a second three-dimensional image of the corresponding target pipeline area; The scraping execution module is configured to determine the residual attachments of the corresponding target attachments based on the second three-dimensional image, and control the scraping execution mechanism to perform scraping operations on the residual attachments.

Images