A boiler plate girder hoisting in-place deviation detection method based on image recognition

Abstract

The application discloses a kind of based on image recognition's boiler plate beam hoisting deviation detection method, it is related to hoisting precision image recognition technical field, comprising: collection hoisting real-time image, hoisting real-time image is benchmark mapping correction, generates hoisting benchmark image set;Plate beam contour gray gradient and sling shelter texture difference are extracted from hoisting benchmark image set, sling shelter area is separated with plate beam boundary area, and according to the continuous trend of plate beam boundary, broken contour is filled, and generates repair contour sequence;From real-time deviation data, deviation decreasing continuation section and deviation reverse mutation section are extracted, and deviation evolution state is determined according to deviation close to stable degree, and deviation evolution state is reviewed and adjusted and prompted in position, and generates hoisting deviation report.The application improves accurate positioning, dynamic monitoring and safety reliability in hoisting process.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology for hoisting accuracy, and in particular to a method for detecting positioning deviations in the hoisting of boiler slab beams based on image recognition. Background Technology

[0002] With the improvement of industrial automation, especially in hoisting operations, the application of image recognition methods has gradually become an important means to improve work accuracy and efficiency. The hoisting positioning deviation detection method based on image recognition uses high-precision images and advanced image processing algorithms to complete real-time monitoring and automated correction in hoisting operations. By acquiring hoisting images in real time and performing benchmark mapping correction, a hoisting benchmark image set can be generated, thereby ensuring the high accuracy and consistency of the detection results.

[0003] Traditional image processing methods typically rely on static image analysis, which cannot fully account for the dynamic changes in images during hoisting. In particular, when there is significant image occlusion or lighting changes, the stability and accuracy of the detection results are limited. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method for detecting the positioning deviation of boiler plate beams based on image recognition, which solves the problem of unstable detection under dynamic changes and occlusion in traditional image processing methods.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides a method for detecting positioning deviations in boiler slab beam hoisting based on image recognition. The method includes: acquiring real-time hoisting images; performing reference mapping correction on the real-time hoisting images to generate a hoisting reference image set; extracting the grayscale gradient of the slab beam contour and the difference in texture due to sling occlusion from the hoisting reference image set, separating the sling occlusion area from the slab beam boundary area, and completing the fracture contour according to the continuous direction of the slab beam boundary to generate a repair contour sequence; extracting the confidence level of the occlusion completion segment and the end-line closure engagement point from the repair contour sequence, and extrapolating the high-confidence engagement point back to the support bearing reference area to generate a slab beam support correspondence map; extracting the engagement point deviation and overlap increment from the slab beam support correspondence map, establishing a unidirectional approach chain based on the consistency of deviation decrease, suppressing the reverse abrupt change correspondence, and generating real-time positioning deviation data; extracting the deviation decrease continuation segment and the deviation reverse abrupt change segment from the real-time positioning deviation data, determining the deviation evolution state according to the degree of deviation approaching stability, performing positioning verification and adjustment prompts on the deviation evolution state, and generating a hoisting deviation report.

[0007] As a preferred embodiment of the image recognition-based boiler plate beam hoisting and positioning deviation detection method of the present invention, the method involves acquiring real-time hoisting images, identifying the image clarity, occlusion integrity and support area visibility of the real-time hoisting images, eliminating invalid hoisting images that cannot identify the plate beam outline and support bearing reference area, and generating a valid hoisting image sequence. Extract the edge of the bearing reference area and the fixed reference position of the hoisting area from the effective hoisting image sequence, match them with the preset reference coordinates, calculate the reference geometric deviation, and identify the image geometric deviation; A reference mapping relationship is established based on the geometric deviation of the images, and the position reference geometry is corrected for the effective hoisting image sequence according to the reference mapping relationship to generate a hoisting reference image set.

[0008] As a preferred embodiment of the image recognition-based boiler slab beam hoisting and positioning deviation detection method of the present invention, the specific steps for extracting the gray-level gradient of the slab beam contour and the difference in texture caused by sling occlusion from the hoisting reference image set are as follows: Local images of the hoisting are extracted from the hoisting reference image set, and grayscale smoothing and local enhancement are performed on the local images to generate enhanced analysis images; The gray-level gradient of the beam outline and the difference in texture of the suspension cable occlusion are extracted from the enhanced analysis image. The gray-level gradient of the beam outline is encoded in the direction sequence according to the boundary extension direction. The difference in texture of the suspension cable occlusion is encoded in the perturbation level according to the perturbation intensity of the neighborhood. The layers are then interlocked using the same pixel coordinate as the anchor point to generate a joint outline texture map.

[0009] As a preferred embodiment of the image recognition-based boiler plate girder hoisting and positioning deviation detection method of the present invention, the steps of separating the sling-obstructed area from the plate girder boundary area and completing the fracture contour according to the continuous direction of the plate girder boundary to generate a repair contour sequence are as follows: From the contour texture joint map, select the initial boundary chain with continuous gray-level transitions and consistent direction extension, and locate the occlusion and breakage positions in the initial boundary chain where gray-level transitions drop abruptly, texture perturbation is enhanced, and boundary direction is interrupted, and generate the breakage candidate chain. Extract the front and rear boundary support segments on both sides of the fracture location from the fracture candidate chain, classify the area with boundary stability characteristics into the plate beam boundary area, classify the area with sudden increase in texture disturbance and interruption of closure into the suspension cable occlusion area, and generate a contour separation mark map. The boundary area of ​​the plate beam in the contour separation mark map is retained, the area obscured by the suspension cable is removed, and the boundary support segments on both sides of the fracture location are connected to generate the contour chain. The completed contour chain is smoothed and closed, and the completed contour chain that matches the check is sequentially spliced ​​with the original plate beam boundary area to generate the repair contour sequence.

[0010] As a preferred embodiment of the image recognition-based boiler beam hoisting and positioning deviation detection method of the present invention, the specific steps for extracting the confidence level of the occlusion completion segment and the end line closure engagement point from the repair contour sequence are as follows: Extract the connection position between the occluded fill segment and the original plate beam boundary area from the repair contour sequence, and verify the connection direction, grayscale continuity and texture attenuation along the end of the plate beam to generate the end fill connection chain. Based on the end-filling support chain, the boundary continuity, residual shading disturbance, and bearing width matching status of the shading filler segment are evaluated in tandem to generate the confidence level of the shading filler segment. High-confidence occlusion completion segments are selected from the confidence scores of the occlusion completion segments, and the boundary support segments on both sides of the high-confidence occlusion completion segments are extended towards each other along the end closure direction to generate end line closure engagement points.

[0011] As a preferred embodiment of the image recognition-based boiler plate beam hoisting and positioning deviation detection method of the present invention, the plate beam support correspondence diagram is obtained by performing a back-hook connection of the end bearing edge line along the bearing boundary normal of the bearing reference area based on the end line closed engagement point, and marking the engagement deviation position and the bearing corresponding position.

[0012] As a preferred embodiment of the image recognition-based boiler plate girder hoisting and positioning deviation detection method of the present invention, the specific steps for extracting the interlocking point deviation and coincidence increment from the plate girder support correspondence diagram are as follows: Read the interlocking deviation position, the corresponding bearing position and the time sequence of the image frames from the corresponding diagram of the plate beam support, and connect the positional correspondence of the same end line closed interlocking point in adjacent image frames to generate an interlocking point frame alignment sequence. The normal deviation distance of the end line closure engagement point relative to the corresponding bearing position is extracted from the inter-frame alignment sequence of the engagement point as the engagement point deviation, and the overlap change between the end bearing edge line and the bearing reference area of ​​the support is extracted as the overlap increment to generate the deviation overlap linkage sequence.

[0013] As a preferred embodiment of the image recognition-based boiler beam hoisting and positioning deviation detection method of the present invention, the specific steps for establishing a unidirectional approach chain based on the consistency of deviation decrease, suppressing the corresponding reverse abrupt change, and generating real-time positioning deviation data are as follows: Based on the deviation coincidence linkage sequence, the near consistency state in adjacent image frames is verified in the same frame to generate a near locking sequence. According to the near locking sequence, the correspondence of continuously decreasing biting point deviation and continuous coincidence increment is linked together, and the correspondence of reverse increase biting point deviation is suppressed to generate a one-way nearing chain. Based on the unidirectional approach chain, the approach consistency state within the chain is preserved, and the deviation changes of the plate beam end relative to the support bearing reference area are sorted according to the time sequence of image frames to generate real-time positioning deviation data.

[0014] As a preferred embodiment of the image recognition-based boiler beam hoisting and positioning deviation detection method of the present invention, the specific steps of extracting the deviation decreasing continuation segment and the deviation reversal abrupt change segment from the real-time positioning deviation data, and determining the deviation evolution state according to the degree of deviation approaching stability, are as follows: The in-place deviation frame sequence is extracted from the real-time in-place deviation data, and the in-place deviation frame sequence is continuously merged to generate a deviation evolution chain. Based on the deviation evolution chain, the decreasing continuity of the remaining positioning deviation and the end swing situation are identified, and the frame segments that continuously decrease and do not swing at the end are marked as deviation decreasing continuation segments. Based on the deviation evolution chain, the reverse change of offset between adjacent frame segments is identified, and the frame segments in which the remaining positioning deviation suddenly increases, the offset direction is reversed, and the stability of the overlapping domain decreases are marked as deviation reverse mutation segments. Based on the deviation reverse mutation segment and the deviation decreasing continuation segment, the decreasing segment is strung together as a stable main chain, and the mutation segment is embedded as a perturbation side chain. The deviation evolution state is divided according to whether the side chain is absorbed by the subsequent decreasing segment.

[0015] As a preferred embodiment of the image recognition-based boiler beam hoisting and positioning deviation detection method of the present invention, the specific steps for performing positioning verification and adjustment prompts on the deviation evolution state and generating a hoisting deviation report are as follows: Extract the frame segment positions corresponding to the near-stable state, the deviation swing state, and the state to be reviewed from the deviation evolution state, and read back the overlap relationship between the end bearing edge line and the support bearing reference area for the frame segment position corresponding to the near-stable state. Extract the frame position corresponding to the deviation swing state from the overlap preservation relationship, track the change in offset direction and the decrease in the stability of the overlap domain in the deviation reverse mutation segment, and generate adjustment prompts. Extract the frame segment positions corresponding to the state to be reviewed from the adjustment prompts, mark the frame segments that do not meet the near-stable state and have not formed a continuous deviation swing state as continuous observation items, and generate a review prompt sequence; Based on the review prompt sequence, the review prompt information is arranged to generate a hoisting deviation report.

[0016] The beneficial effects of this invention are as follows: by separating the area obscured by the sling and the boundary of the plate beam through image recognition and completing the fracture contour, the integrity of the plate beam boundary information and the generation of a high-precision repair contour sequence are completed. At the same time, by tracking the position offset features and the overlap and change of the bearing reference area of ​​the support along the adjacent image frames, the plate beam offset is captured in real time and the decreasing trend and the reverse sudden change are distinguished, which improves the accuracy of positioning, dynamic monitoring and safety and reliability of the hoisting process. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is an overall flowchart of the image recognition-based boiler beam hoisting and positioning deviation detection method.

[0019] Figure 2 Flowchart for hoisting image acquisition, benchmark correction, and contour restoration.

[0020] Figure 3 A flowchart illustrating the mapping process for locating the end bearing edge and aligning it with the plate beam support.

[0021] Figure 4 A flowchart for real-time positioning deviation evolution analysis and hoisting deviation report generation. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4As one embodiment of the present invention, this embodiment provides a method for detecting the positioning deviation of boiler plate beams based on image recognition, including the following steps: S1. Acquire real-time images of the hoisting operation, perform benchmark mapping correction on the real-time images of the hoisting operation, and generate a set of hoisting benchmark images.

[0026] S1.1 Acquire real-time hoisting images, identify the image clarity, occlusion integrity, and support area visibility of the real-time hoisting images, remove invalid hoisting images that cannot identify the outline of the beam and the support bearing reference area, and generate a valid hoisting image sequence.

[0027] It should be noted that real-time images of the hoisting are acquired, and edge gradient distribution, local contrast changes, and high-frequency detail strength are extracted to calculate image sharpness. Then, the area covered by the slings and the continuous area of ​​the beam outline are located, and the length of the sling coverage is compared with the length of the beam outline to identify the integrity of the obstruction. The edges, center landing area, and outer range of the support bearing reference area are located to verify the visibility of the support area. Based on image sharpness, obstruction integrity, and support area visibility, frames are filtered one by one to remove invalid hoisting images where the beam outline and support bearing reference area cannot be identified. The retained images are then arranged in the order of acquisition to generate a valid hoisting image sequence.

[0028] S1.2 Extract the edge of the bearing reference area and the fixed reference position of the hoisting area from the effective hoisting image sequence, and make coordinate correspondence with the preset reference coordinates to calculate the reference geometric deviation and identify the image geometric deviation.

[0029] It should be noted that, from the effective hoisting image sequence, the edge pixels of the support bearing reference area and the fixed reference position of the hoisting area are extracted frame by frame. The edge continuity and feature point distribution are analyzed through edge detection and corner point recognition to establish the mapping relationship between edge points and reference positions. The extracted edge coordinates and hoisting area reference coordinates are matched with the preset reference coordinates point by point to calculate the horizontal deviation, vertical deviation and angular deviation. The reference geometric deviation is quantified by Euclidean distance and direction difference. The image geometric deviation is identified according to the magnitude and direction of the deviation.

[0030] It should also be noted that the reference coordinates are a set of fixed coordinates set based on the construction drawings, the actual installation position of the support, and the first frame calibration image. First, the fixed reference positions of the support bearing reference area edge, center landing area, and hoisting area are selected as reference points. Then, the image coordinate origin, horizontal axis, vertical axis, and scale relationship are determined. The reference points are converted into unified coordinate values ​​for subsequent point-to-point correspondence with the support edge and fixed reference position in the real-time hoisting image to determine the horizontal offset, vertical offset, and angular offset of the image.

[0031] S1.3. Establish a reference mapping relationship based on the geometric deviation of the images, and perform position reference geometric correction on the effective hoisting image sequence according to the reference mapping relationship to generate a hoisting reference image set.

[0032] It should be noted that, based on the geometric deviation of the image, the horizontal displacement, vertical displacement, and rotation angle are calculated frame by frame. The deviation is used as a mapping vector to establish a reference mapping relationship between the effective hoisting image sequence and the preset reference coordinates. According to the reference mapping relationship, the edge points and key feature points of each frame of the effective hoisting image are translated and rotated. The new coordinate positions of each pixel are calculated and resampled to generate a corrected image. At the same time, the edge continuity and feature point alignment error after correction are evaluated to determine whether the correction effect meets the deviation convergence requirement. After continuously processing the entire effective hoisting image sequence, a hoisting reference image set with completed position reference geometric correction is formed.

[0033] S2. Extract the grayscale gradient of the plate beam outline and the difference in texture of the sling occlusion from the hoisting reference image set, separate the sling occlusion area from the plate beam boundary area, and complete the fracture outline according to the continuous direction of the plate beam boundary to generate a repair outline sequence.

[0034] S2.1 Extract local images of the hoisting from the hoisting reference image set, and perform grayscale smoothing and local enhancement on the local images of the hoisting to generate enhanced analysis images.

[0035] It should be noted that the position of the bearing reference area of ​​the support, the position of the plate beam and the coverage of the sling are read frame by frame from the hoisting reference image set. The boundary coverage of the adjacent area of ​​the plate beam end is analyzed along the relative distribution direction of the bearing reference area of ​​the support and the plate beam. The edge stability of the bearing reference area of ​​the support, whether the outline of the plate beam falls within the bearing range and the degree of sling obstruction are judged. The interception boundary including the plate beam boundary, sling obstruction and the adjacent area of ​​the bearing reference area of ​​the support is determined, and the hoisting local image is intercepted according to the unified boundary. The grayscale distribution of the local image is read pixel by pixel. Neighborhood statistics are performed on isolated abrupt grayscale values ​​and fragmented noise points. Grayscale smoothing is completed by adjusting abrupt grayscale values ​​that do not form continuous edges. Then, the local grayscale span and grayscale contrast of adjacent areas are calculated. Positions with small spans and unclear boundary transitions are enhanced to improve the grayscale distinction between the beam outline and the sling occlusion and background, generating an enhanced analysis image.

[0036] The expression for calculating the local grayscale span and the grayscale contrast between adjacent regions is: ; in, For pixels The grayscale enhancement determination value; For pixels Maximum gray value within the neighborhood; For pixels The minimum gray value within the neighborhood; Weights are assigned based on the local grayscale range. The pixels to be analyzed in the local image of the hoisting process; For pixels Centered on the plate beam boundary, the average value of all pixel grayscale values ​​in the adjacent area on one side; In pixels Centered on the plate beam boundary, the average value of all pixel grayscale values ​​in the adjacent area on the other side.

[0037] S2.2 Extract the gray-level gradient of the beam outline and the difference in texture of the suspension cable occlusion from the enhanced analysis image. Encode the gray-level gradient of the beam outline according to the direction of boundary extension. Encode the difference in texture of the suspension cable occlusion according to the perturbation intensity of the neighborhood. Then, use the same pixel coordinate as the anchor point to perform hierarchical embedding to generate a joint outline texture map.

[0038] It should be noted that the gray-level distribution and local texture distribution are read pixel by pixel from the enhanced analysis image. First, the horizontal gray-level change, vertical gray-level change and diagonal gray-level change are calculated along the neighborhood of the pixel. The gray-level change in each direction is jointly evaluated. The evaluation process is based on whether the gray-level transitions between adjacent pixels are concentrated, whether the gray-level transition direction is continuous, and whether the gray-level transition position can be extended into the boundary direction in the neighborhood. The set of pixels with concentrated gray-level transitions, continuous direction and the ability to be extended into the boundary direction are identified as the gray-level gradient of the plate beam contour.

[0039] Texture difference analysis is performed on the local texture distribution in the enhanced analysis image. The analysis process is carried out according to the frequency of gray-level fluctuations, the amplitude of gray-level fluctuations and the degree of texture direction disorder within the pixel neighborhood. The degree of perturbation change of each pixel neighborhood relative to the surrounding neighborhood is calculated. Pixel sets with dense gray-level fluctuations, disordered perturbation directions and weaker texture continuity than the boundary continuity of the plate beam are identified as cable occlusion texture differences. After extracting the grayscale gradient of the slab beam outline, the grayscale gradient of the slab beam outline is sorted point by point along the extension direction of the slab beam boundary. The boundary extension path is established according to the consistency of the extension direction of adjacent gradient points and the continuity of the direction turning point. The extension direction corresponding to each gradient point is written into the direction sequence code according to the path order. After extracting the difference in the texture of the suspension cable occlusion, the disturbance intensity value is calculated with the pixel neighborhood as the range. The difference in the texture of the suspension cable occlusion is classified according to the size of the disturbance intensity value. The disturbance intensity value is written into the disturbance level code from low to high. After completing the direction sequence code and the disturbance level code, the correspondence is established with the same pixel coordinate as the anchor point. The direction sequence code and the disturbance level code at the same pixel coordinate are matched point by point. The hierarchical integration is performed according to the pixel coordinate position, the boundary extension direction and the neighborhood disturbance intensity, so that each pixel coordinate retains the direction attribute of the grayscale gradient of the slab beam outline and the disturbance attribute of the difference in the texture of the suspension cable occlusion. Finally, the outline texture joint map is generated.

[0040] S2.3. Select initial boundary chains with continuous gray-level transitions and consistent directional extension from the contour texture joint map, and locate the occlusion and breakage positions in the initial boundary chains where gray-level transitions drop abruptly, texture perturbation is enhanced, and boundary direction is interrupted, and generate breakage candidate chains.

[0041] It should be noted that, from the contour texture joint image, grayscale transition values, perturbation level codes, and directional sequence codes are read point by point along the contour direction of the beam. Pixels with continuous grayscale transition values, sequential directional sequence codes, and stable perturbation level codes are concatenated in the boundary extension order to form an initial boundary chain. A continuous neighborhood window is set along the initial boundary chain. First, the grayscale transition values ​​of the stable boundary in the previous section are used as a reference. The positions in the neighborhood window where the grayscale transition values ​​change from continuous high values ​​to low values, the grayscale difference between adjacent pixels narrows, and the boundary response weakens are marked as grayscale transition drop candidate points. Then, the perturbation level codes around the grayscale transition drop candidate points are read, and the perturbation level is increased from the stable level to... The locations where the dense perturbation level and texture direction change from unidirectional extension to intersecting and disordered, and where the perturbation range covers the boundary direction, are marked as texture perturbation enhancement candidate points. The direction sequence encoding before and after the texture perturbation enhancement candidate points is read again. The locations where the extension direction of the previous boundary point cannot be connected to the next boundary point, where there is a break in the gap between adjacent boundary points, and where the boundary extension path stops are marked as boundary direction interruption candidate points. The locations where the gray-level transition drop candidate points, texture perturbation enhancement candidate points, and boundary direction interruption candidate points overlap in the same pixel neighborhood are determined as occlusion breakage locations. The neighboring pixels of the breakage are connected in the order of the boundary points before and after the occlusion breakage location to generate a breakage candidate chain.

[0042] S2.4 Extract the front and rear boundary support segments on both sides of the fracture location from the fracture candidate chain, classify the area with boundary stability characteristics into the plate beam boundary area, classify the area with sudden increase in texture disturbance and interruption of closure into the suspension cable occlusion area, and generate a contour separation mark map.

[0043] It should be noted that the fracture location is located point by point from the fracture candidate chain. The front and rear boundary support segments are extracted along both sides of the fracture location in the order of boundary extension. The gray-level transition continuity, directional extension consistency, boundary point spacing change, and closure connection relationship of the front and rear boundary support segments are calculated. Regions with continuous gray-level transition, stable directional extension, stable boundary point spacing, and continuous connection with adjacent boundaries are identified as regions with boundary stability characteristics and are included in the plate-beam boundary region. Through the texture disturbance intensity, disturbance direction disorder, and closure / opening state of the fracture location neighborhood, regions with sudden increases in texture disturbance and interrupted closure are included in the suspension cable occlusion region. Then, corresponding markers are written for the plate-beam boundary region and the suspension cable occlusion region according to the pixel position to generate a contour separation marker map.

[0044] S2.5. Retain the boundary area of ​​the plate beam in the contour separation mark diagram, remove the area obscured by the suspension cable, and connect the boundary support segments on both sides of the fracture location to generate a complete contour chain.

[0045] It should be noted that pixels of the plate-beam boundary region are extracted from the contour separation mark map, pixels of the area obscured by the suspension cable are removed, and boundary support segments are read along both sides of the fracture location. The endpoint spacing, grayscale continuity, and directional extension consistency of the front and rear boundary support segments are used to determine whether the endpoints can be connected to form a continuous boundary. For the front and rear boundary support segments that can be connected, a connection path is established, the boundary extension direction is calculated point by point, and pixels with abnormal spacing are adjusted to make the connection boundary smooth and continuous. The retained plate-beam boundary region and the connected boundary support segments are merged in pixel order to generate a complete contour chain, providing a complete contour for subsequent boundary smoothing and closure verification.

[0046] S2.6 Perform boundary smoothing and closure verification on the supplementary contour chain, and sequentially splice the verified supplementary contour chain with the original plate beam boundary area to generate the repair contour sequence.

[0047] It should be noted that, for each segment of the completed contour chain, the connecting turning points, the spacing between adjacent boundary points, and the boundary extension direction are extracted. The previous boundary point, the current boundary point, and the next boundary point are used to form a sliding boundary segment. The continuity of the extension directions, the continuity of the turning points, and the stability of the spacing changes are verified. The positions where the extension direction changes abruptly are marked as abrupt turning points, and the positions where the spacing deviates from the continuous changes before and after are marked as spacing jump points. The positions of abnormal boundary points are corrected according to the extension direction and spacing trend of the stable boundaries before and after. The connection relationship between the two ends of the completed contour chain and the corresponding ends of the original plate beam boundary area is further verified. The completed contour chain with continuous connection of the ends, unidirectional connection, and closed reconnection without misalignment is embedded into the fracture position to generate a repair contour sequence.

[0048] S3. Extract the confidence level of the occluded and completed segments and the end-line closure points from the repair contour sequence, and deduct the high-confidence closure points back to the support bearing reference area to generate the plate beam support corresponding diagram.

[0049] S3.1 Extract the connection position between the occluded fill segment and the original plate beam boundary area from the repair contour sequence, and verify the connection direction, grayscale continuity and texture attenuation along the end of the plate beam to generate the end fill connection chain.

[0050] It should be explained that, from the repair contour sequence, the endpoints of the occluded completion segment, the endpoints of the original slab beam boundary area, and adjacent boundary pixels are read frame by frame. Positions where the endpoints of the occluded completion segment and the endpoints of the original slab beam boundary area are close in distance and adjacent in their boundary extension directions are found according to the pixel arrangement order, and these positions are marked as continuation positions. Based on the continuation position, the boundary directions before and after the continuation position are read along the end direction of the slab beam, and the continuation direction verification information is obtained by comparing whether the extension direction of the occluded completion segment and the extension direction of the original slab beam boundary area are in the same direction. The verification information for the continuation direction continues to read the grayscale transition values ​​on both sides of the continuation position, and compares whether the grayscale at the end of the occluded and completed section transitions smoothly with the initial grayscale of the original plate beam boundary area to obtain grayscale continuity information; based on the grayscale continuity information, the intensity of texture disturbance in the neighborhood of the continuation position is further read to determine whether the residual texture of the cable occlusion gradually weakens along the end of the plate beam to obtain texture attenuation information; the continuation position, continuation direction verification information, grayscale continuity information and texture attenuation information are connected in series according to the end direction of the plate beam to generate the end completion continuity chain.

[0051] S3.2 Based on the end-filling support chain, the boundary continuity, residual shading disturbance, and bearing width matching status of the shading filler segment are evaluated in tandem to generate the confidence level of the shading filler segment.

[0052] It should be noted that, based on the end-filling connection chain, the connection position, connection direction verification information, grayscale connection information, and texture attenuation information are read item by item. First, the boundary point spacing, extension direction, and grayscale transition continuity are read back along both ends of the occluded filling section to the original plate beam boundary area. The corresponding positions with stable boundary point spacing, unidirectional extension direction, and smooth grayscale transition are marked as valid boundary continuity positions, and boundary continuity evaluation values ​​are formed.

[0053] Based on the boundary continuity evaluation value, the texture disturbance intensity and texture attenuation information of the neighborhood of the occlusion completion segment are read. The corresponding positions where the residual texture of the cable occlusion gradually weakens along the end of the plate beam and does not cover the effective position of the boundary continuity are marked as controlled positions of disturbance residue, and a disturbance residue evaluation value is formed. Based on the disturbance residue evaluation value, the end bearing width corresponding to the occlusion completion segment and the end bearing width corresponding to the original plate beam boundary area are read. The spacing between the two side boundaries, the spacing between the end edge lines and the change of bearing width are compared to see if they are consistent. The positions where the width change is stable and matches the bearing width of the original plate beam boundary area are marked as bearing width matching positions, and a bearing width matching evaluation value is formed. The boundary continuity evaluation value, disturbance residue evaluation value and bearing width matching evaluation value are merged according to the same continuation position. The occlusion completion segment that meets the requirements of continuity, control and matching in the corresponding evaluation is marked with a high confidence mark. The occlusion completion segment with disconnection, residue or width deviation in the corresponding evaluation is marked with a low confidence mark, and the confidence level of the occlusion completion segment is generated.

[0054] S3.3. Select high-confidence occlusion completion segments from the confidence of the occlusion completion segments, and extend the boundary support segments on both sides of the high-confidence occlusion completion segments towards each other along the end closure direction to generate end line closure biting points.

[0055] It should be noted that, from the confidence level of the shading and filling section, the high confidence marker, the continuation position, the boundary continuity evaluation value, the disturbance residual evaluation value, and the load-bearing width matching evaluation value are read item by item. The shading and filling section that simultaneously meets the requirements of boundary continuity stability, controlled shading disturbance, and load-bearing width matching is selected as the high confidence shading and filling section. Based on the continuation position of the high confidence shading and filling section, the stable boundary points in the boundary areas of the adjacent original plate beams are read to both sides respectively. The boundary support sections on both sides of the high confidence shading and filling section are determined according to the requirements of stable boundary point spacing, continuous gray-scale transition, and consistent directional bearing.

[0056] Read the endpoint coordinates and extension direction of the two boundary support sections along the end of the plate beam. Determine the end closure direction based on the direction in which the two endpoints approach each other. Based on the end closure direction, push the two boundary support sections inward along their original extension directions. During the pushing process, verify whether the directional angle, endpoint distance, and gray scale connection of the two extension paths gradually converge. Determine the position where the two extension paths approach each other and form an end closure as the interlocking candidate position. Then, verify the matching state between the interlocking candidate position and the bearing width of the high-confidence shading and filling section. Determine the interlocking candidate position that meets the closure requirements in terms of directional connection, gray scale connection, and bearing width as the end line closure interlocking point.

[0057] S3.4. Based on the end line closed engagement point, the end bearing edge line is hooked back along the bearing boundary normal of the bearing reference area of ​​the support, and the engagement deviation position and the corresponding bearing position are marked to generate the plate beam support correspondence diagram.

[0058] It should be noted that, based on the end-line closure engagement point, the endpoint coordinates, line segment direction, and end-line closure position of the end bearing edge line are read frame by frame, and the bearing boundary, boundary normal, and bearing range of the support bearing reference area are read simultaneously. First, the retraction start position is determined along the end bearing edge line with the end-line closure engagement point as the starting point. Then, the retraction direction is established along the bearing boundary normal of the support bearing reference area. The end bearing edge line is projected onto the bearing boundary of the support bearing reference area according to the retraction direction to obtain the normal correspondence between the end bearing edge line and the bearing boundary. Based on the normal correspondence, the normal distance, tangential misalignment, and end-line drop point offset from the end-line closure engagement point to the bearing boundary are calculated. The position deviating from the bearing boundary is marked as the engagement deviation position, and the position where the end bearing edge line falls into the bearing range of the support bearing reference area after projection is marked as the bearing corresponding position. The end-line closure engagement point, end bearing edge line, engagement deviation position, and bearing corresponding position are connected according to the same image frame and the same end position to generate the plate beam support correspondence diagram.

[0059] S4. Extract the interlocking point deviation and coincidence increment from the plate beam support correspondence diagram, establish a unidirectional approach chain according to the deviation decreasing consistency, suppress the reverse mutation correspondence, and generate real-time positioning deviation data.

[0060] S4.1 Read the interlocking deviation position, bearing corresponding position and image frame time sequence from the plate beam support correspondence diagram, and concatenate the position correspondence of the same end line closed interlocking point in adjacent image frames to generate an interlocking point frame alignment sequence.

[0061] It should be noted that, from the image frame corresponding to the plate beam support, the end line closure engagement point, engagement deviation position, and load-bearing corresponding position are read frame by frame in chronological order. The end line closure engagement point in each image frame is bound with the corresponding engagement deviation position and load-bearing corresponding position to form a single-frame engagement positioning record. Based on the single-frame engagement positioning record, the end line closure engagement point in the previous image frame is used as the tracking starting point. In the next image frame, end line closure engagement points with adjacent end positions, consistent engagement deviation directions, and load-bearing corresponding positions falling within the same support load-bearing reference area are searched. The found end line closure engagement points are determined as the corresponding points in the next frame for the same end line closure engagement point. The search for corresponding points in the next frame is repeated in chronological order according to the image frame. The changes in engagement deviation position, load-bearing corresponding position, and frame sequence relationship of the same end line closure engagement point in adjacent image frames are sequentially concatenated to generate an inter-frame alignment sequence of engagement points.

[0062] S4.2 Extract the normal deviation distance of the end line closed engagement point relative to the corresponding position of the bearing from the inter-frame alignment sequence of the engagement point as the engagement point deviation, and extract the overlap change between the end bearing edge line and the bearing reference area of ​​the support as the overlap increment to generate the deviation overlap linkage sequence.

[0063] It should be noted that, from the inter-frame alignment sequence of the biting point, the end line closure biting point, the corresponding bearing position, the end bearing edge, and the bearing reference area of ​​the support are read frame by frame in the time sequence of the image frames. First, the boundary normal is determined by the bearing boundary of the bearing reference area where the bearing position is located. Then, the end line closure biting point is projected along the boundary normal to the bearing position. The normal interval between the end line closure biting point and the bearing position is read and written into the biting point deviation. Based on the biting point deviation, the overlap range between the end bearing edge and the bearing reference area in the same image frame is read. The coverage length, coverage width, and overlap area position of the end bearing edge falling into the bearing reference area are extracted and compared with the coverage length, coverage width, and overlap area position in the previous image frame. The change in the overlap range is written into the overlap increment. The biting point deviation, overlap increment, end line closure biting point, and image frame time sequence are bound frame by frame to generate a deviation overlap linkage sequence.

[0064] S4.3. Based on the deviation coincidence linkage sequence, perform same-frame verification on the near consistency state in adjacent image frames to generate a near-locking sequence. According to the near-locking sequence, connect the corresponding relationships where the biting point deviation continues to decrease and the coincidence increment continues, and suppress the corresponding relationship where the biting point deviation increases in the opposite direction to generate a unidirectional nearing chain.

[0065] It should be noted that, based on the deviation and coincidence linkage sequence, the end-line closure engagement point, engagement point deviation, coincidence increment, and corresponding bearing position are read frame by frame in the time sequence of the image frames. First, the engagement point deviation in the next image frame is compared with the engagement point deviation in the previous image frame by decreasing, and the position where the engagement point deviation decreases is marked as the deviation decrease position. Based on the deviation decrease position, it is further checked whether the coincidence increment in the next image frame continues to increase or remain continuous relative to the previous image frame, and the position where the coincidence increment is not interrupted is marked as the coincidence continuation position.

[0066] Based on the overlapping and continuing position, continue to check whether the offset direction of the end line closure biting point relative to the bearing corresponding position continuously points to the bearing reference area of ​​the support. Mark the positions where the offset direction continuously approaches as the directional bearing position. Determine the positions where the deviation decreases, the overlapping and continuing position and the directional bearing position coincide within the same image frame as the approaching consistency state, and concatenate the approaching consistency states according to the time sequence of the image frames to generate the approaching locking sequence. According to the approaching locking sequence, connect the corresponding relationships of adjacent image frames where the biting point deviation continuously decreases and the overlapping increment continues in sequence. Mark the corresponding relationships where the biting point deviation increases in the opposite direction or the overlapping increment is interrupted as the reverse mutation correspondence and suppress them from the connection relationship to generate a unidirectional approaching chain.

[0067] S4.4. Based on the unidirectional approach chain, retain the approach consistency state within the chain, and organize the deviation changes of the plate beam end relative to the support bearing reference area according to the time sequence of the image frames to generate real-time positioning deviation data.

[0068] It should be noted that, based on the unidirectional approach chain, the approach consistency state, end-line closure engagement point, engagement point deviation, overlap increment, and load-bearing corresponding position are read frame by frame in the time sequence of the image frames. First, the suppressed reverse mutation corresponding to the unidirectional approach chain is removed, and only the approach consistency state within the unidirectional approach chain, where the engagement point deviation decreases, the overlap increment continues, and the offset direction continuously approaches the support load-bearing reference area, is retained. Based on the retained approach consistency state, the normal deviation, tangential misalignment, overlap range, and end-line landing point of the plate beam end relative to the support load-bearing reference area are read back frame by frame. The changes in normal deviation, tangential misalignment, and overlap range between adjacent image frames are then concatenated in chronological order to form the deviation change of the plate beam end relative to the support load-bearing reference area. Based on the deviation change, the engagement point deviation, overlap increment, load-bearing corresponding position, and approach consistency state corresponding to each image frame are bound and organized to generate real-time positioning deviation data.

[0069] S5. Extract the deviation decreasing continuation segment and the deviation reverse sudden change segment from the real-time positioning deviation data, determine the deviation evolution state according to the degree of deviation approaching stability, perform positioning verification and adjustment prompts on the deviation evolution state, and generate a hoisting deviation report.

[0070] S5.1 Extract the positioning deviation frame sequence from the real-time positioning deviation data, and continuously merge the positioning deviation frame sequences to generate a deviation evolution chain.

[0071] It should be explained that the remaining positioning deviation and offset change trend are read from the real-time positioning deviation data according to the corresponding frame identifier, and the remaining positioning deviations are arranged in the order of the corresponding frame identifiers to form the positioning deviation frame sequence. The remaining positioning deviation, offset direction and offset magnitude of adjacent frames are read frame by frame along the positioning deviation frame sequence. The remaining positioning deviation of the next frame is compared with the remaining positioning deviation of the previous frame. When comparing, it is found that the offset direction is consistent, the offset magnitude is progressive along the same change trend, and the corresponding frame identifiers are arranged continuously.

[0072] If the offset direction of adjacent frames remains continuous, the offset magnitude progresses along the same trend, and the corresponding frame identifiers are arranged continuously, then adjacent frames are grouped into the same continuous frame segment; if the offset direction of adjacent frames is reversed, the offset magnitude changes abruptly, or the corresponding frame identifiers are broken, then the continuous frame segment is broken at the corresponding position; each continuous frame segment is numbered according to the arrangement order of the positioning deviation frame sequence, and the remaining positioning deviation, offset change trend, and corresponding frame identifier of each continuous frame segment are bound together to generate a deviation evolution chain.

[0073] It should also be noted that the deviation evolution chain is used to connect the deviations of each image frame in the real-time positioning deviation data in chronological order, and record the continuous change relationship between the remaining positioning deviation, offset direction, offset magnitude and the corresponding frame identifier, so as to transform the single frame deviation judgment into the continuous frame segment judgment. Through the deviation evolution chain, it is possible to identify whether the deviation is continuously decreasing, whether there is a reverse sudden change, or whether it is close to stabilization, so as to provide a continuous basis for subsequent division of deviation decreasing continuation segment, deviation reverse sudden change segment, and generation of review prompts and adjustment prompts.

[0074] S5.2 Identify the decreasing continuity of the remaining positioning deviation and the end swing situation based on the deviation evolution chain, and mark the frame segment that continuously decreases and does not swing at the end as the deviation decreasing continuation segment.

[0075] It should be noted that, according to the deviation evolution chain, the remaining positioning deviation, offset direction and corresponding frame identifier in the continuous frame segments are extracted segment by segment. The remaining positioning deviation of adjacent frames is verified in frame order to see if it continuously decreases, if the offset direction continuously points to the bearing reference area of ​​the support, and if the offset amplitude shows a converging trend. Frame segments that continuously meet the above trends are identified as frame segments with decreasing continuity. The remaining positioning deviation and offset direction at the end of the continuous frame segments are extracted. The remaining positioning deviation at the end is verified to see if it increases and if the offset direction at the end deviates from the bearing reference area of ​​the support. Frame segments that do not increase and do not deviate are identified as frame segments that do not swing back at the end and are marked as deviation decreasing continuation segments.

[0076] S5.3 Based on the deviation evolution chain, identify the reverse change of offset between adjacent frame segments, and mark the frame segments where the remaining positioning deviation suddenly increases, the offset direction reverses and the stability of the overlapping domain decreases as deviation reverse mutation segments.

[0077] It should be noted that, based on the deviation evolution chain, the remaining positioning deviation, offset direction, and centroid deviation of the overlapping region of adjacent frame segments are read according to the consecutive frame segment numbers. First, the remaining positioning deviation of the starting frame of the next frame segment is compared with the remaining positioning deviation of the ending frame of the previous frame segment to check for abrupt changes. During the comparison, it is determined whether the remaining positioning deviation of the starting frame of the next frame segment is greater than the remaining positioning deviation of the ending frame of the previous frame segment, and whether the increase in deviation disrupts the decreasing trend of the previous frame segment. Then, the offset direction of the starting frame of the next frame segment is compared with the offset direction of the ending frame of the previous frame segment in reverse. During the comparison, it is determined whether the remaining positioning deviation of the starting frame of the next frame segment is greater than the remaining positioning deviation of the ending frame of the previous frame segment, and whether the increase in offset direction disrupts the decreasing trend of the previous frame segment. Check whether the offset direction of the initial frame deviates from the support bearing reference area and whether it forms the opposite direction to the offset direction of the last frame of the previous frame segment; continue to perform a stability check between the centroid deviation of the overlapping domain of the starting frame of the next frame segment and the centroid deviation of the overlapping domain of the last frame of the previous frame segment. When checking, compare whether the centroid deviation of the overlapping domain increases, whether the centroid movement direction of the overlapping domain moves away from the support bearing reference area, and whether the overlapping area changes from continuous coverage to partial separation; mark the frame segment that simultaneously satisfies the following conditions: a sudden increase in the remaining positioning deviation, a reversal of the offset direction, and a decrease in the stability of the overlapping domain as a deviation reversal mutation segment.

[0078] S5.4 Based on the deviation reverse mutation segment and the deviation decreasing continuation segment, the decreasing segment is strung together as a stable main chain, and the mutation segment is embedded as a perturbation side chain. The deviation evolution state is divided according to whether the side chain is absorbed by the subsequent decreasing segment.

[0079] It should be noted that the deviation decreasing continuation segment and the deviation reverse mutation segment are read frame by frame from the deviation evolution chain. The consecutive deviation decreasing continuation segments are sequentially connected in time order to establish a stable main chain. The adjacent deviation reverse mutation segments are located along the main chain, and the coverage of the start and end points of the mutation segment relative to the preceding and following decreasing segments is evaluated. The mutation segments are then embedded into the main chain to form a perturbation side chain.

[0080] Frame by frame, determine whether the perturbation sidechain is covered or absorbed by the subsequent decreasing segment. If the subsequent decreasing segment covers all or most of the perturbation sidechain, the sidechain is determined to be absorbed; otherwise, it is determined not to be absorbed. Mark the deviation evolution state of the corresponding frame segment according to the absorption status of the perturbation sidechain. Mark the frame segment corresponding to the decreasing segment as the stable state and mark the frame segment corresponding to the sidechain that was not absorbed as the swingback or pending verification state, thus generating a complete deviation evolution state sequence.

[0081] S5.5 Extract the frame segment positions corresponding to the near-stable state, the deviation swing state, and the state to be reviewed from the deviation evolution state, and read back the overlap relationship between the end bearing edge line and the support bearing reference area for the frame segment position corresponding to the near-stable state.

[0082] It should be noted that the frame positions corresponding to the near-stable state, the deviation swing state, and the state to be reviewed are extracted frame by frame from the deviation evolution state. The endpoint coordinates of the end bearing edge and the boundary pixels of the overlapping area are extracted sequentially along the frame sequence. The continuity and integrity of the endpoint covering the support bearing reference area are evaluated by the normal distance, tangential position, and longitudinal landing point from the endpoint of the end bearing edge to the edge of the support bearing reference area. It is determined whether the end bearing edge and the support bearing reference area remain continuously overlapping in each frame. Frame segments that are continuously overlapping, have stable normal and tangential distances, and complete coverage are identified as overlapping frames, providing a basis for the generation of subsequent review prompts and adjustment prompts.

[0083] It should also be noted that the near-stable state indicates that the end of the slab beam and the bearing reference area of ​​the support have been continuously approaching each other, the remaining positioning deviation is decreasing and the coincidence relationship remains stable, and it can be used as the main basis for positioning verification.

[0084] The deviation swing state indicates that the end of the plate beam has reversed or the coincidence stability has decreased during the approach process, indicating that the hoisting posture may swing back and an adjustment prompt needs to be triggered.

[0085] The "pending verification" status indicates that the deviation trend has not yet stabilized or formed a clear retracement, and it is necessary to continue to observe the changes in subsequent frames to avoid prematurely determining the in-place status.

[0086] S5.6 Extract the frame position corresponding to the deviation swing state from the overlap maintenance relationship, track the change in offset direction and the decrease in the stability of the overlap domain in the deviation reverse mutation segment, and generate adjustment prompts.

[0087] It should be noted that the frame segment positions corresponding to the deviation swing state are read frame by frame from the overlap maintenance relationship. The lateral offset, longitudinal landing point and tangential misalignment of the end bearing edge line in the deviation reverse change segment are tracked along the frame sequence. The offset direction change of consecutive frames is calculated, and the centroid position of the overlap domain is read simultaneously. The magnitude and continuity of the centroid deviating from the support bearing reference area are evaluated, and the frame segments that continuously increase or reverse direction are identified. The frame segments with obvious offset direction changes and decreased overlap domain stability are recorded point by point and arranged in frame sequence to form adjustment prompts.

[0088] S5.7 Extract the frame segment positions corresponding to the state to be reviewed from the adjustment prompts, mark the frame segments that do not meet the near-stable state and have not formed a continuous deviation swing state as continuous observation items, and generate a review prompt sequence.

[0089] It should be noted that the frame positions corresponding to the state to be reviewed are read frame by frame from the adjustment prompts. The endpoint positions, offset directions, normal deviations, and centroid deviations of the end bearing edge lines are analyzed along the frame sequence to determine whether each frame meets the requirements of a near-stable state or a continuous deviation swing state. Frame segments that have not achieved stable overlap, whose endpoint offset directions continue to change, or whose centroids of the overlap region have not formed continuous swings are identified as abnormal frame segments and marked as continuous observation items in chronological order. Finally, all continuous observation items are arranged in frame order to generate a review prompt sequence.

[0090] S5.8. Based on the review prompt sequence, arrange the review prompt information and generate a hoisting deviation report.

[0091] It should be noted that, according to the review prompt sequence, the continuous observation items, adjustment prompt items, and near-stable state frame positions are read frame by frame. Information such as end bearing edge line offset, normal deviation, tangential misalignment, angular deflection, and centroid deviation of the overlapping region are integrated according to the frame order. The deviation evolution trend and abnormal change points of each frame are analyzed, and deviation reduction segments, reverse sudden change segments, and unstabilized frame segments are identified. Different types of frame segments are classified and summarized and identification information is generated. Then, the review prompt content is arranged according to the frame order and type order to form a coherent and traceable frame segment deviation description. The sorted hoisting deviation report is then output.

[0092] In summary, this invention achieves completeness of the beam boundary information and high-precision repair contour sequence generation by separating the sling-obscured area from the beam boundary and completing the fracture contour through image recognition. At the same time, it tracks the position offset features and coincidence changes of the support bearing reference area along adjacent image frames, and achieves real-time capture of beam offset and distinguishes between decreasing trends and reverse abrupt changes, thereby improving the accuracy of positioning, dynamic monitoring and safety reliability of the hoisting process.

[0093] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for detecting positioning deviations during boiler beam hoisting based on image recognition, characterized in that, include: Acquire real-time images of the hoisting operation, perform benchmark mapping correction on the real-time images of the hoisting operation, and generate a set of hoisting benchmark images; Extract the grayscale gradient of the plate beam outline and the difference in texture of the sling occlusion from the hoisting reference image set, separate the sling occlusion area from the plate beam boundary area, and complete the fracture outline according to the continuous direction of the plate beam boundary to generate a repair outline sequence. Extract the confidence level of the occluded and complete segments and the end-line closure and interlocking points from the repair contour sequence, and deduct the high-confidence interlocking points back to the support bearing reference area to generate the plate beam support corresponding map; Extract the interlocking point deviation and coincidence increment from the plate beam support correspondence diagram, establish a unidirectional approach chain according to the deviation decreasing consistency, suppress the reverse sudden change correspondence, and generate real-time positioning deviation data. Extract the decreasing deviation continuation segment and the reverse abrupt deviation segment from the real-time positioning deviation data, determine the deviation evolution state according to the degree of deviation approaching stability, perform positioning verification and adjustment prompts on the deviation evolution state, and generate a hoisting deviation report.

2. The method for detecting positioning deviations in boiler beam hoisting based on image recognition as described in claim 1, characterized in that, The process of acquiring real-time hoisting images, performing reference mapping correction on these images, and generating a hoisting reference image set involves the following steps: Real-time images of hoisting are acquired, and the image clarity, occlusion integrity, and support area visibility of the real-time images are identified. Invalid hoisting images that cannot identify the outline of the beam and the support bearing reference area are removed, and a valid hoisting image sequence is generated. Extract the edge of the bearing reference area and the fixed reference position of the hoisting area from the effective hoisting image sequence, match them with the preset reference coordinates, calculate the reference geometric deviation, and identify the image geometric deviation; A reference mapping relationship is established based on the geometric deviation of the images, and the position reference geometry is corrected for the effective hoisting image sequence according to the reference mapping relationship to generate a hoisting reference image set.

3. The method for detecting positioning deviations in boiler beam hoisting based on image recognition as described in claim 2, characterized in that, The specific steps for extracting the grayscale gradient of the beam outline and the difference in texture caused by sling occlusion from the hoisting reference image set are as follows: Local images of the hoisting are extracted from the hoisting reference image set, and grayscale smoothing and local enhancement are performed on the local images to generate enhanced analysis images; The gray-level gradient of the beam outline and the difference in texture of the suspension cable occlusion are extracted from the enhanced analysis image. The gray-level gradient of the beam outline is encoded in the direction sequence according to the boundary extension direction. The difference in texture of the suspension cable occlusion is encoded in the perturbation level according to the perturbation intensity of the neighborhood. The layers are then interlocked using the same pixel coordinate as the anchor point to generate a joint outline texture map.

4. The method for detecting positioning deviations in boiler beam hoisting based on image recognition as described in claim 1, characterized in that, The separation of the sling-covered area and the boundary area of ​​the plate beam is performed, and the fracture contour is completed according to the continuous direction of the plate beam boundary to generate a repair contour sequence. The specific steps are as follows: From the contour texture joint map, select the initial boundary chain with continuous gray-level transitions and consistent direction extension, and locate the occlusion and breakage positions in the initial boundary chain where gray-level transitions drop abruptly, texture perturbation is enhanced, and boundary direction is interrupted, and generate the breakage candidate chain. Extract the front and rear boundary support segments on both sides of the fracture location from the fracture candidate chain, classify the area with boundary stability characteristics into the plate beam boundary area, classify the area with sudden increase in texture disturbance and interruption of closure into the suspension cable occlusion area, and generate a contour separation mark map. The boundary area of ​​the plate beam in the contour separation mark map is retained, the area obscured by the suspension cable is removed, and the boundary support segments on both sides of the fracture location are connected to generate the contour chain. The completed contour chain is smoothed and closed, and the completed contour chain that matches the check is sequentially spliced ​​with the original plate beam boundary area to generate the repair contour sequence.

5. The method for detecting positioning deviations in boiler beam hoisting based on image recognition as described in claim 4, characterized in that, The specific steps for extracting the confidence score of the occluded completion segment and the end-line closure biting point from the repaired contour sequence are as follows: Extract the connection position between the occluded fill segment and the original plate beam boundary area from the repair contour sequence, and verify the connection direction, grayscale continuity and texture attenuation along the end of the plate beam to generate the end fill connection chain. Based on the end-filling support chain, the boundary continuity, residual shading disturbance, and bearing width matching status of the shading filler segment are evaluated in tandem to generate the confidence level of the shading filler segment. High-confidence occlusion completion segments are selected from the confidence scores of the occlusion completion segments, and the boundary support segments on both sides of the high-confidence occlusion completion segments are extended towards each other along the end closure direction to generate end line closure engagement points.

6. The method for detecting positioning deviations in boiler plate beam hoisting based on image recognition as described in claim 1, characterized in that, The plate beam support correspondence diagram is obtained by performing a back-hook connection of the end bearing edge line along the bearing boundary normal of the bearing reference area based on the end line closed engagement point, and marking the engagement deviation position and the bearing corresponding position.

7. The method for detecting positioning deviations in boiler beam hoisting based on image recognition as described in claim 6, characterized in that, The specific steps for extracting the interlocking point deviation and coincidence increment from the corresponding diagram of the plate beam support are as follows: Read the interlocking deviation position, the corresponding bearing position and the time sequence of the image frames from the corresponding diagram of the plate beam support, and connect the positional correspondence of the same end line closed interlocking point in adjacent image frames to generate an interlocking point frame alignment sequence. The normal deviation distance of the end line closure engagement point relative to the corresponding bearing position is extracted from the inter-frame alignment sequence of the engagement point as the engagement point deviation, and the overlap change between the end bearing edge line and the bearing reference area of ​​the support is extracted as the overlap increment to generate the deviation overlap linkage sequence.

8. The method for detecting positioning deviations in boiler beam hoisting based on image recognition as described in claim 7, characterized in that, The steps for establishing a unidirectional approach chain based on decreasing deviation consistency, suppressing reverse mutation correspondences, and generating real-time in-place deviation data are as follows: Based on the deviation coincidence linkage sequence, the near consistency state in adjacent image frames is verified in the same frame to generate a near locking sequence. According to the near locking sequence, the correspondence of continuously decreasing biting point deviation and continuous coincidence increment is linked together, and the correspondence of reverse increase biting point deviation is suppressed to generate a one-way nearing chain. Based on the unidirectional approach chain, the approach consistency state within the chain is preserved, and the deviation changes of the plate beam end relative to the support bearing reference area are sorted according to the time sequence of image frames to generate real-time positioning deviation data.

9. The method for detecting positioning deviations in boiler beam hoisting based on image recognition as described in claim 1, characterized in that, The steps for extracting the decreasing deviation continuation segment and the reverse abrupt deviation segment from the real-time positioning deviation data, and determining the deviation evolution state according to the degree of deviation approaching stability, are as follows: The in-place deviation frame sequence is extracted from the real-time in-place deviation data, and the in-place deviation frame sequence is continuously merged to generate a deviation evolution chain. Based on the deviation evolution chain, the decreasing continuity of the remaining positioning deviation and the end swing situation are identified, and the frame segments that continuously decrease and do not swing at the end are marked as deviation decreasing continuation segments. Based on the deviation evolution chain, the reverse change of offset between adjacent frame segments is identified, and the frame segments in which the remaining positioning deviation suddenly increases, the offset direction is reversed, and the stability of the overlapping domain decreases are marked as deviation reverse mutation segments. Based on the deviation reverse mutation segment and the deviation decreasing continuation segment, the decreasing segment is strung together as a stable main chain, and the mutation segment is embedded as a perturbation side chain. The deviation evolution state is divided according to whether the side chain is absorbed by the subsequent decreasing segment.

10. The method for detecting positioning deviations in boiler beam hoisting based on image recognition as described in claim 9, characterized in that, The specific steps for verifying and adjusting the deviation evolution status and generating a hoisting deviation report are as follows: Extract the frame segment positions corresponding to the near-stable state, the deviation swing state, and the state to be reviewed from the deviation evolution state, and read back the overlap relationship between the end bearing edge line and the support bearing reference area for the frame segment position corresponding to the near-stable state. Extract the frame position corresponding to the deviation swing state from the overlap preservation relationship, track the change in offset direction and the decrease in the stability of the overlap domain in the deviation reverse mutation segment, and generate adjustment prompts. Extract the frame segment positions corresponding to the state to be reviewed from the adjustment prompts, mark the frame segments that do not meet the near-stable state and have not formed a continuous deviation swing state as continuous observation items, and generate a review prompt sequence; Based on the review prompt sequence, the review prompt information is arranged to generate a hoisting deviation report.

Images