A large-scale ring structure hoisting deformation monitoring method and system based on panoramic vision and deep learning

Abstract

The application provides a large-scale ring structure hoisting deformation monitoring method and system based on panoramic vision and deep learning. The method uses a single panoramic camera to collect video streams synchronously with the structure, and reconstructs a full-circumferential distortion-free window in real time through double-cubic projection technology. To meet the real-time requirements of the hoisting site, a lightweight YOLOv8 model integrating partial convolution and efficient attention mechanism is used to improve the dynamic response speed of the edge computing terminal for target identification. During the hoisting operation, the system uses sub-pixel positioning of the inner and outer targets and uses a difference algorithm to immediately eliminate the rigid body displacement interference caused by the movement of the camera with the structure, thereby directly calculating the real elastic deformation of the structure under the hoisting state. The application realizes an automatic operation mode of "hoisting, monitoring and feedback at the same time", effectively solving the problem of real-time and accurate acquisition of deformation data of large-scale components in a moving base environment.

Description

Technical Field

[0001] This invention relates to the field of structural health monitoring and computer vision measurement technology, and in particular to a method and system for monitoring the deformation of a large-scale ring structure during hoisting based on panoramic vision and deep learning. Background Technology

[0002] With the development of industrialized construction, the overall hoisting of large steel cages and ring-shaped steel structures is becoming increasingly common. During the hoisting process, the structure is prone to deformation due to its own weight and dynamic loads. Excessive deformation may lead to structural yielding or even failure. Therefore, real-time monitoring of deformation during the hoisting process is crucial.

[0003] Traditional deformation monitoring methods mainly rely on contact sensors (such as displacement gauges and strain gauges), which require laying a large number of cables on the structural surface. This is not only cumbersome and inefficient to install, but also easily interferes with construction operations and makes it difficult to achieve full-site coverage. Although multi-camera vision measurement systems can provide non-contact measurement, the system hardware costs are high, and the process of multi-camera synchronization and joint calibration is complex, making it inconvenient to deploy quickly on the construction site.

[0004] Panoramic cameras, with their advantage of providing a 360-degree field of view with a single unit, are increasingly being used in monitoring. However, their application in dynamic hoisting monitoring still faces three major technical bottlenecks:

[0005] (1) Problem of measuring moving base: Panoramic cameras are usually installed on the suspended object and move synchronously with the structure. The camera's own shaking causes the measured displacement to include rigid body motion components, making it difficult to separate the real elastic deformation of the structure.

[0006] (2) Image distortion and small target recognition: The panoramic image has severe distortion, and the background of the hoisting site is extremely complex (including tower cranes, scaffolding, and strong light interference), making it difficult for standard algorithms to stably track small targets at a distance.

[0007] (3) Insufficient real-time performance: The data volume of high-resolution panoramic video is huge, and traditional methods are difficult to achieve high frame rate real-time calculation and early warning on edge devices on site. Summary of the Invention

[0008] Purpose of the invention: To propose a method for monitoring the deformation of large-scale ring structure hoisting based on panoramic vision and deep learning, and to propose a system for implementing this method. The aim is to use a panoramic camera in conjunction with an edge computing terminal, through dual projection reconstruction, lightweight deep learning network and relative displacement difference algorithm, to solve the measurement problem under dynamic base conditions and achieve high-precision, low-cost and real-time full-field deformation monitoring.

[0009] To achieve the above objectives, this invention provides a method for monitoring the deformation of a large-scale ring structure during hoisting based on panoramic vision and deep learning. This method includes the following steps: S1. A panoramic camera is rigidly fixed at the geometric center of the large-scale ring structure to be hoisted, and feature targets are arranged on the inner and outer nodes in the same radial direction of the structure to acquire panoramic video streams in real time as the panoramic camera moves synchronously with the structure. S2. Using the dual cube projection method, a mapping relationship between the panoramic spherical coordinate system and the planar perspective projection is established, and each frame of the panoramic video stream is decomposed and reconstructed into several distortion-free planar monitoring windows covering the circumference of the structure. S3. Construct and train an improved YOLOv8 deep learning network model. Input the distortion-free planar monitoring window obtained in step S2 into the deep learning network model. Under the complex background of hoisting dynamics, identify the feature targets at the inner nodes and the feature targets at the outer nodes in real time, and output the region of interest (ROI) containing the feature targets. S4. Perform image preprocessing on the region of interest (ROI) obtained in step S3, extract the target contour using Otsu adaptive thresholding and morphological operations, and calculate the sub-pixel level pixel coordinates of the target center using the image moment algorithm. S5. Based on the virtual camera imaging model, the target pixel displacement is converted into physical space displacement, eliminating the rigid body displacement component generated by the panoramic camera with the structure movement, and calculating the real elastic deformation of the structure at the outer node relative to the inner node.

[0010] As a preferred solution, a fisheye panoramic camera is used to capture the panoramic video stream of the hoisting process. The fisheye panoramic camera is equipped with two ultra-wide-angle fisheye lenses, each with a field of view of more than 180°, which together cover a 360° spherical space. The image projection module adopts dual cube projection technology, which reconstructs the panoramic field of view into eight independent distortion-free planar monitoring windows through two sets of perspective projections that are rotated 45° relative to each other, ensuring that all targets are located in the low-distortion center area of ​​the window.

[0011] As a preferred embodiment, the double cube projection method in step S2 specifically includes: With 0°, 90°, 180°, and 270° as the main optical axes, the first layer of perspective projection is executed to generate the first set of four planar windows; Rotate 45° clockwise relative to the first projection, with 45°, 135°, 225°, and 315° as the main optical axis directions, and perform the second perspective projection to generate the second set of 4 planar windows; Combining the two sets of windows above forms a sequence of eight planar monitoring windows that cover the entire 360-degree field and whose targets are all located in the low-distortion center region.

[0012] As a preferred embodiment, the improved YOLOv8 deep learning network model described in step S3 introduces a C2f-Faster-EMA module into the backbone network. This module integrates a partially convolutional FasterNet Block to reduce computational redundancy and adapt to edge computing power; it also embeds an efficient multi-scale attention mechanism to enhance the ability to extract features from small-sized targets through cross-spatial learning. A C2f-Faster module is introduced into the neck network to achieve lightweighting of the feature fusion stage.

[0013] As a preferred embodiment, the pattern design of the feature target is as follows: a white circle is contained in the center of a black square background, and the diameter of the white circle is 2 / 3 of the side length of the square; the target detection and tracking module performs grayscale conversion, Gaussian filtering, Otsu adaptive threshold binarization and morphological opening operation on the extracted target ROI in sequence, and calculates the sub-pixel level coordinates of the target center through the image moment algorithm.

[0014] As a preferred embodiment, the calculation of the sub-pixel coordinates of the target center in step S4 specifically includes: Convert the region of interest image to grayscale and then apply Gaussian filtering; The optimal threshold is calculated using the Otsu algorithm, and the image after Gaussian filtering is binarized to separate the white circular area of ​​the target from the background. Morphological opening operations are performed on the binarized image to remove noise and fill holes, and the contour of the largest connected component is extracted. Calculate the zeroth moment of the contour of the largest connected region. and first moment , Through formula , Obtain the sub-pixel coordinates of the target center .

[0015] As a preferred solution, the rigid body displacement component generated by the panoramic camera due to the structural motion is eliminated by using the differential calculation formula. Specifically, this includes: calculating the difference between the measured displacement of the feature target at the outer node and the feature target at the inner node in the same direction, thereby mathematically eliminating the rigid body displacement component generated by the panoramic camera due to the structural motion.

[0016] As a preferred embodiment, step S5 specifically includes: Let the rigid body displacement of the panoramic camera as a function of the structure be... The actual elastic deformations of the feature targets at the inner and outer nodes are respectively... and ; Based on the virtual camera intrinsic parameters, the absolute measurement position of the feature target at the inner node observed by the panoramic camera is calculated. And the absolute displacement of the characteristic target at the outer node. ; By calculating the relative displacement difference In mathematics, the rigid body displacement term is eliminated. ,get This refers to the actual elastic deformation of the structure at the outer node relative to the inner node.

[0017] Furthermore, this invention also proposes a large-scale ring structure hoisting deformation monitoring system, which includes: The panoramic acquisition unit includes a panoramic camera rigidly fixed at the geometric center of the large-scale ring structure to be hoisted, used to acquire panoramic video streams in real time as the panoramic camera moves synchronously with the structure. Several feature targets are arranged on the inner and outer nodes of the same radial direction of the large-scale ring structure to be hoisted, in order to provide visual features; An edge computing terminal is electrically connected to the panoramic acquisition unit; the edge computing terminal integrates a graphics processor, which is used to run the above-mentioned large-scale ring structure hoisting deformation monitoring method based on panoramic vision and deep learning in real time and output the monitoring results. The display terminal is used to read the monitoring results output by the edge computing terminal and display them as deformation time history curves.

[0018] As a preferred embodiment, the edge computing terminal includes: The image projection module uses a dual cube projection method to establish a mapping relationship from the panoramic spherical coordinate system to the planar perspective projection, decomposing and reconstructing each frame of the panoramic video stream into several distortion-free planar monitoring windows covering the circumference of the structure. The target detection and tracking module, based on an improved YOLOv8 deep learning network model, automatically extracts the region of interest (ROI) containing the target. The displacement calculation module is used to perform sub-pixel positioning and relative displacement calculation, eliminate panoramic camera motion errors, and generate deformation curves.

[0019] Furthermore, the present invention also proposes an electronic device, which includes a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, it implements the above-mentioned method for monitoring the deformation of large-scale ring structure hoisting based on panoramic vision and deep learning.

[0020] Furthermore, the present invention also proposes a computer-readable storage medium storing at least one executable instruction, which, when executed on an electronic device, causes the electronic device to perform the aforementioned method for monitoring the deformation of a large-scale ring structure hoisting based on panoramic vision and deep learning.

[0021] The present invention achieves the following beneficial effects compared to the prior art: (1) A single panoramic camera replaces the traditional multi-camera array or a large number of contact sensors, greatly reducing hardware costs and on-site installation complexity, and realizing "single-machine full-field monitoring".

[0022] (2) Image edge distortion was eliminated by double cube projection; the problem of complex background interference and small target recognition at the construction site was solved by the improved YOLOv8 network, and the measurement accuracy within ±10mm was achieved by combining subpixel positioning technology.

[0023] (3) A relative displacement difference elimination algorithm was proposed, which cleverly eliminated the rigid body displacement error caused by the camera swaying with the hoisting structure, and the real deformation could be measured without setting an external static reference point.

[0024] (4) The algorithm has been optimized for edge computing and can output deformation curves and provide audible and visual alarms in real time on site, providing timely safety decision-making basis for hoisting command. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the on-site layout disclosed in the embodiments of the present invention.

[0026] Figure 2 This is a schematic diagram of an improved YOLOv8 network structure disclosed in an embodiment of the present invention.

[0027] Figure 3 Displaying detection graphs for YOLOv8n and improved YOLOv8n.

[0028] Figure 4 This is a schematic diagram of the target center identification process disclosed in the embodiments of the present invention.

[0029] Figure 5 This is a schematic diagram of the camera recognition principle disclosed in the embodiments of the present invention. Detailed Implementation

[0030] In the following description, numerous specific details are set forth in order to provide a more thorough understanding of the invention. However, it will be apparent to those skilled in the art that the invention can be practiced without one or more of these details. In other instances, certain technical features well-known in the art have not been described in order to avoid obscuring the invention.

[0031] The purpose of this invention is to provide a large-scale ring structure hoisting deformation monitoring system based on panoramic vision and deep learning, so as to solve the problems existing in the prior art, realize high-precision and efficient full-field deformation monitoring during the hoisting of large-scale ring structures, and ensure hoisting safety.

[0032] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0033] This invention provides a large-scale ring structure hoisting deformation monitoring system and method based on panoramic vision and deep learning. The system utilizes a single dual-fisheye panoramic camera as the front-end acquisition unit, coupled with an edge computing terminal equipped with a high-performance GPU, and employs an improved lightweight deep learning network to achieve real-time tracking and displacement calculation of multiple targets across the entire field.

[0034] I. Site Layout and Hardware Architecture like Figure 1 The diagram shown is a schematic representation of the layout of a large annular steel mesh hoisting site according to an embodiment of the present invention.

[0035] Overall scene ( Figure 1 a, Figure 1 b): At the site of large steel structure prefabrication, the annular steel mesh to be hoisted is laid flat on the ground. The monitoring system of this invention is installed at the geometric center of the structure and is hoisted as an integral part of the structure.

[0036] Data Acquisition and Computation Unit ( Figure 1 c): The panoramic acquisition unit uses a single dual-fisheye panoramic camera, fixed with a rigid bracket. To verify accuracy, multiple traditional real-time monitoring cameras can be deployed simultaneously on-site for comparison. The panoramic camera is connected to an edge computing terminal via a high-speed data interface. This terminal is equipped with a high-performance embedded GPU, capable of running deep learning models on-site and outputting data in real time.

[0037] Target components ( Figure 1 d): Several specially designed targets are placed at key nodes along the circumference of the structure. The targets are designed as high-contrast geometric patterns: a white circle is contained in the center of a black square background, and the diameter of the white circle is approximately 2 / 3 of the side length of the square. This design ensures that the target features can still be clearly identified by the algorithm under complex construction backgrounds (including tower cranes, scaffolding, strong light, etc.).

[0038] II. Improved YOLOv8 Deep Learning Network Construction To achieve high frame rate detection on resource-constrained edge devices and address the challenges of identifying small targets and dealing with background interference, this embodiment first constructs a system as follows: Figure 2 The improved YOLOv8 network structure shown serves as the core algorithm foundation for subsequent target tracking.

[0039] Overall architecture design: such as Figure 2 As shown above, the network mainly consists of three parts: the backbone network, the neck network, and the detection head.

[0040] Backbone network improvement: Responsible for feature extraction. This invention replaces the original C2f module with a C2f-Faster-EMA module, while retaining spatial pyramid pooling to enhance multi-scale features.

[0041] Improved neck network: Responsible for feature fusion. A path aggregation network structure was adopted, and the C2f module was replaced with a lighter C2f-Faster module. Deep and shallow feature fusion was achieved through upsampling and concatenation.

[0042] Detection head: Responsible for outputting detection results, it includes three detection heads of different scales, used to detect large, medium and small targets respectively.

[0043] Core module technical details: CFE module: such as Figure 2 As shown, this module is the core of the backbone network. It integrates the Faster-EMA Block.

[0044] Lightweight design: Spatial feature extraction is first performed using partial convolution. Partial convolution only convolves a portion of the input channels, leaving the remaining channels unchanged, thereby significantly reducing floating-point operations and memory access latency, and improving the inference speed at the edge.

[0045] Attention Enhancement: Introducing efficient multi-scale attention, this module can capture the dependencies between channels more precisely through feature grouping and cross-space learning, significantly enhancing the network's sensitivity to target geometric features and reducing the false negative rate.

[0046] The C2f-Faster module is primarily used for the neck region of the network. Its structure is similar to the Faster-EMA Block, but it removes the EMA attention mechanism, retaining only some convolutions and subsequent convolutional structures. This design minimizes the computational burden on the model while ensuring efficient feature fusion.

[0047] III. Target Center Identification and Tracking Process Based on the improved YOLOv8 network constructed above, this system adopts the following... Figure 3 The target center identification process shown accurately extracts displacement data from the panoramic video stream. This process consists of two main stages: 1. Coarse localization and tracking based on deep learning ( Figure 4 a): Panoramic projection: First, the dual fisheye images captured by the panoramic camera are decomposed into 8 distortion-free planar windows using dual cube projection technology.

[0048] Model inference: The decomposed image sequence is input into the trained improved YOLOv8 network in real time. For example... Figure 4 As shown in Figure a, despite the complex background of the original image (left side), the model can still accurately identify all targets in the image and output bounding boxes, as shown in Figure a. Figure 4 As shown on the right, each target is labeled "target" and accompanied by a confidence level.

[0049] Region of interest extraction: Based on the bounding box coordinates, automatically extract a local image containing the target to prepare for subsequent fine localization.

[0050] 2. Precise localization based on image processing ( Figure 4 b): Preprocessing and binarization: The extracted region of interest image is converted to grayscale and denoised using Gaussian filtering. Then, the Otsu adaptive thresholding algorithm is used for binarization to separate the white circular target from the black background and environmental noise, thus obtaining a binary mask.

[0051] Central computation: Morphological opening is performed on the binary image to remove noise, and edge detection is used to extract the contour of the largest connected component. Figure 4 (b) the green circle on the right, and use the image moment algorithm to calculate the sub-pixel centroid of the contour ( Figure 4 (b) The red dot on the right side serves as the final high-precision target center coordinates.

[0052] IV. Principles of Displacement Calculation and Error Elimination After obtaining the sub-pixel coordinates of the target center, the system performs physical displacement calculations. For example... Figure 5 The diagram shown is a schematic diagram illustrating the camera recognition and displacement calculation principle of this embodiment.

[0053] Coordinate system definition: and These represent the positions of the panoramic camera's optical center at the initial and current moments, respectively. (The target is also represented.) and target The initial position; and This represents the position of the target after it has moved. and target This represents the direction of the camera's line of sight.

[0054] Measurement principle and relative deformation calculation: Since the panoramic camera is mounted on the suspended structure, when the structure shakes, the optical center of the camera will undergo a rigid body displacement. The strategy for arranging measuring points on a ring structure is to set up two targets in the same direction, and then calculate three displacement values, such as... Figure 5 As shown. Among them, This indicates the displacement of target 1 relative to the position of the panoramic camera. This represents the displacement of target 2 relative to the position of the panoramic camera. This represents the relative displacement between target 1 and target 2. At this time, the measured displacement observed by the camera (e.g.) In reality, it is a true structural deformation. With camera rigid body displacement Superposition:

[0055]

[0056] In order to eliminate Due to the influence of the system, this system calculates the relative displacement difference between the two monitoring points. .(like Figure 5 (As shown).

[0057]

[0058] This differential calculation mathematically eliminates the common-mode interference caused by the camera's movement with the structure, thereby accurately obtaining the relative deformation within the structure.

[0059] V. Edge Computing and Real-Time Display All of the above algorithm modules are integrated and deployed on the edge computing terminal on site.

[0060] Real-time processing: Thanks to the lightweight design of PConv and C2f-Faster, the model inference time is greatly reduced, and the system can achieve a processing speed of more than 15 frames per second.

[0061] Dynamic early warning: The terminal is connected to a display screen to draw the time history curve of Δy in real time. When the detected deformation exceeds the preset safety threshold (e.g., ±20mm), the system immediately triggers an audible and visual alarm, realizing a real-time closed loop of "monitoring-analysis-early warning" and effectively ensuring the structural safety of the hoisting process.

[0062] The underlying logic of the large-scale ring structure hoisting deformation monitoring method disclosed in the above embodiments can be implemented entirely or partially through software, hardware, firmware, or other arbitrary combinations. When implemented using software, the above embodiments can be implemented entirely or partially in the form of a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions according to the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another.

[0063] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0064] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0065] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0066] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A method for monitoring deformation of large-scale ring structure hoisting based on panoramic vision and deep learning, characterized in that, Includes the following steps: S1. A panoramic camera is rigidly fixed at the geometric center of the large-scale ring structure to be hoisted, and feature targets are arranged on the inner and outer nodes in the same radial direction of the structure to acquire panoramic video streams in real time as the panoramic camera moves synchronously with the structure. S2. Using the dual cube projection method, a mapping relationship between the panoramic spherical coordinate system and the planar perspective projection is established, and each frame of the panoramic video stream is decomposed and reconstructed into several distortion-free planar monitoring windows covering the circumference of the structure. S3. Construct and train an improved YOLOv8 deep learning network model. Input the distortion-free planar monitoring window obtained in step S2 into the deep learning network model. Under the complex background of hoisting dynamics, identify the feature targets at the inner nodes and the feature targets at the outer nodes in real time, and output the region of interest (ROI) containing the feature targets. S4. Perform image preprocessing on the region of interest (ROI) obtained in step S3, extract the target contour using Otsu adaptive thresholding and morphological operations, and calculate the sub-pixel level pixel coordinates of the target center using the image moment algorithm. S5. Based on the virtual camera intrinsic parameters, the target pixel displacement is converted into physical space displacement, eliminating the rigid body displacement component generated by the panoramic camera with the structure movement, and calculating the actual elastic deformation of the structure at the outer node relative to the inner node.

2. The large-scale ring structure hoisting deformation monitoring method based on panoramic vision and deep learning according to claim 1, characterized in that, The dual cube projection method described in step S2 specifically includes: With 0°, 90°, 180°, and 270° as the main optical axes, the first layer of perspective projection is executed to generate the first set of four planar windows; Rotate 45° clockwise relative to the first projection, with 45°, 135°, 225°, and 315° as the main optical axis directions, and perform the second perspective projection to generate the second set of 4 planar windows; Combining the two sets of windows above forms a sequence of eight planar monitoring windows that cover the entire 360-degree field and whose targets are all located in the low-distortion center region. 3.The large-scale ring structure hoisting deformation monitoring method based on panoramic vision and deep learning according to claim 1, characterized in that, The improved YOLOv8 deep learning network model described in step S3 has the following improved features: In the backbone network, the original C2f module is replaced by a C2f-Faster-EMA module that integrates partial convolution and an efficient multi-scale attention mechanism. Partial convolution is used to reduce computational redundancy to adapt to edge computing terminals, and EMA is used to enhance the ability to extract features of small-sized targets. In the neck network, a C2f-Faster module with integrated partial convolutions is used to replace the original C2f module to achieve lightweight feature fusion. 4.The large-scale ring structure hoisting deformation monitoring method based on panoramic vision and deep learning of claim 1, wherein, The calculation of the sub-pixel coordinates of the target center in step S4 specifically includes: Convert the region of interest image to grayscale and then apply Gaussian filtering; The optimal threshold is calculated using the Otsu algorithm, and the image after Gaussian filtering is binarized to separate the white circular area of ​​the target from the background. Morphological opening operations are performed on the binarized image to remove noise and fill holes, and the contour of the largest connected component is extracted. computing the zeroth moment of the maximum connected component contour and the first moment , obtaining sub-pixel coordinates of the target center by the formula , .​ 5. The large-scale ring structure hoisting deformation monitoring method based on panoramic vision and deep learning according to claim 1, characterized in that, The rigid body displacement components caused by the panoramic camera's movement with the structure are eliminated using a differential calculation formula, specifically including: The difference between the measured displacements of the feature targets at the outer node and the inner node in the same direction is calculated to mathematically eliminate the rigid body displacement components generated by the panoramic camera as it moves with the structure.

6. The large-scale ring structure hoisting deformation monitoring method based on panoramic vision and deep learning according to claim 1 or 5, characterized in that, Step S5 specifically includes: The rigid body displacement of the panoramic camera generated with the structure motion is set as The structural true elastic deformations of the feature targets at the inner nodes and the feature targets at the outer nodes are and respectively; Based on the virtual camera intrinsic parameters, the absolute measurement positions of the feature targets at the inner side nodes observed by the panoramic camera are calculated , and the absolute measurement displacement of the feature targets at the outer side nodes ; By calculating the relative displacement difference Mathematically eliminate the rigid body displacement term , we get That is the true elastic deformation of the structure relative to the outside node to the inside node.

7. A large-scale ring structure hoisting deformation monitoring system characterized in that, include: The panoramic acquisition unit includes a panoramic camera rigidly fixed at the geometric center of the large-scale ring structure to be hoisted, used to acquire panoramic video streams in real time as the panoramic camera moves synchronously with the structure. Several feature targets are arranged on the inner and outer nodes of the same radial direction of the large-scale ring structure to be hoisted, in order to provide visual features; An edge computing terminal is electrically connected to the panoramic acquisition unit; the edge computing terminal integrates a graphics processor for running the large-scale ring structure hoisting deformation monitoring method based on panoramic vision and deep learning as described in any one of claims 1 to 6 in real time, and outputting monitoring results; The display terminal is used to read the monitoring results output by the edge computing terminal and display them as deformation time history curves.

8. The large-scale ring structure hoisting deformation monitoring system according to claim 7, characterized in that, The edge computing terminal includes: The image projection module uses a dual cube projection method to establish a mapping relationship from the panoramic spherical coordinate system to the planar perspective projection, decomposing and reconstructing each frame of the panoramic video stream into several distortion-free planar monitoring windows covering the circumference of the structure. The target detection and tracking module, based on an improved YOLOv8 deep learning network model, automatically extracts the region of interest (ROI) containing the target. The displacement calculation module is used to perform sub-pixel positioning and relative displacement calculation, eliminate panoramic camera motion errors, and generate deformation curves.

9. An electronic device, comprising: It includes a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, it implements the method for monitoring the deformation of a large-scale ring structure hoisting based on panoramic vision and deep learning as described in any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that, The storage medium stores at least one executable instruction, which, when executed on an electronic device, causes the electronic device to perform the large-scale ring structure hoisting deformation monitoring method based on panoramic vision and deep learning as described in any one of claims 1 to 6.

Images