A method and system for monitoring and protecting construction site wells based on YOLOv8 and Siamese Network

Abstract

This invention relates to the field of computer vision safety monitoring technology, and discloses a method and system for monitoring the protection of construction site manholes based on YOLOv8 and Siamese Network. The method includes: acquiring a sequence of target construction images at the construction site; performing detection using a YOLOv8 target detection network; extracting geometric feature vectors of the manhole protection structure; performing structural matching based on Siamese Network; and determining anomalies in the manhole protection structure. Compared to existing technologies that rely solely on target detection to determine whether a manhole is covered by fencing, this method addresses the technical problem of inaccurately identifying abnormal states of the manhole protection structure, especially when the fencing partially shifts but still covers the manhole. This application improves the accuracy of construction site manhole protection monitoring by introducing the spatial geometric relationship between the manhole and the fencing, combined with a structural consistency matching mechanism.

Description

Technical Field

[0001] This invention relates to the field of computer vision safety monitoring technology, and in particular to a method and system for monitoring the protection of construction site wells based on YOLOv8 and SiameseNetwork. Background Technology

[0002] Currently, to prevent accidents caused by personnel or equipment accidentally entering wells at construction sites, protective fences, warning signs, or temporary covers are typically installed around them. Meanwhile, to improve safety management at construction sites, some sites are gradually introducing video surveillance systems. These systems use cameras to monitor the construction area in real time and combine this with target detection algorithms to automatically identify hazardous areas. However, existing vision-based well safety detection technologies still have certain shortcomings.

[0003] For example, most existing technologies use traditional target detection methods to identify manholes or protective fences in construction images. These methods can only determine whether a manhole area exists or whether a fence has been detected, lacking the ability to analyze the spatial structural relationship between the manhole and the fence. When protective fences around manholes on construction sites experience slight shifts, tilts, or localized loosening due to collisions with construction vehicles, movement of construction personnel, or long-term use, the fence may still partially cover the manhole area. In such cases, traditional detection methods will still classify it as a normal state of "fence presence," making it difficult to identify potential safety hazards in a timely manner. Furthermore, the complex environment of construction sites, including changes in lighting, occlusion interference, and camera angle changes, can significantly impact traditional visual detection methods, easily leading to misjudgments or missed detections.

[0004] Therefore, there is an urgent need for a construction site well protection monitoring method that can still stably identify the spatial relationship of the well protection structure and determine abnormal conditions even in complex construction environments and when the well protection structure undergoes spatial displacement or structural changes, so as to improve the accuracy of well protection monitoring and safety early warning capabilities at construction sites. Summary of the Invention

[0005] To address the aforementioned technical shortcomings, the purpose of this invention is to propose a construction site well protection monitoring method based on YOLOv8 and SiameseNetwork. This method aims to solve the technical problem that existing monitoring methods, which rely solely on target detection to determine whether a well is covered by a fence, are unable to accurately identify abnormal states of the well protection structure, especially when the fence at the construction site has partially shifted but still partially covers the well.

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: The present invention provides a method for monitoring and protecting construction site wells based on YOLOv8 and Siamese Network.

[0007] The construction site well protection monitoring method based on YOLOv8 and Siamese Network includes:

[0008] Step S10: Obtain the target construction image sequence, and perform the well and fence structure analysis task based on the target construction image sequence using a collaborative mechanism of target detection and structural geometry extraction, and output the initial spatial structure parameter set;

[0009] Step S20: Based on the initial set of spatial structure parameters, the joint modeling mechanism of the geometric relationship between the well and the fence structure is used to perform the spatial relationship analysis task of the well protection structure, and output the geometric feature vector of the well protection structure;

[0010] Step S30: Based on the geometric feature vector of the well protection structure, a historical sample statistical modeling mechanism is used to execute the task of generating a standard template for the well protection structure, and output the template features of the well protection structure. template covariance matrix ;

[0011] Step S40: Based on the features of the well shaft protection structure template template covariance matrix A structural consistency matching mechanism based on Siamese Network is used to perform the similarity evaluation task of well hole protection structure and output the similarity score of well hole protection structure;

[0012] Step S50: Obtain the spatial offset of the well enclosure from the geometric feature vector of the well protection structure. Based on the spatial offset of the well enclosure The similarity score between the well and the protective structure is used to determine the abnormality of the well protection status using the structural offset confidence propagation judgment mechanism, and the monitoring results of the safety status of the well protection are output.

[0013] Preferably, step S10, which involves acquiring a target construction image sequence, performing a joint target detection and structural geometry extraction mechanism based on the target construction image sequence to analyze the well and fence structures, and outputting an initial set of spatial structure parameters, specifically includes:

[0014] Step S101: Obtain the target construction image sequence, and use the YOLOv8 target detection network to perform the detection task of the well hole area and the well hole protective fence structure based on the target construction image sequence, and output the well hole detection box set and the fence detection box set;

[0015] Step S102: Based on the well hole detection box set, use Python's OpenCV image processing library and NumPy numerical computing library to perform well hole geometric contour modeling task, and obtain well hole edge point set. Geometric center of the well ;

[0016] Step S103: Based on the fence detection box set, a structural boundary resolution mechanism based on fence corner point localization is used to perform the fence structure key point extraction task, and obtain the fence corner point set. With the geometric center of the fence ;

[0017] Step S104: Based on the set of points at the edge of the well. Geometric center of the well Fence corner point set With the geometric center of the fence Construct and output the initial set of spatial structure parameters.

[0018] Preferably, step S20, which involves performing the spatial relationship analysis task of the well protection structure based on the initial set of spatial structural parameters using a joint modeling mechanism for the geometric relationship between the well and the fence structure, and outputting the geometric feature vector of the well protection structure, specifically includes:

[0019] Step S201: Based on the geometric center of the well shaft With the geometric center of the fence The spatial offset of the well enclosure is calculated using a two-dimensional Euclidean distance method. ;

[0020] Step S202: Based on the geometric center of the well shaft With the geometric center of the fence The first angle of the fence structure is calculated using a polar coordinate angle calculation method based on the angular direction relationship. The angular features of the well enclosure at each corner point ;

[0021] Step S203: Adjust the spatial offset of the well enclosure. Angular features of the well enclosure Perform unified vector encoding and splicing fusion processing to output the geometric feature vector of the well protection structure.

[0022] Preferably, in step S202, based on the geometric center of the well... With the geometric center of the fence The first angle of the fence structure is calculated using a polar coordinate angle calculation method based on the angular direction relationship. The angular features of the well enclosure at each corner point The steps specifically include:

[0023] Step S2021: Using the geometric center of the well shaft As the origin of the polar coordinates, and with the geometric center of the well... Pointing to the geometric center of the fence The direction is used as the polar coordinate reference direction to establish the polar coordinate reference coordinate system of the well protection structure;

[0024] Step S2022: Obtain the fence structure The spatial positions of the corner points in the polar coordinate reference system are calculated. Each corner point is relative to the geometric center of the well. The direction deflection relationship vector;

[0025] Step S2023: Calculate the fence structure based on the direction deflection relationship vector. The polar angle values ​​of each corner point in the polar coordinate reference system are determined, and these polar angle values ​​are used to define the fence structure. The angle characteristics of the well enclosure corresponding to each corner point .

[0026] Preferably, in step S30, a historical sample statistical modeling mechanism is used to perform the task of generating a standard template for the well protection structure based on the geometric feature vector of the well protection structure, and the template features of the well protection structure are output. template covariance matrix The steps specifically include:

[0027] Step S301: Obtain images of well protection under normal protection status from historical construction monitoring data, and construct a standard training sample set of well protection structures based on the geometric feature vectors of well protection structures corresponding to the images of well protection.

[0028] Step S302: Based on the standard training sample set of well protection structures, statistical analysis is performed using the statistical mean modeling method to output the average structural features. Based on the average structural features, the feature vector centering modeling method is used to generate template features for well protection structures. Features of well protection structure template Used to characterize the typical spatial structural relationship of well protection structures under normal conditions;

[0029] Step S303: Based on the standard training sample set of well protection structures, the dispersion of well protection image samples is statistically analyzed using covariance statistical modeling to generate a template covariance matrix. Template covariance matrix This is used to characterize the range of variation of the geometric features of the well protection structure under normal conditions and the correlation between the features.

[0030] Preferably, in step S40, based on the features of the well shaft protection structure template... template covariance matrix The steps for performing the similarity evaluation task of well shaft protection structures using a Siamese Network-based structural consistency matching mechanism and outputting the similarity score of the well shaft protection structures specifically include:

[0031] Step S401: Pre-set a structural feature embedding Siamese network, which includes: an input feature fusion layer, used to perform unified feature fusion processing on the geometric feature vector of the well protection structure, the template features of the well protection structure, and the template covariance matrix to form a structural input feature representation; a feature mapping layer, used to perform nonlinear mapping processing on the structural input feature representation to extract the high-dimensional spatial feature representation of the well protection structure; a structural embedding encoding layer, used to encode the high-dimensional spatial feature representation into a structural embedding feature vector to form a deep structural expression of the well protection structure; a two-branch feature comparison layer, used to perform structural feature comparison processing between two shared weight branches of the structural feature embedding Siamese network to calculate the feature distance relationship between the structural features to be detected and the structural template features; and a similarity evaluation output layer, used to calculate the similarity score of the well protection structure based on the feature distance relationship.

[0032] Step S402: Obtain the historical training input dataset and the historical training output dataset. The historical training input dataset includes the historical geometric feature vector of the well protection structure, the historical template features of the well protection structure, and the historical template covariance matrix. Based on the historical training input dataset and the historical training output dataset, perform a pre-training process on the Siamese network by combining the contrastive loss function and the sample pair distance minimization training method.

[0033] Step S403: Combine the geometric feature vector of the well protection structure and the template features of the well protection structure. template covariance matrix Input a pre-trained structural feature embedding twin network, and output a similarity score for the well protection structure.

[0034] Preferably, in step S50, the spatial offset of the well enclosure is obtained from the geometric feature vector of the well protection structure. Based on the spatial offset of the well enclosure The steps for determining anomalies in well protection status and outputting well protection safety status monitoring results using a structural offset confidence propagation mechanism based on the similarity score of the well protection structure include:

[0035] Step S501: Obtain the spatial offset of the well enclosure from the geometric feature vector of the well protection structure. Spatial offset of well enclosure Used to characterize the degree of spatial position change of the current fence structure relative to the standard well protection structure;

[0036] Step S502: Based on the spatial offset of the well enclosure The similarity score of the well protection structure is corrected by confidence propagation based on covariance constraint, and the offset confidence index of the well protection structure is output.

[0037] Step S503: When the offset confidence index of the well shaft protection structure is greater than the preset structural offset anomaly threshold, it is determined that the well shaft protection structure is in an abnormal protection state; when the offset confidence index of the well shaft protection structure is less than or equal to the preset structural offset anomaly threshold, it is determined that the well shaft protection structure is in a normal protection state; finally, the well shaft protection safety status monitoring result is output based on the judgment result.

[0038] This invention also provides a construction site well protection monitoring system based on YOLOv8 and Siamese Network, comprising:

[0039] The well and fence structure analysis module is used to acquire the target construction image sequence. Based on the target construction image sequence, it uses a collaborative mechanism of target detection and structural geometry extraction to perform the well and fence structure analysis task and output the initial spatial structure parameter set.

[0040] The spatial structure relationship modeling module is used to perform the spatial relationship analysis task of the well protection structure based on the initial set of spatial structure parameters and the joint modeling mechanism of the geometric relationship between the well and the fence structure, and outputs the geometric feature vector of the well protection structure.

[0041] The structural template generation module is used to perform the task of generating standard templates for well and tunnel protection structures based on the geometric feature vectors of the well and tunnel protection structures and employing a historical sample statistical modeling mechanism, outputting the template features of the well and tunnel protection structures. template covariance matrix ;

[0042] The structural consistency matching module is used to match the features of the well shaft protection structure template. template covariance matrix A structural consistency matching mechanism based on Siamese Network is used to perform the similarity evaluation task of well hole protection structure and output the similarity score of well hole protection structure;

[0043] The protection status determination module is used to obtain the spatial offset of the well enclosure from the geometric feature vector of the well protection structure. Based on the spatial offset of the well enclosure The similarity score between the well and the protective structure is used to determine the abnormality of the well protection status using the structural offset confidence propagation judgment mechanism, and the monitoring results of the safety status of the well protection are output.

[0044] The present invention also provides a construction site well protection monitoring device based on YOLOv8 and Siamese Network, comprising: a memory, a processor, and a construction site well protection monitoring program based on YOLOv8 and Siamese Network stored in the memory and executable on the processor. When the construction site well protection monitoring program based on YOLOv8 and Siamese Network is executed by the processor, a construction site well protection monitoring method based on YOLOv8 and Siamese Network is implemented.

[0045] The present invention also provides a computer program product, including a construction site well protection monitoring program based on YOLOv8 and Siamese Network. When the construction site well protection monitoring program based on YOLOv8 and Siamese Network is executed by a processor, it implements the construction site well protection monitoring method based on YOLOv8 and Siamese Network.

[0046] The beneficial effects of this invention are as follows: By introducing the YOLOv8 target detection network and the geometric relationship modeling mechanism of the well-hole-fence structure, this invention jointly analyzes the well-hole area and the protective fence structure in the construction site image, and further constructs the spatial geometric relationship feature vector between the well-hole and the fence. This not only identifies whether the well-hole is covered by the fence, but also analyzes the changes in the spatial structural relationship between the well-hole and the fence. Compared with the traditional monitoring method that only relies on target detection to determine whether the well-hole is enclosed, this invention can effectively identify hidden anomalies such as fence tilting, loosening or overall displacement, thus improving the accuracy and reliability of well-hole protection monitoring.

[0047] This invention constructs standard template features and template covariance matrix for well protection structures, and combines them with SiameseNetwork for structural consistency matching. At the same time, it introduces a structural offset confidence propagation judgment mechanism to jointly evaluate the spatial offset of well enclosures and structural similarity scores, thereby realizing intelligent judgment of the stability of well protection structures. This avoids the false alarm or missed alarm problems caused by traditional judgments based on fixed thresholds or simple rules, and improves the stability and anomaly identification capability of well protection safety status monitoring. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. 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.

[0049] Figure 1 This is a flowchart illustrating the first embodiment of a construction site well protection monitoring method based on YOLOv8 and Siamese Network according to the present invention.

[0050] Figure 2 This is a geometric analytical diagram of the well and fence structure of a first embodiment of a construction site well protection and monitoring method based on YOLOv8 and Siamese Network according to the present invention.

[0051] Figure 3 This is a schematic diagram illustrating the joint modeling of the geometric relationship between the well and the fence structure in the first embodiment of the construction site well protection and monitoring method based on YOLOv8 and Siamese Network of the present invention.

[0052] Figure 4 This is a schematic diagram of statistical modeling of a well protection structure template, representing the first embodiment of a construction site well protection monitoring method based on YOLOv8 and Siamese Network according to the present invention.

[0053] Figure 5 This diagram illustrates the structural offset confidence propagation anomaly determination in the first embodiment of a construction site well protection monitoring method based on YOLOv8 and Siamese Network according to the present invention.

[0054] Figure 6 This is a schematic diagram of the equipment for a construction site well protection monitoring method based on YOLOv8 and Siamese Network according to the present invention. Detailed Implementation

[0055] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] Example 1: As Figure 1 The diagram shown is a flowchart of the first embodiment of the construction site well protection and monitoring method based on YOLOv8 and Siamese Network of the present invention. The first embodiment of the construction site well protection and monitoring method based on YOLOv8 and Siamese Network of the present invention is presented.

[0057] In the first embodiment, the construction site well protection monitoring method based on YOLOv8 and Siamese Network includes:

[0058] Step S10: Obtain the target construction image sequence, and perform the well and fence structure analysis task based on the target construction image sequence using a collaborative mechanism of target detection and structural geometry extraction, and output the initial spatial structure parameter set;

[0059] It should be noted that the collaborative mechanism of target detection and structural geometry extraction refers to a joint processing mechanism that, based on the identification of manholes and fences in construction site images, further extracts spatial geometric features and analyzes structural parameters of the identified target structures. This mechanism includes multiple processing steps such as target detection, structural contour extraction, key point localization, and spatial structural parameter organization. Specifically, target detection primarily uses a pre-trained YOLOv8 target detection network to identify manhole areas and manhole protective fence structures in construction images, obtaining sets of manhole detection boxes and fence detection boxes. Structural contour extraction is used to analyze the edge contours of manholes within the detection box areas, thereby obtaining the set of manhole edge points and the geometric center of the manhole. Key point localization is used to identify and locate the corner points of the fence structure, obtaining the set of fence corner points and the geometric center of the fence. Spatial structural parameter organization is used to uniformly and structurally describe the information such as the geometric center of the manhole, the set of manhole edge points, the set of fence corner points, and the geometric center of the fence, forming an initial set of spatial structural parameters, thus providing basic data for subsequent modeling of the spatial relationship between the manhole and fence structures.

[0060] Understandably, by introducing a structural geometry extraction collaborative mechanism in the target detection stage, it is possible not only to identify the presence of manhole areas and manhole protective fences in construction images, but also to further obtain key spatial structural information between the manhole and the fence structure, such as the center position of the manhole, the distribution of corner points of the fence structure, and the outline shape of the manhole edge. This allows for a structural characterization of the spatial relationship between the manhole and the fence. In this way, the traditional coarse judgment based solely on the location of the detection frame can be upgraded to a refined analysis based on structural geometric relationships. This provides a reliable data foundation for subsequently constructing geometric feature vectors of the manhole protection structure and conducting structural consistency matching, thereby improving the stability and accuracy of manhole protection monitoring.

[0061] It should be understood that, compared to traditional construction safety monitoring methods that typically rely solely on target detection models to determine whether a well area is covered by a fence, this step, by extracting geometric features and analyzing spatial structural parameters of the well and fence structures, can identify changes in the spatial layout of the fence structure. For example, when the fence as a whole undergoes slight translation, rotation, or partial tilting, traditional detection methods, relying only on detection frame information, often still classify it as a normal protective state. However, this implementation method, by introducing structural information such as well edge point sets and fence corner point sets, can more accurately reflect the true spatial relationship between the well and the fence, thus providing more reliable data support for subsequent anomaly identification.

[0062] For example, such as Figure 2 As shown, the circular area represents the manhole area at the construction site, with the center of the circle indicating the geometric center of the manhole. The rectangular structures represent the protective fences surrounding the manholes; the dashed rectangles indicate the fence's position under standard protection conditions, while the solid rectangles indicate the currently detected fence structure position. When the construction site image is input, the YOLOv8 object detection network first identifies the detection bounding box of the manhole area and the fence structure. Based on this, the edge contour of the manhole is further analyzed to obtain the manhole edge point set and calculate the manhole's geometric center. Simultaneously, by locating corner points and analyzing the fence structure, key corner point information is extracted, and the fence's geometric center is calculated based on the fence corner point set. Subsequently, the spatial offset of the manhole-fence is calculated based on the spatial relationship between the manhole's geometric center and the fence's geometric center, as shown by the arrows in the figure. This offset reflects the degree of change in the fence structure's position relative to the manhole. Under standard protection conditions, the fence structure should be symmetrically distributed around the center of the manhole, and the distance between the geometric center of the manhole and the geometric center of the fence is usually small and stable. However, when the fence structure is subjected to collisions with construction vehicles, movement of construction personnel, or loosening due to long-term use, the fence may shift in a certain direction, at which point the spatial distance between the center of the fence and the center of the manhole will change. The set of manhole edge points, fence corner points, and corresponding geometric center information extracted in step S10 can intuitively reflect the spatial structural relationship between the manhole and the fence. This structural information is organized into an initial set of spatial structural parameters, providing basic data for subsequent steps such as constructing the geometric feature vector of the manhole protection structure, modeling the structural template, and matching structural consistency.

[0063] Step S20: Based on the initial set of spatial structure parameters, the joint modeling mechanism of the geometric relationship between the well and the fence structure is used to perform the spatial relationship analysis task of the well protection structure, and output the geometric feature vector of the well protection structure;

[0064] It should be noted that the joint modeling mechanism for the geometric relationship between the manhole and the fence structure refers to a spatial relationship analysis method that, after obtaining the initial set of spatial structural parameters output in step S10, jointly models the spatial positional relationship, angular relationship, and structural distribution relationship between the manhole and the fence. This includes processing steps such as spatial offset relationship modeling, angular direction relationship modeling, and structural feature vector construction. Specifically, spatial offset relationship modeling mainly calculates the spatial offset of the manhole and fence based on the spatial distance between their geometric centers, representing the translational change of the fence structure relative to the manhole center. Angular direction relationship modeling mainly calculates the angular distribution characteristics of each corner point of the fence relative to the manhole center based on the directional relationship between the manhole's geometric center and the fence's corner points, reflecting whether the fence structure has rotated or undergone structural morphological changes. Structural feature vector construction is used to uniformly encode the aforementioned spatial offsets and angular features, thereby forming a geometric feature vector of the manhole protection structure that comprehensively describes the spatial relationship of the manhole protection structure, providing a unified feature representation for subsequent manhole protection structure template modeling and structural consistency matching.

[0065] Understandably, by jointly modeling the spatial geometric relationship between the manhole and the fence, the structural information in the original image can be converted into structural feature vectors with clear spatial meaning, thus enabling the spatial layout characteristics of the manhole protection structure to be characterized at the geometric level. For example, by analyzing the spatial offset between the center of the manhole and the center of the fence, it can be reflected whether the fence has moved as a whole; by analyzing the angular distribution characteristics between the center of the manhole and the corner points of the fence, it can be reflected whether the fence structure has rotated or deformed. By uniformly encoding this structural relationship information into a geometric feature vector of the manhole protection structure, a stable and comparable feature representation can be provided for subsequent structural template generation and structural consistency matching, thereby improving the stability and accuracy of manhole protection status identification.

[0066] It should be understood that, compared to traditional construction safety monitoring methods that typically determine whether a well opening is covered by a fence simply by checking the position of the detection frame, this step, by constructing a spatial geometric relationship model between the well opening and the fence, can identify the spatial change trend of the fence structure. For example, when the fence structure as a whole shifts but still covers the well opening area, traditional detection methods would still classify it as a normal protection state. However, this implementation method, by calculating the spatial offset of the well opening fence and the directional distribution of the fence corner points, can more accurately reflect the spatial positional changes of the fence structure, thus providing more reliable data for subsequent identification of anomalies in the well opening protection structure.

[0067] For example, such as Figure 3As shown, the circular area represents the well area of ​​the construction site, and the center of the circle represents the geometric center of the well. Four corner points of the fence surrounding the well together constitute the well protection fence structure, and the four corner points are connected sequentially to form the outline of the fence structure. The figure uses dashed lines of different colors to represent the spatial directional relationships between the geometric center of the well and each corner point of the fence. These directional relationships are used to characterize the angular distribution characteristics between the center of the well and the fence structure. Simultaneously, the arrows in the figure indicate the spatial offset direction between the geometric center of the well and the geometric center of the fence. This spatial offset reflects the overall translational change of the fence structure relative to the position of the well. In the actual processing, firstly, the geometric center of the well and the set of fence corner points are obtained based on the initial spatial structure parameter set obtained in step S10. Then, using the geometric center of the well as a reference point, the directional relationships between the center of the well and each corner point of the fence are calculated, thus obtaining the angular characteristic distribution of the fence corner points relative to the center of the well. At the same time, the geometric center of the fence structure is calculated based on the set of fence corner points, and the spatial offset relationship between the geometric center of the well and the geometric center of the fence is further calculated. Through the above process, the spatial offset of the manhole enclosure and the angular distribution characteristics of the enclosure corner points can be obtained simultaneously, thus forming geometric feature information that comprehensively describes the spatial structural relationship between the manhole and the enclosure. In this way, this step can transform the structural information in the original construction images into structural relationship features with clear spatial meaning, thereby constructing a geometric feature vector for the manhole protection structure. For example, when the enclosure is in a normal state, the directional distribution of the four corner points relative to the center of the manhole is usually relatively stable, and the spatial distance between the geometric center of the enclosure and the geometric center of the manhole remains within a reasonable range; however, when the enclosure is affected by external forces and moves or rotates, the directional distribution of the enclosure corner points and the position of the enclosure center will change. By jointly modeling these spatial relationships, the spatial layout state of the manhole protection structure can be accurately reflected, thus providing a reliable feature basis for subsequent manhole protection structure template construction, structural consistency matching, and abnormal state determination.

[0068] Step S30: Based on the geometric feature vector of the well protection structure, a historical sample statistical modeling mechanism is used to execute the task of generating a standard template for the well protection structure, and output the template features of the well protection structure. template covariance matrix ;

[0069] It should be noted that the historical sample statistical modeling mechanism refers to a data modeling mechanism that, based on the current geometric feature vector of the well protection structure, introduces well protection samples in normal protection status from historical construction monitoring data to statistically analyze and build a model of the spatial geometric features of the well protection structure. This mechanism includes historical sample screening, structural feature statistical modeling, and structural distribution feature analysis. Specifically, historical sample screening is used to select image samples of the well protection structure in normal status from long-term monitoring data of the construction site, and extracts the corresponding geometric feature vectors of the well protection structure according to the processing flow of steps S10 and S20, thereby constructing a standard training sample set for the well protection structure. Structural feature statistical modeling is used to statistically analyze the geometric feature vectors in the training sample set, calculate the average structural features of the well protection structure in normal status, and form the template features of the well protection structure. Structural distribution feature analysis constructs a template covariance matrix by calculating the statistical dispersion between geometric feature vectors, which describes the spatial variation range of the well protection structure in normal status and the correlation between different structural features. Through the above methods, a standard structural template reflecting the spatial distribution law of the well protection structure in normal status can be constructed.

[0070] Understandably, by introducing a historical sample statistical modeling mechanism, a large amount of data on manhole enclosure structures under normal protection conditions at construction sites can be statistically analyzed in a unified manner, thereby constructing a standard structural template that can represent the spatial relationship of normal manhole protection structures. This template not only includes the typical geometric features of manhole protection structures but also reflects the changing trends and fluctuation ranges between different structural features through the template covariance matrix. Therefore, when conducting a consistency assessment of manhole protection structures in the future, the currently detected structural features can be compared with this standard template to determine whether the current manhole protection structure deviates from the normal structural distribution state. In this way, a certain tolerance for structural changes is achieved, thereby avoiding misjudgment problems caused by factors such as changes in camera angle, image noise, or detection errors. For example, during long-term monitoring of a construction site, a large number of image samples of manhole protection fences in a normal state can be selected from continuously acquired construction images, such as acquiring hundreds or thousands of images of manhole protection in a normal state, and extracting the corresponding geometric feature vectors of the manhole protection structure through steps S10 and S20. By statistically analyzing these geometric feature vectors, the average structural characteristics of the manhole protection structure under normal conditions can be calculated, such as the typical spatial distance between the manhole center and the fence center, and the typical angular distribution of the fence corner points relative to the manhole center. Simultaneously, by calculating the statistical dispersion among these features, a template covariance matrix can be obtained, which describes the reasonable range of variation allowed for the manhole protection structure under normal conditions. For example, in actual construction environments, the positions of fence corner points may fluctuate slightly due to factors such as camera position, lighting conditions, or image detection errors. The template covariance matrix can reflect this normal fluctuation range, thus distinguishing between normal and abnormal variations during structural consistency assessments, improving the accuracy and stability of manhole protection monitoring.

[0071] For example, such as Figure 4As shown in the figure, the horizontal axis represents the spatial offset characteristics of the manhole enclosure, reflecting the degree of spatial offset between the geometric center of the enclosure and the geometric center of the manhole; the vertical axis represents the angular distribution characteristics of the enclosure structure, reflecting the overall structural orientation distribution of each corner point of the enclosure relative to the center of the manhole. The blue scatter points in the figure represent multiple geometric feature samples of the manhole protection structure extracted from historical monitoring data of the construction site. These samples all come from construction scenarios where the manhole protection structure is in a normal protection state. Statistical analysis of these historical samples shows that the geometric features of the manhole protection structure under normal conditions are usually distributed within a certain range and exhibit a relatively stable distribution pattern. The orange dots located at the center of the sample distribution represent the structural template features obtained from historical sample statistics. These template features reflect the typical spatial structural relationships of the manhole protection structure under normal conditions, such as the typical offset distance between the manhole center and the enclosure center, and the average directional distribution of the enclosure structure corner points. The dashed elliptical area drawn around the template features represents the structural variation range corresponding to the template covariance matrix obtained from historical sample statistics. This elliptical area reflects the reasonable fluctuation range that the manhole protection structure may experience under normal construction conditions. For example, due to factors such as camera installation location, shooting angle, changes in lighting, or detection algorithm errors, even if the fence structure remains in a normal state, its spatial offset characteristics or angular distribution characteristics may fluctuate to a certain extent. The template covariance matrix can statistically describe this normal fluctuation, thereby forming the normal distribution area of ​​the well hole protection structure.

[0072] Step S40: Based on the features of the well shaft protection structure template template covariance matrix A structural consistency matching mechanism based on Siamese Network is used to perform the similarity evaluation task of well hole protection structure and output the similarity score of well hole protection structure;

[0073] It should be noted that the Siamese Network-based structural consistency matching mechanism is a structural consistency assessment method that utilizes Siamese networks to perform deep feature matching on the similarity between the currently detected features of the well protection structure and the features of the standard structural template. This mechanism constructs a dual-branch neural network with shared parameters. The geometric feature vector of the current well protection structure and the template features are input into two structurally consistent network branches for feature embedding representation learning. Inside the network, high-dimensional structural feature representations of the well protection structure are extracted through multi-layer nonlinear mapping. At the output layer, distance or similarity calculations are performed on the embedded features obtained from the two branches to obtain the well protection structure similarity score. In this process, the template covariance matrix obtained in step S30 can be used to normalize or adjust the weights of the input structural features, ensuring that the calculation of structural similarity fully considers the variation range of different structural feature dimensions under normal conditions, thereby improving the stability of the structural consistency assessment.

[0074] Understandably, by introducing a Siamese Network-based structural consistency matching mechanism, the currently detected features of the manhole protection structure can be compared with historically statistically obtained standard structural templates at the deep feature level, thereby determining whether the current manhole protection structure is consistent with the normal structural pattern. Compared to judging whether a structure is abnormal solely through simple distance calculation or fixed rules, this method can learn the similarity relationships between manhole protection structures in a high-dimensional feature space, enabling more accurate identification of subtle changes in the manhole protection structure, such as slight offsets, local deformations, or angular changes in the fence structure, thus improving the identification and adaptability of manhole protection monitoring. Compared to traditional construction safety monitoring methods that typically use fixed thresholds or manual rules to judge the spatial distance between the manhole and the fence, this step, by introducing a Siamese network structure, can learn the spatial structural features of the manhole protection structure under normal conditions through a data-driven approach and identify anomalies based on the similarity relationships between structural features. In this way, even under conditions of changes in lighting, camera angle, or errors in the detection frame at the construction site, the true changes in the manhole protection structure can still be identified through deep feature matching, thereby reducing false positives or false negatives and improving the reliability of manhole protection monitoring. For example, in a construction site, the geometric feature vector of the current manhole protection structure is first obtained according to step S20, and the template feature of the manhole protection structure constructed in step S30 is used as a standard reference structure. Then, the current structural features and the template features are respectively input into two branches of a Siamese network for feature extraction and embedding representation. When the fence structure is in a normal state, the two sets of structural features have high consistency in the feature space output by the network, resulting in a high calculated structural similarity score. However, when the fence shifts due to a collision with a construction vehicle or part of the fence is removed, the spatial relationship between the manhole and the fence changes significantly, and the structural features extracted by the Siamese network also differ, leading to a significant decrease in the calculated structural similarity score. By analyzing this similarity score, it is possible to determine whether there are abnormal changes in the manhole protection structure, providing a reliable basis for subsequent assessment of the manhole protection safety status.

[0075] Step S50: Obtain the spatial offset of the well enclosure from the geometric feature vector of the well protection structure. Based on the spatial offset of the well enclosure The similarity score between the well and the protective structure is used to determine the abnormality of the well protection status using the structural offset confidence propagation judgment mechanism, and the monitoring results of the safety status of the well protection are output.

[0076] It should be noted that the structural offset confidence propagation judgment mechanism is a judgment method that comprehensively evaluates the abnormal state of a well hole protection structure based on obtaining the spatial offset of the well hole fence and the similarity score of the well hole protection structure. The spatial offset of the well hole fence characterizes the degree of spatial offset of the geometric center of the fence structure relative to the center of the well hole, and the similarity score of the well hole protection structure characterizes the degree of consistency between the overall shape of the current well hole protection structure and the historical standard structural template. Specifically, firstly, the similarity score of the well hole protection structure is corrected using a confidence propagation correction method based on covariance constraints according to the spatial offset of the well hole fence to obtain an offset confidence index. Based on this, the offset confidence index is further fused and modeled in conjunction with the similarity score of the well hole protection structure to obtain a comprehensive risk confidence score that characterizes the overall abnormal risk level of the well hole protection structure. Finally, the comprehensive risk confidence score is compared with a preset structural offset anomaly threshold. When the comprehensive risk confidence score is greater than the structural offset anomaly threshold, the well shaft protection structure is determined to be in an abnormal protection state; when the comprehensive risk confidence score is less than or equal to the structural offset anomaly threshold, the well shaft protection structure is determined to be in a normal protection state, and the well shaft protection safety status monitoring result is output. It should be noted that the offset confidence score is an intermediate variable in the confidence propagation process and is not directly used as the final judgment basis.

[0077] Understandably, by introducing a structural offset confidence propagation mechanism, the spatial position change information and structural morphology change information of the manhole enclosure can be comprehensively analyzed to achieve a multi-dimensional assessment of the manhole protection status. For example, when the enclosure structure remains intact but undergoes a small-scale translation, although the structural similarity score is still high, the spatial offset may increase significantly; while when the enclosure structure is partially missing or tilted, the structural similarity score will decrease significantly.

[0078] It should be understood that, compared to traditional construction safety monitoring methods that typically determine whether a well is in a safe protection state based solely on simple distance thresholds or fence detection results, this implementation method, by introducing a structural offset confidence propagation mechanism, can comprehensively analyze the well protection structure in terms of both spatial location changes and structural morphological changes. When the fence structure exhibits slight displacement but the overall structure remains intact, the abnormal risk can be appropriately suppressed based on the structural similarity score. However, when the fence structure undergoes significant displacement accompanied by structural morphological changes, both the spatial offset and the structural similarity score will simultaneously reflect the abnormal state. The confidence propagation mechanism can rapidly improve the confidence level of the abnormal risk, thereby promptly identifying anomalies in the well protection structure.

[0079] For example, such as Figure 5As shown in the figure, the horizontal axis represents the spatial offset of the manhole enclosure, which characterizes the degree of change in the spatial position of the geometric center of the enclosure structure relative to the center of the manhole; the vertical axis represents the similarity score of the manhole protection structure, which reflects the degree of consistency between the overall shape of the current manhole protection structure and the standard structural template. The scatter points in the figure represent the distribution of the manhole protection structure status obtained during continuous monitoring. By analyzing the structural features extracted at different monitoring times, the overall distribution relationship of the manhole protection structure in terms of spatial offset features and structural similarity features can be formed. The area boundary shown by the dashed line in the figure represents the risk judgment boundary formed in the structural offset confidence propagation model, which is used to distinguish whether the manhole protection structure is in a normal state area or an abnormal risk area. It should be noted that the area boundary is not based on a fixed threshold division result of a single spatial offset or structural similarity score, but is a judgment boundary adaptively formed by the structural offset confidence propagation fusion model combined with historical statistical features. Under normal circumstances, when the manhole enclosure structure at the construction site remains stable and the enclosure shape is intact, its spatial offset is usually small, and the overall structural similarity score is at a high level. Therefore, the corresponding structural state points tend to be concentrated near the normal structure area in the feature space. However, in practical applications, the anomaly determination of well protection structures is not solely determined by the numerical values ​​of spatial offset and structural similarity score. Instead, it is based on the comprehensive risk confidence score calculated by the structural offset confidence propagation fusion model. Therefore, even if certain structural state points have small spatial offsets and high structural similarity scores, they may still be judged as abnormal states if they deviate from the normal structural pattern under historical statistical distribution or covariance constraints, or if their comprehensive risk confidence score exceeds a preset threshold. This approach allows for more accurate identification of potential anomaly risks in well protection structures, improving the robustness and reliability of the monitoring system. Furthermore, during actual system operation, the anomaly determination threshold and model parameters can be dynamically adjusted according to different construction scenarios or safety strategies. Therefore, when the threshold is tightened or the determination strategy is enhanced, even structural states with small spatial offsets and high structural similarity scores may be judged as abnormal states if their comprehensive risk confidence score exceeds the current threshold. When the fence structure experiences significant displacement, partial fence loss, or tilting, the spatial offset of the fence's geometric center relative to the well center gradually increases, while the structural similarity score decreases, causing the structural state point to gradually move towards the abnormal risk area. This state change trend is comprehensively analyzed using a structural offset confidence propagation mechanism to calculate the overall risk confidence of the well protection structure. This process not only identifies whether there are obvious structural anomalies in the well protection structure but also allows for the early identification of potential abnormal risks based on the distribution trend of the structural state in the feature space.For example, when the fence structure gradually approaches the abnormal area but has not yet fully entered the abnormal area, the risk growth trend can be identified through the confidence propagation mechanism, thereby issuing a safety warning in advance and improving the reliability and timeliness of well protection safety monitoring.

[0080] Example 2: Furthermore, the present invention provides a construction site well protection monitoring system based on YOLOv8 and Siamese Network, employing a construction site well protection monitoring method based on YOLOv8 and Siamese Network from the above embodiments, which can solve the technical problem of construction site well protection monitoring based on YOLOv8 and Siamese Network. The beneficial effects of the construction site well protection monitoring system based on YOLOv8 and Siamese Network provided by the present invention are the same as those of the construction site well protection monitoring method based on YOLOv8 and Siamese Network provided in the above embodiments, and other technical features of the construction site well protection monitoring system based on YOLOv8 and Siamese Network are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0081] Example 3: This invention provides a construction site well protection monitoring device based on YOLOv8 and Siamese Network. Please refer to... Figure 6A construction site well protection and monitoring device based on YOLOv8 and Siamese Network includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform the construction site well protection and monitoring method based on YOLOv8 and Siamese Network described in Embodiment 1 above. The construction site well protection and monitoring device based on YOLOv8 and Siamese Network in this embodiment of the invention may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), vehicle terminals (e.g., vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. This construction site well protection and monitoring device based on YOLOv8 and Siamese Network is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the invention. A construction site well protection and monitoring device based on YOLOv8 and Siamese Network may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in read-only memory 1002 or a program loaded from storage device 1003 into random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the YOLOv8 and Siamese Network-based construction site well protection and monitoring device. The processing unit 1001, read-only memory 1002, and random access memory 1004 are interconnected via a bus 1005. An I / O interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows a YOLOv8 and Siamese Network-based construction site well protection monitoring device to wirelessly or wiredly communicate with other devices to exchange data.While the diagram illustrates a YOLOv8 and Siamese Network-based site well protection monitoring system with various systems, it should be understood that implementation of all the systems shown is not required. More or fewer systems may be implemented alternatively.

[0082] Example 4: This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the construction site well protection monitoring method based on YOLOv8 and Siamese Network described above. The computer program product provided by this invention can solve the technical problem of construction site well protection monitoring based on YOLOv8 and Siamese Network. Compared with the prior art, the beneficial effects of the computer program product provided by this invention are the same as those of the construction site well protection monitoring method based on YOLOv8 and Siamese Network provided in the above embodiments, and will not be repeated here.

[0083] In particular, according to the embodiments disclosed in this invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this invention.

[0084] It should be understood that the various parts disclosed in this invention can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics may be combined in any suitable manner in one or more embodiments or examples.

[0085] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for monitoring and protecting construction site wells and shafts based on YOLOv8 and Siamese Network, characterized in that, The methods include: Step S10: Obtain the target construction image sequence, and perform the well and fence structure analysis task based on the target construction image sequence using a collaborative mechanism of target detection and structural geometry extraction, and output the initial spatial structure parameter set; Step S20: Based on the initial set of spatial structure parameters, the joint modeling mechanism of the geometric relationship between the well and the fence structure is adopted to perform the spatial relationship analysis task of the well protection structure, and output the geometric feature vector of the well protection structure; Step S30: Based on the geometric feature vector of the well protection structure, a historical sample statistical modeling mechanism is used to execute the task of generating a standard template for the well protection structure, and output the template features of the well protection structure. template covariance matrix ; Step S40: Based on the features of the well shaft protection structure template template covariance matrix A structural consistency matching mechanism based on SiameseNetwork is used to perform the similarity evaluation task of well hole protection structure, and the similarity score of well hole protection structure is output. Step S50: Obtain the spatial offset of the well enclosure from the geometric feature vector of the well protection structure. Based on the spatial offset of the well enclosure The similarity score between the well and the protective structure is used to determine the abnormality of the well protection status using the structural offset confidence propagation judgment mechanism, and the monitoring results of the safety status of the well protection are output.

2. The construction site well protection monitoring method based on YOLOv8 and Siamese Network as described in claim 1, characterized in that, Step S10 involves acquiring the target construction image sequence, and then using a collaborative mechanism of target detection and structural geometry extraction to perform the analysis of the well and fence structures, outputting the initial set of spatial structure parameters. This step specifically includes: Step S101: Obtain the target construction image sequence, and use the YOLOv8 target detection network to perform the detection task of the well hole area and the well hole protective fence structure based on the target construction image sequence, and output the well hole detection box set and the fence detection box set; Step S102: Based on the well hole detection box set, use Python's OpenCV image processing library and NumPy numerical computing library to perform well hole geometric contour modeling task, and obtain well hole edge point set. Geometric center of the well ; Step S103: Based on the fence detection box set, a structural boundary resolution mechanism based on fence corner point localization is used to perform the fence structure key point extraction task, and obtain the fence corner point set. With the geometric center of the fence ; Step S104: Based on the set of points at the edge of the well. Geometric center of the well Fence corner point set With the geometric center of the fence Construct and output the initial set of spatial structure parameters.

3. The construction site well protection monitoring method based on YOLOv8 and Siamese Network as described in claim 2, characterized in that, Step S20, which involves performing the spatial relationship analysis task of the well protection structure based on the initial set of spatial structural parameters using a joint modeling mechanism for the geometric relationship between the well and the fence structure, and outputting the geometric feature vector of the well protection structure, specifically includes: Step S201: Based on the geometric center of the well shaft With the geometric center of the fence The spatial offset of the well enclosure is calculated using a two-dimensional Euclidean distance method. ; Step S202: Based on the geometric center of the well shaft With the geometric center of the fence The first angle of the fence structure is calculated using a polar coordinate angle calculation method based on the angular direction relationship. The angular features of the well enclosure at each corner point ; Step S203: Adjust the spatial offset of the well enclosure. Angular features of the well enclosure Perform unified vector encoding and splicing fusion processing to output the geometric feature vector of the well protection structure.

4. The construction site well protection monitoring method based on YOLOv8 and Siamese Network as described in claim 3, characterized in that, In step S202, based on the geometric center of the well... With the geometric center of the fence The first angle of the fence structure is calculated using a polar coordinate angle calculation method based on the angular direction relationship. The angular features of the well enclosure at each corner point The steps specifically include: Step S2021: Using the geometric center of the well shaft As the origin of the polar coordinates, and with the geometric center of the well... Pointing to the geometric center of the fence The direction is used as the polar coordinate reference direction to establish the polar coordinate reference coordinate system of the well protection structure; Step S2022: Obtain the fence structure The spatial positions of the corner points in the polar coordinate reference system are calculated. Each corner point is relative to the geometric center of the well. The direction deflection relationship vector; Step S2023: Calculate the fence structure based on the direction deflection relationship vector. The polar angle values ​​of each corner point in the polar coordinate reference system are determined, and these polar angle values ​​are used to define the fence structure. The angle characteristics of the well enclosure corresponding to each corner point .

5. The construction site well protection monitoring method based on YOLOv8 and Siamese Network as described in claim 1, characterized in that, In step S30, based on the geometric feature vector of the well protection structure, a historical sample statistical modeling mechanism is used to perform the task of generating a standard template for the well protection structure, and the template features of the well protection structure are output. template covariance matrix The steps specifically include: Step S301: Obtain images of well protection under normal protection status from historical construction monitoring data, and construct a standard training sample set of well protection structures based on the geometric feature vectors of well protection structures corresponding to the images of well protection. Step S302: Based on the standard training sample set of well protection structures, statistical analysis is performed using the statistical mean modeling method to output the average structural features. Based on the average structural features, the feature vector centering modeling method is used to generate template features for well protection structures. Features of well protection structure template Used to characterize the typical spatial structural relationship of well protection structures under normal conditions; Step S303: Based on the standard training sample set of well protection structures, the dispersion of well protection image samples is statistically analyzed using covariance statistical modeling to generate a template covariance matrix. Template covariance matrix This is used to characterize the range of variation of the geometric features of well protection structures under normal conditions and the correlation between the features.

6. The construction site well protection monitoring method based on YOLOv8 and Siamese Network as described in claim 1, characterized in that, In step S40, based on the features of the well shaft protection structure template... template covariance matrix The steps for performing the similarity evaluation task of well shaft protection structures using a Siamese Network-based structural consistency matching mechanism and outputting the similarity score of the well shaft protection structures specifically include: Step S401: Pre-set a structural feature embedding Siamese network, which includes: an input feature fusion layer, used to perform unified feature fusion processing on the geometric feature vector of the well protection structure, the template features of the well protection structure, and the template covariance matrix to form a structural input feature representation; a feature mapping layer, used to perform nonlinear mapping processing on the structural input feature representation to extract the high-dimensional spatial feature representation of the well protection structure; a structural embedding encoding layer, used to encode the high-dimensional spatial feature representation into a structural embedding feature vector to form a deep structural expression of the well protection structure; a two-branch feature comparison layer, used to perform structural feature comparison processing between two shared weight branches of the structural feature embedding Siamese network to calculate the feature distance relationship between the structural features to be detected and the structural template features; and a similarity evaluation output layer, used to calculate the similarity score of the well protection structure based on the feature distance relationship. Step S402: Obtain the historical training input dataset and the historical training output dataset. The historical training input dataset includes the historical geometric feature vector of the well protection structure, the historical template features of the well protection structure, and the historical template covariance matrix. Based on the historical training input dataset and the historical training output dataset, perform a pre-training process on the Siamese network by combining the contrastive loss function and the sample pair distance minimization training method. Step S403: Combine the geometric feature vector of the well protection structure and the template features of the well protection structure. template covariance matrix Input a pre-trained structural feature embedding twin network, and output a similarity score for the well protection structure.

7. The construction site well protection monitoring method based on YOLOv8 and Siamese Network as described in claim 1, characterized in that, In step S50, the spatial offset of the well enclosure is obtained from the geometric feature vector of the well protection structure. Based on the spatial offset of the well enclosure The steps for determining anomalies in well protection status and outputting well protection safety status monitoring results using a structural offset confidence propagation mechanism based on the similarity score of the well protection structure include: Step S501: Obtain the spatial offset of the well enclosure from the geometric feature vector of the well protection structure. Spatial offset of well enclosure Used to characterize the degree of spatial position change of the current fence structure relative to the standard well protection structure; Step S502: Based on the spatial offset of the well enclosure The similarity score of the well protection structure is corrected by confidence propagation based on covariance constraint, and the offset confidence index of the well protection structure is output. Step S503: When the offset confidence index of the well shaft protection structure is greater than the preset structural offset anomaly threshold, it is determined that the well shaft protection structure is in an abnormal protection state; when the offset confidence index of the well shaft protection structure is less than or equal to the preset structural offset anomaly threshold, it is determined that the well shaft protection structure is in a normal protection state; finally, the well shaft protection safety status monitoring result is output based on the judgment result.

8. A construction site well and shaft protection monitoring system based on YOLOv8 and Siamese Network, applied to the construction site well and shaft protection monitoring method based on YOLOv8 and Siamese Network as described in any one of claims 1 to 7, characterized in that, The construction site well protection monitoring system based on YOLOv8 and Siamese Network includes: The well and fence structure analysis module is used to acquire the target construction image sequence. Based on the target construction image sequence, it uses a collaborative mechanism of target detection and structural geometry extraction to perform the well and fence structure analysis task and output the initial spatial structure parameter set. The spatial structure relationship modeling module is used to perform the spatial relationship analysis task of the well protection structure based on the initial set of spatial structure parameters and the joint modeling mechanism of the geometric relationship between the well and the fence structure, and outputs the geometric feature vector of the well protection structure. The structural template generation module is used to perform the task of generating standard templates for well and tunnel protection structures based on the geometric feature vectors of the well and tunnel protection structures and employing a historical sample statistical modeling mechanism, outputting the template features of the well and tunnel protection structures. template covariance matrix ; The structural consistency matching module is used to match the features of the well shaft protection structure template. template covariance matrix A structural consistency matching mechanism based on Siamese Network is used to perform the similarity evaluation task of well hole protection structure and output the similarity score of well hole protection structure; The protection status determination module is used to obtain the spatial offset of the well enclosure from the geometric feature vector of the well protection structure. Based on the spatial offset of the well enclosure The similarity score between the well and the protective structure is used to determine the abnormality of the well protection status using the structural offset confidence propagation judgment mechanism, and the monitoring results of the safety status of the well protection are output.

9. A construction site well and shaft protection monitoring device based on YOLOv8 and Siamese Network, characterized in that, The construction site well protection and monitoring device based on YOLOv8 and Siamese Network includes: a memory, a processor, and a construction site well protection and monitoring program based on YOLOv8 and Siamese Network stored in the memory and executable on the processor. When the construction site well protection and monitoring program based on YOLOv8 and Siamese Network is executed by the processor, it implements the construction site well protection and monitoring method based on YOLOv8 and Siamese Network according to any one of claims 1 to 7.

10. A computer program product, characterized in that, The computer program product includes a construction site well protection and monitoring program based on YOLOv8 and Siamese Network. When the construction site well protection and monitoring program based on YOLOv8 and Siamese Network is executed by a processor, it implements a construction site well protection and monitoring method based on YOLOv8 and Siamese Network as described in any one of claims 1 to 7.

Images