Unitized curtain wall component matching method and system combined with AI recognition

Abstract

The application provides a kind of unit curtain wall component matching method and system combined with AI identification, first, obtain the field image data set of the unit curtain wall to be matched containing multiple component local image units with spatial position marks, then carry out image feature extraction processing to the field image data set, obtain the surface texture feature and structure contour feature of component local image unit, then call the pre-trained component matching model to carry out joint matching analysis on surface texture feature and structure contour feature, generate standard component type matching result, then determine the matching consistency information of each component local image unit and standard component library according to standard component type matching result, finally, generate matching optimization instruction containing component replacement or adjustment instruction based on the matching consistency information, and feedback to curtain wall construction management system to trigger component configuration adjustment operation, realize the automation and intelligentization of unit curtain wall component matching, improve matching efficiency and precision.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method and system for matching unitized curtain wall components combined with AI recognition. Background Art

[0002] In the field of architectural curtain wall engineering, unitized curtain walls have been widely used as a highly efficient and aesthetically pleasing form of building envelope. Unitized curtain walls are assembled from numerous different components, and their installation precision and quality directly impact the aesthetics, safety, and functionality of the entire building. Ensuring the precise matching of individual components is crucial during curtain wall construction.

[0003] Currently, the traditional method for matching unitized curtain wall components relies primarily on manual experience, with construction workers determining component matching through visual observation and comparison with drawings. However, this method has numerous drawbacks. On the one hand, manual judgment is easily influenced by subjective factors, and the experience and judgment standards of different construction workers vary, making it difficult to ensure the consistency and accuracy of matching results. On the other hand, with the rapid development of the construction industry, the variety and complexity of curtain wall components are constantly increasing. Manual matching methods are inefficient and cannot meet the requirements of large-scale, high-precision construction. In addition, traditional methods are unable to promptly detect potential problems in the component matching process, which can easily lead to component installation errors and rework during subsequent construction, increasing construction costs and the risk of delays. Summary of the Invention

[0004] In view of the above-mentioned problems, in combination with the first aspect of the present invention, an embodiment of the present invention provides a method for matching unitized curtain wall components in combination with AI recognition, the method comprising:

[0005] Acquire a set of on-site image data of a unitized curtain wall to be matched, wherein the set of on-site image data includes a plurality of component local image units with spatial position marks;

[0006] Performing image feature extraction processing on the on-site image data set to obtain surface texture features and structural contour features of the local image units of the component;

[0007] Calling a pre-trained component matching model to perform a joint matching analysis on the surface texture features and the structural contour features to generate a standard component type matching result corresponding to the component local image unit;

[0008] Determining matching consistency information between the local image unit of each component in the unitized curtain wall to be matched and the standard component library according to the standard component type matching result;

[0009] A matching optimization instruction including a component replacement or adjustment instruction is generated based on the matching consistency information, and the matching optimization instruction is fed back to the curtain wall construction management system to trigger a component configuration adjustment operation.

[0010] On the other hand, an embodiment of the present invention also provides a unit curtain wall component matching system combined with AI recognition, including a processor and a machine-readable storage medium, the machine-readable storage medium is connected to the processor, the machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0011] Based on the above aspects, the embodiment of the present invention obtains a set of on-site image data of the unitized curtain wall to be matched, and performs image feature extraction processing on these local image units of the components containing spatial position marks, comprehensively obtains the surface texture features and structural contour features of the local image units of the components, and calls a pre-trained component matching model to perform joint matching analysis on the extracted features. It can comprehensively consider various feature information of the components to improve the accuracy and reliability of matching, determine the matching consistency information of the local image units of each component and the standard component library based on the matching results, generate matching optimization instructions containing component replacement or adjustment instructions based on the matching consistency information, and feed back the matching optimization instructions to the curtain wall construction management system to trigger the component configuration adjustment operation, thereby realizing the automation and intelligence of the unitized curtain wall component matching, greatly improving the matching efficiency and accuracy, reducing the errors caused by manual intervention, and can effectively avoid construction problems caused by component matching errors, reduce construction costs and the risk of construction delays, and ensure the installation quality and overall performance of the unitized curtain wall. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a schematic diagram of the execution flow of the unitized curtain wall component matching method combined with AI recognition provided in an embodiment of the present invention.

[0013] Figure 2 Schematic diagram of exemplary hardware and software components of a unitized curtain wall component matching system combined with AI recognition provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0014] The present invention will be described in detail below with reference to the accompanying drawings. Figure 1 This is a flow chart of a method for matching unitized curtain wall components in combination with AI recognition provided by an embodiment of the present invention. The method for matching unitized curtain wall components in combination with AI recognition is introduced in detail below.

[0015] Step S110: obtaining a set of on-site image data of the unitized curtain wall to be matched, wherein the set of on-site image data includes a plurality of component local image units with spatial position marks.

[0016] In this embodiment, in order to achieve accurate matching of unitized curtain wall components, it is first necessary to obtain a set of on-site image data of the unitized curtain wall to be matched. For example, a variety of image acquisition devices can be selected, such as high-definition cameras, drones equipped with high-definition cameras, etc. Different acquisition strategies can be adopted for unitized curtain walls of different scales and scenes. For example, in a unitized curtain wall project in a large business center, if the curtain wall height is low and the surrounding environment is open, a high-definition camera can be used to shoot multi-angle shots around the curtain wall. The shooting positions can be distributed in different directions such as the front, back, left, right, top, and bottom of the curtain wall to ensure that images of all parts of the curtain wall can be fully captured. If the curtain wall is located on a high-rise building, it will be more efficient to use a drone for image acquisition. The drone can fly and shoot according to a pre-set route, and the route planning should cover every corner and different levels of the curtain wall.

[0017] During image acquisition, each component's local image unit needs to be spatially labeled. This can be achieved, for example, by using a high-precision positioning system like the Global Positioning System (GPS) combined with the building's internal coordinate system. For indoor unitized curtain walls, a three-dimensional coordinate system is established with a fixed reference point in the building as the origin. Using equipment such as laser rangefinders and angle measuring instruments, the coordinate position of each component's local image unit relative to the origin is precisely measured, thereby assigning an accurate spatial location marker. In this way, the resulting on-site image data set contains multiple component local image units with spatial location markers.

[0018] Step S120: performing image feature extraction processing on the on-site image data set to obtain surface texture features and structural contour features of the local image units of the component.

[0019] After acquiring the on-site image data set, the next step is to perform image feature extraction to obtain the surface texture features and structural contour features of the component's local image units. These two features play a key role in subsequent component matching. Surface texture features can reflect information such as the component's material and manufacturing process, while structural contour features can reflect geometric characteristics such as the component's shape and size.

[0020] Step S121: performing image preprocessing operations on the on-site image data set to obtain preprocessed component local image units, wherein the image preprocessing operations include grayscale processing and noise filtering processing.

[0021] In this embodiment, grayscale processing converts the local image units of the colored components into grayscale images. This reduces the interference of color information on subsequent feature extraction and allows the processing to focus more on the brightness information of the image. A common grayscale conversion method is the weighted averaging method. Based on the different sensitivities of the human eye to different color channels, different weights are assigned to the three color channels of red (R), green (G), and blue (B). Typically, the red channel has a weight of 0.299, the green channel has a weight of 0.587, and the blue channel has a weight of 0.114. For each pixel in the image, its grayscale value (Gray) can be calculated using the formula Gray = 0.299 × R + 0.587 × G + 0.114 × B.

[0022] After the grayscale processing is completed, noise filtering is performed. The image may be affected by various factors during the acquisition process, such as lighting changes, sensor noise, etc., resulting in noise points in the image. In order to improve the image quality, the Gaussian filtering algorithm can be used. The principle of Gaussian filtering is to perform weighted averaging on each pixel point and its neighborhood in the image according to the Gaussian function. With a certain pixel point as the center, a neighborhood window of a set size is selected. The weight of each pixel point in the neighborhood window is determined by its distance from the center pixel point. The closer the distance, the greater the weight. By performing weighted summation on the pixels in the neighborhood window, the new value of the pixel point after filtering is obtained. After grayscale processing and noise filtering, the pre-processed component local image unit is obtained.

[0023] Step S122: performing texture feature capture processing on the pre-processed local image unit of the component, and extracting the texture distribution regularity characteristics of the local image unit of the component as surface texture features through a local binary pattern algorithm. The surface texture features include texture repetition period parameters and texture contrast parameters.

[0024] Step S1221: Divide the pre-processed component local image unit into a plurality of non-overlapping texture analysis window units, each texture analysis window unit containing a fixed number of pixels.

[0025] After obtaining the preprocessed component local image unit, it is divided into multiple non-overlapping texture analysis window units. The size and shape of the window should be determined based on the specific characteristics of the image and the needs of subsequent analysis. For example, for images with fine textures, a smaller window size can be selected to more accurately capture texture details; for images with coarser textures, a larger window size can be selected. Each texture analysis window unit contains a fixed number of pixels, ensuring a consistent analysis standard in subsequent processing.

[0026] Step S1222: For each texture analysis window unit, the grayscale value of the central pixel of the texture analysis window unit is used as a reference value, and the magnitude relationship between other pixels in the texture analysis window unit and the reference value is calculated to generate a binary code sequence.

[0027] For each divided texture analysis window unit, the grayscale value of its central pixel is used as the reference value. Then, the grayscale values ​​of all pixels in the window, except the central pixel, are compared with the reference value. If the grayscale value of a pixel is greater than or equal to the reference value, it is recorded as 1; if it is less than the reference value, it is recorded as 0. The pixels in the window are compared in a set order (e.g., clockwise), and the comparison results are arranged to form a binary code sequence. For example, for a 3×3 texture analysis window unit, there are 8 pixels surrounding the central pixel. After comparison, an 8-bit binary code sequence can be obtained.

[0028] Step S1223: converting the binary code sequence into a decimal value as the local binary pattern feature value of the texture analysis window unit.

[0029] After obtaining the binary code sequence, it is converted into a decimal value. The decimal value is the local binary pattern eigenvalue of the texture analysis window unit. The conversion method is to follow the conventional binary to decimal algorithm, multiply each bit in the binary code sequence by the corresponding weight (power of 2), and then add the results.

[0030] Step S1224: Count the occurrence frequencies of the local binary pattern feature values ​​of all texture analysis window units to generate a histogram distribution of the local binary pattern feature values.

[0031] In this embodiment, a counter can be used to record the number of times each eigenvalue appears. The frequency of occurrence of each eigenvalue is then divided by the total number of texture analysis window units. A histogram is plotted with the eigenvalue as the horizontal axis and the frequency of occurrence as the vertical axis to generate a histogram distribution of the local binary pattern eigenvalues. This histogram distribution can intuitively reflect the distribution of texture eigenvalues ​​in the image.

[0032] Step S1225: Determine a texture repetition period parameter and a texture contrast parameter of the local image unit of the component based on the histogram distribution, wherein the texture repetition period parameter is determined by the periodic fluctuation law of the eigenvalues ​​in the histogram distribution, and the texture contrast parameter is determined by the distribution concentration degree of the eigenvalues ​​in the histogram distribution.

[0033] In this embodiment, the texture repetition period parameter is determined by observing the periodic fluctuation pattern of the eigenvalues ​​in the histogram distribution. If certain eigenvalues ​​in the histogram show periodic repetition, the texture repetition period parameter can be obtained by analyzing the length of these periods. The texture contrast parameter is determined by the degree of concentration of the eigenvalues ​​in the histogram distribution. If the eigenvalues ​​are relatively concentrated in the histogram, it means that the texture contrast is low, that is, the texture changes relatively slowly; if the eigenvalue distribution is relatively dispersed, it means that the texture contrast is high, that is, the texture changes relatively drastically.

[0034] Step S123: performing contour edge detection processing on the pre-processed component local image unit, using an edge detection algorithm to identify contour boundary information of the component local image unit, and generating structural contour features based on the contour boundary information, wherein the structural contour features include contour closure integrity parameters and contour curvature change parameters.

[0035] Step S1231: Perform smoothing filtering on the preprocessed local image unit of the component, and calculate the gradient amplitude and gradient direction of each pixel in the smoothed image, where the gradient amplitude is calculated by the convolution operation results in the horizontal and vertical directions, and the gradient direction is calculated by the relative relationship between the horizontal and vertical gradients.

[0036] Smoothing filtering is performed on the preprocessed local component image units to further reduce image noise and enhance edge detection accuracy. A Gaussian filter algorithm can also be used, similar in principle to the noise filtering used in image preprocessing. After smoothing filtering, the gradient magnitude and gradient direction are calculated for each pixel in the smoothed image. The gradient magnitude reflects the rate of grayscale change at that pixel, while the gradient direction indicates the direction of the greatest grayscale change.

[0037] Horizontal and vertical convolution operations are typically used to calculate the gradient magnitude. The Sobel operator can be used, which includes a horizontal Sobel operator and a vertical Sobel operator. The horizontal Sobel operator is convolved with the image to obtain the horizontal gradient value (Gx); the vertical Sobel operator is convolved with the image to obtain the vertical gradient value (Gy). The gradient magnitude is then calculated using the formula GradientMagnitude = √(Gx² + Gy²). The gradient direction can be calculated using the formula GradientDirection = arctan(Gy / Gx), where arctan is the inverse tangent function.

[0038] Step S1232: performing non-maximum suppression processing on the gradient amplitude, retaining the local maximum pixel points in the gradient direction, and obtaining a set of candidate contour edge pixel points.

[0039] After obtaining the gradient magnitude and gradient direction for each pixel, non-maximum suppression is performed on the gradient magnitude. The purpose of non-maximum suppression is to refine the edges and retain only the pixels with local maxima along the gradient direction. For example, the gradient magnitude of each pixel can be compared with that of adjacent pixels along the gradient direction. If the gradient magnitude of the pixel is not a local maximum, its gradient magnitude is set to 0, and only the pixel with the maximum gradient magnitude is retained. After non-maximum suppression, a set of candidate contour edge pixels is obtained, which may constitute the contour edge of the image.

[0040] Step S1233: using a dual-threshold detection method to perform edge connection processing on the candidate contour edge pixel point set to generate a continuous contour boundary line as the contour boundary information.

[0041] A dual-threshold detection method is used to perform edge connection processing on the candidate contour edge pixel set. The dual-threshold detection method requires setting two thresholds: a high threshold (HighThreshold) and a low threshold (LowThreshold). For each pixel in the candidate contour edge pixel set, if its gradient amplitude is greater than the high threshold, it is marked as a strong edge pixel; if its gradient amplitude is less than the low threshold, it is marked as a non-edge pixel; if its gradient amplitude is between the high and low thresholds and the pixel is connected to a strong edge pixel, it is marked as a weak edge pixel and is also considered an edge pixel. In this way, the edge pixels in the candidate contour edge pixel set are connected to form continuous contour boundary lines, which serve as contour boundary information.

[0042] Step S1234: Determine the contour closure integrity parameter and contour curvature variation parameter of the local image unit of the component based on the continuous contour boundary line as the structural contour feature. The contour closure integrity parameter is determined by the positional relationship between the starting point and the end point of the contour boundary line, and the contour curvature variation parameter is determined by the curvature variation law of each point on the contour boundary line.

[0043] In this embodiment, the contour closure integrity parameter is determined by observing the positional relationship between the starting and ending points of the contour boundary line. If the starting and ending points coincide or are very close, the contour boundary line is closed and the contour closure integrity parameter is high. If the starting and ending points are far apart, the contour boundary line may be broken and the contour closure integrity parameter is low.

[0044] The profile curvature variation parameter is determined by analyzing the variation in curvature at each point along the contour boundary. The curvature of each point on the contour boundary can be calculated, reflecting the degree of curvature at that point. By observing the variation in curvature, such as its magnitude and frequency, the profile curvature variation parameter can be derived, thereby forming the structural profile characteristics of the component's local image unit.

[0045] Step S124: performing correlation analysis on the surface texture features and the structural contour features across feature dimensions to obtain a correlation feature set having a unified feature space representation.

[0046] After obtaining the surface texture and structural contour features of a component's local image units, these two features need to be analyzed for cross-dimensional correlation. While surface texture and structural contour features describe component characteristics from different perspectives, they may have certain correlations. Correlation analysis can uncover this potential correlation information and integrate it into a unified feature space.

[0047] Surface texture features and structural contour features can be fused using feature fusion methods, such as linear combination and nonlinear mapping. For example, surface texture features and structural contour features are first normalized to the same scale range. Then, different weights are assigned to different features based on their importance, and a weighted combination is performed. This approach yields a set of correlated features with a unified feature space representation. The features in this set can more comprehensively and accurately describe the characteristics of the component's local image units.

[0048] Step S125: Based on the importance evaluation results of different feature dimensions in the associated feature set, dynamically adjust the feature fusion weight parameters of the surface texture feature and the structural contour feature in the joint matching analysis.

[0049] In this embodiment, different feature dimensions may have different importance in the component matching process, so their weights need to be adjusted according to actual conditions.

[0050] Machine learning methods, such as decision trees and random forests, can be used to assess the importance of different feature dimensions within a set of associated features. These methods calculate an importance score for each feature dimension based on its performance in classification or regression tasks. Based on these importance scores, the feature fusion weight parameters for surface texture and structural contour features are dynamically adjusted. For example, if the evaluation results indicate that surface texture features are more important in the current matching task, the weight of surface texture features is increased; if structural contour features are more important, the weight of structural contour features is increased. By dynamically adjusting the weight parameters, the accuracy and efficiency of joint matching analysis can be improved.

[0051] Step S130: calling a pre-trained component matching model to perform a joint matching analysis on the surface texture features and the structural contour features, and generating a standard component type matching result corresponding to the component local image unit.

[0052] In the previous steps, surface texture features reflect information such as the component's surface material and manufacturing process, and are reflected by parameters such as texture repetition period and texture contrast. Structural contour features reflect geometric characteristics such as the component's shape and size, and include parameters such as contour closure integrity and contour curvature variation. Next, a pre-trained component matching model will be used to perform a joint matching analysis on these two features to determine the standard component type corresponding to the component's local image unit.

[0053] Step S131: input the texture distribution law of the surface texture feature and the contour morphological feature of the structural contour feature into the feature alignment module of the component matching model, and the feature alignment module performs coordinate alignment processing on the texture distribution law and the contour morphological feature based on pre-trained spatial mapping parameters, so that the key area of ​​the texture distribution and the boundary area of ​​the contour morphology correspond one-to-one in spatial position, and generate a dual feature alignment group with a position correspondence relationship.

[0054] The feature alignment module of the component matching model is a crucial step in the entire matching process. Once the texture distribution patterns of surface texture features and the contour morphology of structural profile features are input into the module, it performs coordinate alignment based on pre-trained spatial mapping parameters.

[0055] The pre-trained spatial mapping parameters are obtained through training with a large amount of sample data. During the training phase, image data of various types of unitized curtain wall components were collected. The surface texture features and structural contour features in these data were analyzed and annotated to identify the spatial correspondence between texture distribution patterns and contour morphological features, thereby obtaining a set of parameters that accurately describe this correspondence.

[0056] The specific coordinate alignment process is as follows: First, the feature alignment module considers texture distribution patterns and contour morphological features as feature representations in different spatial coordinate systems. For texture distribution patterns, key regions may include specific locations where textures recur or areas where texture contrast varies significantly. For contour morphological features, boundary regions are the edges of component contours. The module then transforms and aligns the coordinate systems of the texture distribution patterns and contour morphological features based on pre-trained spatial mapping parameters.

[0057] For example, for a key area A in the texture distribution law, the corresponding boundary area B is found in the coordinate system of the contour morphological features through the spatial mapping parameters. This process involves complex coordinate transformation calculations, including operations such as translation, rotation, and scaling. Assume that the coordinate system where the texture distribution law is located is coordinate system T, and the coordinate system where the contour morphological features are located is coordinate system C. The spatial mapping parameters include the transformation matrix M from coordinate system T to coordinate system C. For the coordinates (xT, yT) of the key area A in coordinate system T, the matrix operation (xC, yC) = M*(xT, yT) is used to obtain the corresponding coordinates in coordinate system C, thereby achieving position correspondence between the key area and the boundary area.

[0058] After this coordinate alignment process, the key areas of texture distribution correspond to the boundary areas of the contour morphology in spatial position, and finally a dual-feature alignment group with positional correspondence is generated. This dual-feature alignment group integrates the spatial correspondence information of surface texture features and structural contour features.

[0059] Step S132: Input the dual-feature alignment group into the context fusion layer of the component matching model, extract the spatial position mark of the local image unit of the component through the context fusion layer, and obtain the structural topology information corresponding to the spatial position mark from the standard component library, and generate a context enhancement feature containing the curtain wall structure position information by feature association encoding the structural topology information and the dual-feature alignment group.

[0060] Step S1321: obtaining the spatial position mark of the local image unit of the component. The spatial position mark is directly obtained from the spatial position mark in the on-site image data set and is used to indicate the specific installation coordinates of the component in the entire curtain wall.

[0061] When the dual-feature alignment group is input into the context fusion layer, it directly extracts the spatial position markers of the component's local image units from the on-site image dataset. During the previous acquisition of the on-site image dataset, spatial position markers were added to each component's local image unit. These markers were obtained using a high-precision positioning system, such as the Global Positioning System (GPS), combined with the building's internal coordinate system. For indoor unitized curtain walls, a three-dimensional coordinate system is established with a fixed reference point in the building as the origin. Using equipment such as laser rangefinders and angle measuring instruments, the coordinate position of each component's local image unit relative to the origin is precisely measured, thereby assigning an accurate spatial position marker. The context fusion layer directly reads these spatial position markers, which accurately indicate the component's specific installation coordinates within the overall curtain wall.

[0062] Step S1322: querying the structural topology information corresponding to the spatial position mark from the standard component library, wherein the structural topology information includes the horizontal connection level and the vertical adjacent component type of the spatial position mark in the curtain wall.

[0063] Based on the spatial position markers of the component's local image units, the context fusion layer queries the standard component library to obtain the corresponding structural topology information. The standard component library is a database that stores detailed information about all standard components, including the structural topology corresponding to different spatial locations. This structural topology information reflects the positional relationships and connections of the components within the curtain wall structure, specifically including the horizontal connection levels and vertical adjacent component types of the spatial position marker within the curtain wall.

[0064] The horizontal connection level describes the horizontal connection hierarchy of components in the curtain wall. For example, in a multi-story unitized curtain wall, components may be located at different horizontal connection levels, with each layer potentially having different connection methods and structural characteristics. By querying the standard component library, you can determine the horizontal connection level of a component and understand its horizontal connection method and relationship with other components.

[0065] The longitudinally adjacent component type specifies the type of other components that are adjacent to the component in the longitudinal direction. In curtain wall structures, longitudinally adjacent components must match and coordinate with each other to ensure the overall stability and functionality of the curtain wall. By querying the standard component library, you can obtain information about the component types that are adjacent to the component in the longitudinal direction.

[0066] Step S1323: Perform feature association encoding on the surface texture features in the dual-feature alignment group and the lateral connection levels in the structural topology information to generate hierarchical association features that reflect the adaptability of texture distribution and connection levels. The feature association encoding is achieved by learning typical texture distribution patterns corresponding to different connection levels in historical installation data.

[0067] After obtaining the dual-feature alignment group and the structural topology information, the context fusion layer performs feature correlation encoding on the surface texture features in the dual-feature alignment group and the horizontal connection hierarchy in the structural topology information. The goal of this process is to generate hierarchical correlation features that reflect the adaptability of the texture distribution and the connection hierarchy.

[0068] To achieve feature association encoding, the context fusion layer learns typical texture distribution patterns corresponding to different connection levels from historical installation data. This historical data contains a large number of curtain wall installation cases, which record the surface texture characteristics of components at different connection levels. By analyzing and learning from this historical data, typical texture distribution patterns corresponding to different connection levels can be summarized.

[0069] The specific feature association encoding process is as follows: First, the surface texture features in the dual-feature alignment group are decomposed and extracted to obtain parameters related to texture distribution, such as texture repetition period and texture contrast. Then, based on the horizontal connection level information, corresponding patterns are found from the learned typical texture distribution patterns. For each parameter in the surface texture feature, it is compared and matched with the typical pattern of the corresponding connection level.

[0070] For example, for the texture repetition period parameter, a typical texture repetition period P corresponding to a certain horizontal connection level is found in historical installation data. The texture repetition period parameter p of the surface texture feature in the dual-feature alignment group is compared with P. If p is close to P, it indicates that the texture repetition period is well-suited to the connection level. Through this comparison and matching, a suitability score is assigned to each surface texture feature parameter.

[0071] Finally, these adaptability scores are combined and encoded to generate hierarchical correlation features that reflect the adaptability of texture distribution and connection hierarchy. This hierarchical correlation feature can reflect the intrinsic relationship between surface texture features and lateral connection hierarchy.

[0072] Step S1324: Perform feature association coding on the structural contour features in the dual-feature alignment group and the longitudinal adjacent component types in the structural topology information to generate adjacency association features that reflect the matching of contour morphology and adjacent components. The feature association coding is obtained by learning typical contour morphology patterns corresponding to different adjacent component types in historical installation data.

[0073] In addition to associating and encoding surface texture features with lateral connection levels, the context fusion layer also associates and encodes the structural contour features in the dual-feature alignment group with the vertical adjacent component types in the structural topology information to generate adjacency association features that reflect the matching between the contour morphology and the adjacent components.

[0074] Similarly, the context fusion layer learns typical contour morphological patterns corresponding to different adjacent component types from historical installation data. By analyzing and learning from historical installation data, which records the structural contour features of components under different combinations of adjacent component types, typical contour morphological patterns corresponding to different adjacent component types can be summarized.

[0075] During feature association encoding, the structural contour features within the dual-feature alignment group are first decomposed and extracted to obtain parameters such as contour closure integrity and contour curvature variation. Then, based on the information about the type of longitudinally adjacent components, corresponding patterns are found from the learned typical contour morphological patterns. For each parameter within the structural contour feature, it is compared and matched with the typical pattern of the corresponding adjacent component type.

[0076] For example, for the profile closure integrity parameter, a typical profile closure integrity parameter C is found in historical installation data for a certain combination of longitudinally adjacent component types. The profile closure integrity parameter c of the structural profile feature in the dual-feature alignment group is compared with C. If c is close to C, it indicates that the profile closure integrity of the structural profile feature matches the adjacent component type well. Through this comparison and matching, a matching score is assigned to each structural profile feature parameter.

[0077] Finally, these matching scores are combined and coded to generate an adjacency association feature that reflects the matching between the profile morphology and adjacent components. This adjacency association feature can reflect the intrinsic relationship between the structural profile characteristics and the types of longitudinally adjacent components.

[0078] Step S1325: Dynamically weight the hierarchical association features and the adjacency association features through the attention mechanism module of the context fusion layer to generate context-enhanced features containing curtain wall structure position information. The weight allocation is determined based on the degree of influence of different connection levels and adjacent component types in the historical installation data on the component matching results.

[0079] After generating hierarchical association features and adjacency association features, the attention mechanism module of the context fusion layer dynamically assigns weights to these two features to generate context-enhanced features containing the curtain wall structure location information.

[0080] The attention mechanism module determines weights based on the degree to which different connection levels and adjacent component types influence component matching results in historical installation data. Historical installation data records a large number of curtain wall installation cases, including component matching results at different connection levels and adjacent component types. By analyzing and statistically analyzing this historical data, we can calculate the degree to which different connection levels and adjacent component types influence component matching results.

[0081] For example, through analysis of historical data, we found that in some cases, the horizontal connection level has a greater impact on component matching results, while in other cases, the vertical adjacent component types have a more critical impact. The attention mechanism module dynamically adjusts the weights of hierarchical and adjacent features based on the specific situation of the current component.

[0082] The specific weight allocation process is as follows: First, the attention mechanism module calculates the influence factors of different connection levels and adjacent component types based on historical data. For hierarchical correlation features, the weight w1 is determined by the influence factor of the horizontal connection level; for adjacency correlation features, the weight w2 is determined by the influence factor of the vertical adjacent component type. Then, the hierarchical correlation features and adjacency correlation features are multiplied by the corresponding weights to obtain the weighted features.

[0083] Finally, the weighted hierarchical and adjacency-related features are concatenated and fused to generate context-enhanced features that contain the curtain wall structure's positional information. This context-enhanced feature comprehensively considers the compatibility of surface texture features with horizontal connection levels and the matching of structural profile features with vertical adjacent component types, providing more accurate and comprehensive context for subsequent feature filtering and matching analysis.

[0084] Step S133: Input the context enhancement feature into the constraint filtering layer of the component matching model, extract the standard constraint conditions corresponding to the installation position according to the structural topology information through the constraint filtering layer, and generate the constraint enhancement feature focusing on the key installation features by filtering the redundant features in the context enhancement feature that do not meet the standard constraint conditions.

[0085] When context-enhanced features are fed into the constraint filtering layer, they extract standard constraints for the corresponding installation locations based on the structural topology information. These standard constraints are formulated based on curtain wall design requirements and installation specifications to ensure the correctness and compatibility of components in specific installation locations.

[0086] Structural topology information includes information such as the component's specific installation location within the curtain wall and its connections to other components. Based on this information, the constraint filtering layer extracts specific constraints for the corresponding installation location from the standard constraint database. For example, a specific installation location may have constraints on component size, shape, and connection methods.

[0087] After extracting the standard constraints, the constraint filtering layer filters the context-enhanced features, removing redundant features that do not meet the standard constraints. The specific filtering process is as follows: First, the context-enhanced features are decomposed and extracted to obtain individual feature parameters. Each feature parameter is then compared and matched against the standard constraints.

[0088] For example, if a dimension parameter in a context-enhanced feature exceeds the range specified by the standard constraint, the feature corresponding to that parameter is considered redundant and requires filtering. Through this comparison and filtering operation, features that do not meet the standard constraint are removed, and only key features that meet the requirements are retained.

[0089] Finally, the retained key features are recombined and encoded to generate a constraint-enhanced feature that focuses on the key installation features. This constraint-enhanced feature only contains key features that match the standard constraints of the corresponding installation location.

[0090] Step S134: Input the constraint enhancement feature into the feature aggregation layer of the component matching model, and perform cross-dimensional weighted fusion on the texture distribution law, contour morphological features and structural position information in the constraint enhancement feature through the feature aggregation layer to generate a spatial enhancement feature vector that can comprehensively reflect the relationship between component features and installation scenarios.

[0091] After the constraint-enhanced features are obtained, they are input into the feature aggregation layer of the component matching model. The main task of the feature aggregation layer is to perform a cross-dimensional weighted fusion of the texture distribution patterns, contour morphological characteristics, and structural position information in the constraint-enhanced features to generate a spatially enhanced feature vector that comprehensively reflects the relationship between component features and installation scenarios.

[0092] First, the feature aggregation layer further decomposes and extracts the constraint-enhanced features, separating and organizing the texture distribution patterns, contour morphological features, and structural position information. Texture distribution patterns include parameters related to surface texture features, such as texture repetition period and texture contrast; contour morphological features include parameters related to structural contour features, such as contour closure integrity and contour curvature variation; and structural position information is composed of the spatial location markers and structural topology information extracted previously.

[0093] The feature aggregation layer then assigns different weights to each feature dimension. These weights are determined based on historical data and experimental results, reflecting the importance of each feature dimension in the component matching process. For example, in some installation scenarios, texture distribution patterns may have a greater impact on component matching and therefore be assigned a higher weight; in other scenarios, contour morphology may be more critical and therefore receive a higher weight accordingly.

[0094] The specific weighted fusion process is as follows: For each parameter in the texture distribution law, it is multiplied by the corresponding weight to obtain the weighted texture distribution parameter. Similarly, each parameter in the contour morphological characteristics and structural position information is weighted. Then, the weighted texture distribution parameters, contour morphological parameters, and structural position parameters are spliced ​​and combined.

[0095] For example, suppose there are n parameters for texture distribution, m parameters for contour morphology, and k parameters for structural position information. The weighted parameters are t1, t2, ..., tn, p1, p2, ..., pm, s1, s2, ..., sk. Concatenating these parameters in a certain order yields a vector of length n + m + k.

[0096] Finally, the concatenated vectors are normalized to a uniform scale and range, resulting in a spatially enhanced feature vector that comprehensively reflects the relationship between component characteristics and installation scenarios. This spatially enhanced feature vector integrates cross-dimensional information such as texture distribution patterns, contour morphology, and structural position.

[0097] Step S135: Call the classification output layer of the component matching model to perform standard component type confidence prediction processing on the spatial enhancement feature vector, and generate a standard component type matching result including a standard component type confidence distribution map, wherein the standard component type confidence distribution map is used to represent the matching confidence corresponding to different standard component types.

[0098] For example, step S1351: obtaining a reference confidence distribution curve of all standard component types in the standard component library, wherein the reference confidence distribution curve is obtained by statistically analyzing the characteristic distribution frequency of each standard component type in historical training data.

[0099] Before calling the classification output layer of the component matching model to perform confidence prediction, it is necessary to obtain the baseline confidence distribution curves of all standard component types in the standard component library. These baseline confidence distribution curves are generated based on historical training data.

[0100] The historical training data contains a large amount of feature data of standard components and the corresponding standard component type labels. Through statistical analysis of these historical data, the distribution frequency of the features of each standard component type in different feature dimensions is calculated.

[0101] The specific statistical process is as follows: First, historical training data is classified according to standard component types. For each standard component type, feature data is extracted, decomposed, and statistically analyzed. For example, for surface texture features, the distribution frequency of texture repetition period parameters and texture contrast parameters is statistically analyzed; for structural contour features, the distribution frequency of contour closure completeness parameters and contour curvature variation parameters is statistically analyzed.

[0102] Then, based on these distribution frequencies, a confidence distribution curve is constructed for each standard component type. This confidence distribution curve represents the confidence level of the features of that standard component type across various feature dimensions. In this way, a baseline confidence distribution curve for all standard component types in the standard component library is obtained, providing a reference standard for subsequent confidence predictions.

[0103] Step S1352: input the spatial enhancement feature vector into the confidence calibration submodule of the classification output layer, calculate the Kullback-Leibler divergence between the spatial enhancement feature vector and each benchmark confidence distribution curve, and generate an initial confidence distribution curve.

[0104] After the spatial enhancement feature vector is input into the confidence calibration submodule of the classification output layer, it calculates the Kullback-Leibler divergence between the spatial enhancement feature vector and each baseline confidence distribution curve. The Kullback-Leibler divergence is a metric used to measure the difference between two confidence distributions. It reflects the degree of difference between the feature distribution represented by the spatial enhancement feature vector and the baseline confidence distribution of each standard component type.

[0105] The specific calculation process is as follows: For each standard component type's baseline confidence distribution curve B, the spatial enhancement feature vector S is transformed and processed to obtain its corresponding confidence distribution curve P. Then, the Kullback-Leibler divergence D(P||B) between P and B is calculated.

[0106] The Kullback-Leibler divergence is calculated as: D(P||B)=∑P(x)*log(P(x) / B(x)), where x represents the feature dimension and x spans all feature dimensions. In actual calculations, the confidence value for each feature dimension is calculated and the results are accumulated to obtain the total Kullback-Leibler divergence.

[0107] By calculating the Kullback-Leibler divergence between the spatial enhancement feature vector and the baseline confidence distribution curve for each standard component type, a set of divergence values ​​can be obtained. These divergence values ​​reflect the degree of difference between the spatial enhancement feature vector and the feature distribution of different standard component types. The smaller the divergence value, the closer the feature distribution represented by the spatial enhancement feature vector is to the baseline confidence distribution of the standard component type.

[0108] Based on the calculated divergence values, an initial confidence distribution curve is generated. This is achieved by transforming and normalizing the divergence values. First, the reciprocal of each divergence value is taken to produce a new set of values. This is because smaller divergence values ​​indicate a higher degree of match, and taking the reciprocal transforms small divergence values ​​into larger values ​​that better reflect the confidence level. These new values ​​are then normalized so that their sum is 1. This normalization is achieved by dividing each new value by the sum of all new values.

[0109] After this processing, a set of confidence values ​​corresponding to different standard component types is obtained. By plotting a curve with the standard component type as the horizontal axis and the confidence value as the vertical axis, we obtain an initial confidence distribution curve. This initial confidence distribution curve represents the initial matching confidence of the spatial enhancement feature vector corresponding to different standard component types.

[0110] Step S1353: extracting standard component type matching results of other component local image units spatially adjacent to the current component local image unit in the on-site image data set, wherein the standard component type matching results of the adjacent components include confidence distribution curves of the adjacent components.

[0111] The standard component type matching results for other component partial image units spatially adjacent to the current component partial image unit are extracted from the scene image dataset. Previously, a standard component type matching analysis was performed on each component partial image unit in the scene image dataset, resulting in corresponding matching results. These matching results are presented as confidence distribution curves, reflecting the matching confidence level of each component partial image unit for different standard component types.

[0112] Since adjacent components in actual unitized curtain walls often have certain correlations and their types may be similar or complementary, the standard component type matching results of adjacent components can provide important contextual information for the matching of the current component.

[0113] To extract the standard component type matching results for adjacent components, it is necessary to determine adjacent components based on the spatial position marks of the current component's local image units. The spatial position marks accurately record the specific location of each component in the curtain wall. By comparing the relative relationships between the spatial position marks, it is possible to determine which components are adjacent to the current component.

[0114] For example, in a three-dimensional coordinate system, two components are considered adjacent if the coordinate differences between their spatial position markers in the x, y, and z directions are within a certain range. For components identified as adjacent, the corresponding standard component type matching results, i.e., the confidence distribution curves for the adjacent components, are extracted from the on-site image dataset. These confidence distribution curves contain confidence information for the adjacent components corresponding to different standard component types, providing a foundation for subsequent spatial context constraint processing.

[0115] Step S1354: performing spatial context constraint processing on the initial confidence distribution curve based on the confidence distribution curve of the adjacent components, wherein the constraint processing adjusts the peak position of the initial confidence distribution curve by calculating the cross entropy loss value between the confidence distribution curve of the adjacent components and the initial confidence distribution curve.

[0116] The initial confidence distribution curve is subjected to spatial context constraint processing based on the confidence distribution curves of adjacent components, in order to make the initial confidence distribution curve more consistent with the confidence distribution of adjacent components, thereby improving the accuracy of the matching results.

[0117] Cross-entropy loss is a metric used to measure the difference between two confidence distributions. In this step, the peak position of the initial confidence distribution curve is adjusted by calculating the cross-entropy loss between the confidence distribution curves of adjacent components and the initial confidence distribution curve.

[0118] The specific calculation process is as follows: For each adjacent component's confidence distribution curve, compare it with the initial confidence distribution curve. Assuming the adjacent component's confidence distribution curve is Q and the initial confidence distribution curve is P, the cross entropy loss is calculated as: H(P,Q)=-∑P(x)*log(Q(x)), where x traverses all standard component types.

[0119] In actual calculations, the confidence values ​​corresponding to each standard component type are calculated and the results are accumulated to obtain the total cross-entropy loss value. The larger the cross-entropy loss value, the greater the difference between the two confidence distributions.

[0120] If the confidence distribution curves of adjacent components show a high confidence for a certain standard component type, while the initial confidence distribution curve shows a low confidence for that standard component type, then the cross-entropy loss will be large. In this case, it is necessary to adjust the initial confidence distribution curve to increase the confidence for that standard component type and decrease the confidence for other standard component types to reduce the cross-entropy loss.

[0121] The adjustment process can be performed iteratively. First, the current cross-entropy loss value is calculated. Then, based on the magnitude of the loss value, the confidence values ​​of each standard component type in the initial confidence distribution curve are slightly adjusted. For example, the confidence values ​​of the standard component types corresponding to high-confidence adjacent components can be increased by a certain step size, while the confidence values ​​of other standard component types are correspondingly decreased to ensure that the total confidence value of the confidence distribution curve is always 1.

[0122] The adjusted cross entropy loss is recalculated. If the loss decreases, the adjustment continues. If the loss stops decreasing or reaches the preset convergence condition, the adjustment stops. By continuously adjusting the initial confidence distribution curve to make it more consistent with the confidence distribution of adjacent components, spatial context constraints are formed.

[0123] Step S1355: The adjusted confidence distribution curve is used as the standard component type matching result, and the confidence peak position of the standard component type matching result forms a spatial compatibility constraint relationship with the standard component type of the adjacent component.

[0124] The confidence distribution curve adjusted after spatial context constraint processing is used as the standard component type matching result. The standard component type matching result is presented in the form of a confidence distribution curve, which intuitively shows the matching confidence corresponding to different standard component types.

[0125] In actual unitized curtain walls, adjacent components must be compatible in type to ensure the proper functioning of the overall structure and functionality. Therefore, through spatial context constraint processing, the confidence peak position of the adjusted confidence distribution curve forms a spatial compatibility constraint relationship with the standard component type of the adjacent components.

[0126] For example, if most adjacent components are of a specific standard type, then after applying spatial context constraints, the confidence peak for that standard component type in the adjusted confidence distribution curve will be higher. This means that the component is more likely to be of that specific standard type, ensuring that the final matching result meets the actual spatial layout and usage requirements.

[0127] The confidence distribution curve of the standard component type matching results provides an important basis for determining the consistency of the matching information between the local image units of each component in the unitized curtain wall to be matched and the standard component library. Through further analysis and processing of the confidence distribution curve, the degree of match between the local image units of the component and each standard component in the standard component library can be accurately determined.

[0128] Step S140: determining matching consistency information between the local image unit of each component in the unitized curtain wall to be matched and the standard component library according to the standard component type matching result.

[0129] Step S141: parsing the standard component type confidence distribution diagram in the standard component type matching result, detecting matching confidence peak points exceeding a preset confidence level and their corresponding standard component type identifiers.

[0130] Analyze the standard component type confidence distribution map in the standard component type matching results. In the standard component type confidence distribution map, each standard component type corresponds to a confidence value, which reflects the likelihood of matching the component's local image unit with the corresponding standard component type. By detecting the peak points in the standard component type confidence distribution map, find the matching confidence peak points that exceed the preset confidence level.

[0131] The preset confidence level is a pre-set threshold used to screen for reliable matches. When the match confidence level for a standard component type exceeds the preset confidence level, the local image unit of the component is considered to have a high probability of matching that standard component type. The standard component type identifiers corresponding to these peak match confidence levels are recorded. These standard component type identifiers are used to uniquely identify each standard component type, facilitating subsequent query and processing.

[0132] Step S142: extracting the reference surface texture features and reference structural contour features of the corresponding standard component from the standard component library according to the standard component type identifier corresponding to the matching confidence peak point.

[0133] Based on the standard component type identifier corresponding to the recorded matching confidence peak point, the corresponding standard component's reference surface texture features and reference structural contour features are extracted from the standard component library. The standard component library stores detailed information on all standard components, including their surface texture features and structural contour features.

[0134] Using the standard component type identifier, the corresponding standard component can be quickly and accurately located, and its reference surface texture features and reference structural contour features can be extracted. These reference surface texture features and reference structural contour features are typical features of the standard component and serve as a reference for comparison with the features of the component's local image units.

[0135] Step S143: calculating a first feature similarity between the surface texture feature of the component local image unit and the reference surface texture feature, wherein the first feature similarity is determined by a matching degree between a texture repetition period parameter and a texture contrast parameter.

[0136] Step S1431: extracting a texture repetition period parameter and a texture contrast parameter from the surface texture feature, and recording them as a first period parameter and a first contrast parameter respectively.

[0137] From the surface texture features of the acquired component's local image units, the texture repetition period parameter and texture contrast parameter are extracted. The texture repetition period parameter reflects the spatial repetition pattern of the texture, while the texture contrast parameter reflects the degree of difference between the bright and dark areas of the texture. By analyzing and processing the surface texture features, these two key parameters are separated, and the texture repetition period parameter is recorded as the first period parameter, and the texture contrast parameter is recorded as the first contrast parameter. This extraction process can be implemented using image processing algorithms and statistical analysis methods. For example, frequency analysis of texture features can obtain the texture repetition period parameter, and distribution statistics of texture grayscale values ​​can obtain the texture contrast parameter.

[0138] Step S1432: extracting a texture repetition period parameter and a texture contrast parameter from the reference surface texture feature, and recording them as a second period parameter and a second contrast parameter respectively.

[0139] Similarly, texture repetition period parameters and texture contrast parameters are extracted from the reference surface texture features extracted from the standard component library. Using the same or similar method as that used to extract the surface texture features of the component's local image units, the texture repetition period parameter and texture contrast parameter are separated from the reference surface texture features and recorded as the second period parameter and second contrast parameter, respectively. This is done to enable subsequent accurate comparison of the surface texture features of the component's local image units with the reference surface texture features.

[0140] Step S1433: analyzing a first matching degree between the first period parameter and the second period parameter, where the first matching degree is determined by similarity of period fluctuation patterns.

[0141] After obtaining the first and second periodic parameters, their periodic fluctuation patterns are analyzed in detail to determine the first degree of match. First, the first and second periodic parameters are considered as sequences that vary with spatial position or other related variables. Spectral analysis is performed on these two sequences, and through methods such as Fourier transform, they are converted from the time domain to the frequency domain to obtain their frequency distributions. The two frequency distributions are compared for features such as shape, peak position, and peak amplitude. If the two frequency distributions have similar shapes, close peak positions, and relatively similar peak amplitudes, then the periodic fluctuation patterns of the first and second periodic parameters are similar, and the first degree of match is high. Conversely, if the frequency distributions differ significantly, such as if the peak positions differ widely or the peak amplitudes differ significantly, then the first degree of match is low.

[0142] Furthermore, we can analyze the phase relationship within periodic sequences. By calculating the phase difference between two periodic sequences, we can determine their temporal or spatial synchronization. A small phase difference indicates good synchronization between the two periodic sequences during their fluctuations, which also improves the primary matching level.

[0143] Step S1434: analyzing a second matching degree between the first contrast parameter and the second contrast parameter, where the second matching degree is determined by similarity in distribution concentration.

[0144] The second degree of matching is determined primarily by analyzing the distribution concentration of the first and second contrast parameters. First, the distribution of the first and second contrast parameters is statistically analyzed, for example, by calculating their mean, variance, standard deviation, and other statistics. The mean reflects the average level of the contrast parameter, while the variance and standard deviation reflect the dispersion of the contrast parameter, that is, the degree of concentration of the distribution.

[0145] Compare the means and variances of the two contrast parameters. If their means are similar, the average contrast levels are similar. If their variances are also similar, the concentration of their distributions is similar, that is, the range of contrast variation is similar. This similarity can be quantified by calculating the area of ​​overlap between the two distributions. Treat the distributions of the first and second contrast parameters as probability density functions and calculate the area of ​​their overlap. The larger the overlapping area, the more similar the concentration of the distributions of the two contrast parameters is, and the higher the second degree of match. Conversely, the smaller the overlapping area, the lower the second degree of match.

[0146] Step S1435: Based on the first matching degree and the second matching degree, comprehensively determine a first feature similarity of the surface texture feature.

[0147] After determining the first and second matching degrees, the two matching degrees need to be combined to determine the first feature similarity of the surface texture features. To achieve this comprehensive evaluation, a weighted average method can be used. First, different weights are assigned to the first and second matching degrees based on the actual situation. If, in actual applications, the texture repetition period parameter is considered more important for matching surface texture features, a higher weight can be assigned to the first matching degree; if the texture contrast parameter is considered more critical, a higher weight can be assigned to the second matching degree.

[0148] Assume that the weight assigned to the first matching degree is W1, the weight assigned to the second matching degree is W2, and W1 + W2 = 1. The first feature similarity S1 can then be calculated using the formula S1 = W1 × first matching degree + W2 × second matching degree. This weighted average approach comprehensively considers the matching of the texture repetition period parameter and the texture contrast parameter, resulting in a first feature similarity value that comprehensively reflects the matching degree of surface texture features.

[0149] Step S144: calculating a second feature similarity between the structural contour feature of the component local image unit and the reference structural contour feature, wherein the second feature similarity is determined by a matching degree of a contour closure integrity parameter and a contour curvature variation parameter.

[0150] Step S1441: extracting the contour closure integrity parameter and the contour curvature variation parameter from the structural contour feature, and recording them as the first closure parameter and the first curvature parameter respectively.

[0151] The contour closure integrity parameter and contour curvature variation parameter are extracted from the structural contour features of the component's local image unit. The contour closure integrity parameter is used to describe whether the contour forms a closed shape, while the contour curvature variation parameter reflects the change in the degree of curvature of the contour at different locations.

[0152] The contour closure integrity parameter can be extracted by examining the positional relationship between the start and end points of the contour boundary line. If the start and end points coincide or are very close, the contour is closed, and a higher value can be assigned to the contour closure integrity parameter. If the start and end points are far apart, the contour may be broken, and a lower value can be assigned to the contour closure integrity parameter.

[0153] To extract the profile curvature variation parameter, differential geometry analysis can be performed on the profile boundary. The curvature of each point on the profile boundary is calculated; the curvature represents the degree of curvature at that point. By performing statistical analysis on the curvature values ​​along the entire profile boundary, such as calculating the mean and variance of the curvature, a first curvature parameter reflecting the profile curvature variation is obtained.

[0154] Step S1442: extracting the contour closure integrity parameter and the contour curvature variation parameter from the reference structure contour feature, and recording them as the second closure parameter and the second curvature parameter respectively.

[0155] Using the same method used to extract the structural contour features of the component's local image unit, extract the contour closure integrity parameter and contour curvature variation parameter from the reference structural contour features, denoted as the second closure parameter and second curvature parameter, respectively. Ensure that the same standards and algorithms are used during the extraction process to facilitate accurate comparison later.

[0156] Step S1443: analyzing a third matching degree between the first closure parameter and the second closure parameter, wherein the third matching degree is determined by the similarity between the start and end position relationships of the contour boundary line.

[0157] When analyzing the third degree of match between the first and second closure parameters, focus on the similarity between the start and end positions of the contour boundaries. First, compare the positions of the starting points of the two contour boundaries. The distance between the starting points can be calculated. A small distance indicates that the starting points are close in position. Similarly, compare the positions of the ending points of the two contour boundaries and calculate the distance between them.

[0158] Next, the relative positional relationship between the starting and ending points is comprehensively considered. Features such as the angle and length ratio of the line connecting the starting and ending points can be calculated. If the relative positional relationships between the starting and ending points of the two contour boundaries are similar, for example, the angle and length ratio of the connecting lines are similar, then the third match degree is high. Conversely, if the positions of the starting and ending points are significantly different and the relative positional relationships are also different, then the third match degree is low.

[0159] Step S1444: analyzing a fourth matching degree between the first curvature parameter and the second curvature parameter, wherein the fourth matching degree is determined by the similarity of the changing rules of the curvature degree of the contour boundary line.

[0160] The fourth degree of matching is determined by analyzing the similarity between the first and second curvature parameters in the curvature variation patterns of the contour boundary line. First, the first and second curvature parameters are considered as sequences that vary with the position of the contour boundary line. These two sequences are then curve-fitted, for example using a polynomial fit or a spline curve fit, to obtain a curve that describes the curvature variation pattern.

[0161] Compare the two fitted curves based on their shape, slope, and curvature trends. If the two curves have similar shapes, consistent slope trends, and similar curvature increases and decreases, then the first and second curvature parameters reflect similar curvature patterns, indicating a high fourth degree of fit. Conversely, if the curves have significantly different shapes, and different slope and curvature trends, then the fourth degree of fit is low.

[0162] You can also calculate the distance between two curves, for example using Euclidean distance or Manhattan distance, to quantify the degree of difference between them. The smaller the distance, the more similar the two curves are, and the higher the degree of match.

[0163] Step S1445: Based on the third matching degree and the fourth matching degree, comprehensively determine the second feature similarity of the structural contour feature.

[0164] After determining the third and fourth matching degrees, the second feature similarity of the structural profile feature is determined by combining these two matching degrees. Similarly, a weighted average method is used to assign different weights to the third and fourth matching degrees. Assume that the weight assigned to the third matching degree is W3, and the weight assigned to the fourth matching degree is W4, and that W3 + W4 = 1.

[0165] The second feature similarity S2 can be calculated using the formula S2 = W3 × third matching degree + W4 × fourth matching degree. In this way, the matching of the contour closure integrity parameter and the contour curvature change parameter is comprehensively considered to obtain a second feature similarity value that can fully reflect the matching degree of the structural contour features.

[0166] Step S145: Based on the first feature similarity and the second feature similarity, a weighted synthesis method is used to calculate the comprehensive matching consistency between the component local image unit and the standard component, and the comprehensive matching consistency is used to represent the matching consistency information between the component local image unit and the standard component library.

[0167] Based on the first feature similarity and the second feature similarity, a weighted synthesis method is used to calculate the comprehensive matching consistency between the component local image unit and the standard component. In the weighted synthesis process, different weights are assigned to the first feature similarity and the second feature similarity, and then they are weighted combined.

[0168] The weight assignment can be adjusted based on actual conditions. For example, if surface texture features are considered more important in the matching process, a higher weight can be assigned to the first feature similarity; if structural contour features are considered more important in the matching process, a higher weight can be assigned to the second feature similarity. The comprehensive matching consistency value obtained through weighted combination can fully reflect the degree of matching between the component local image unit and the corresponding standard component in the standard component library, and thus serve as the matching consistency information between the component local image unit and the standard component library.

[0169] Step S150: generating a matching optimization instruction including a component replacement or adjustment instruction based on the matching consistency information, and feeding back the matching optimization instruction to the curtain wall construction management system to trigger a component configuration adjustment operation.

[0170] Step S151: determining a matching level corresponding to the component local image unit according to the comprehensive matching consistency in the matching consistency information, where the matching level includes complete matching, partial matching, and non-matching.

[0171] After obtaining the comprehensive matching consistency between the component's local image unit and the standard component, the matching level corresponding to the component's local image unit needs to be determined based on a preset threshold. First, two thresholds are set: a high threshold and a low threshold. The high threshold is used to distinguish between a complete match and a partial match, while the low threshold is used to distinguish between a partial match and a mismatch.

[0172] When the overall matching consistency exceeds the high threshold, the component's local image unit is considered a complete match. This means that the component's surface texture and structural profile characteristics are highly consistent with those of the corresponding standard component in the standard component library and can be used directly in curtain wall construction without additional adjustments.

[0173] When the overall matching consistency falls between the high and low thresholds, it is considered a partial match. The characteristics of the component's local image units at this level are somewhat similar to those of the standard component, but there may be some subtle differences that require appropriate adjustments to meet the requirements of curtain wall construction.

[0174] When the overall matching consistency falls below the low threshold, it is considered mismatched. The characteristics of such components differ significantly from those of standard components and cannot be directly used in curtain wall construction, requiring replacement.

[0175] Step S152: For the component partial image unit whose matching level is complete match, generate indication information for retaining the current component configuration.

[0176] When the matching level of a component's local image unit is fully matched, it is necessary to generate information indicating that the current component configuration should be retained. This information includes the component's location identifier and operation type identifier. The location identifier is used to specify the component's specific location within the curtain wall and corresponds one-to-one with the spatial location marker in the on-site image data set; the operation type identifier specifies that the component's operation type is retained.

[0177] The specific generation process is as follows: First, the component's location identifier is extracted from the matching consistency information. Then, the operation type identifier "Retain" is added to the instruction information. Finally, the location identifier and operation type identifier are combined into a complete instruction message, such as "Location identifier: [specific location], Operation type: Retain." This instruction clearly informs the curtain wall construction management system that the component can be installed in its current position and state, helping to improve construction efficiency and avoid unnecessary operations.

[0178] Step S153: For the component partial image unit whose matching level is partial matching, generate instruction information for adjusting the current component installation position or angle.

[0179] For component partial image units with a partial match level, instructions for adjusting the current component installation position or angle need to be generated. Before generating the instructions, the differences between the characteristics of the component partial image unit and the standard component characteristics need to be further analyzed to determine the specific adjustment direction and degree.

[0180] For installation position adjustments, the component's horizontal and vertical offsets are determined by comparing the structural profile features of the component's local image units with those of standard components. For example, if the component's profile is found to be offset to the left by a certain distance in the horizontal direction, the rightward adjustment distance must be clearly stated in the instructions; if it is offset upward in the vertical direction, the downward adjustment distance must be specified.

[0181] Adjustment of the installation angle is determined by analyzing the directional differences between surface texture features and structural contour features. Image processing algorithms, such as feature point matching and rotation angle calculation, can be used to determine the required component rotation angle. For example, if the component's grain direction differs from that of the standard component, the rotation angle and direction (clockwise or counterclockwise) must be clearly stated in the instructions.

[0182] The generated instructions also include the component's location and operation type. The operation type specifies that the component's operation type is adjustment, and the instructions also detail the specific adjustment details, for example, "Location: [specific location], Operation type: Adjustment, Installation position: Horizontally adjust right by [X] distance, vertically adjust downward by [Y] distance; Installation angle: Rotate clockwise by [Z] degrees."

[0183] Step S154: For the component partial image unit whose matching level is mismatch, generate instruction information for replacing it with the corresponding standard component type.

[0184] When the matching level of the component partial image unit is mismatched, it is necessary to generate instruction information for replacing with the corresponding standard component type. The instruction information includes the component position identifier, operation type identifier and replacement standard component type identifier.

[0185] First, the component location identifier is extracted from the matching consistency information. Then, based on the standard component type matching results, the standard component type to be replaced is determined and its corresponding standard component type identifier is obtained. Finally, the operation type identifier "Replace" is added to the instruction information.

[0186] An example of the generated instruction is as follows: "Location ID: [specific location], Operation Type: Replace, Replacement Standard Component Type ID: [specific ID]." This instruction clearly informs the curtain wall construction management system that the component needs to be replaced with a suitable standard component from the inventory.

[0187] Step S155: Fusion the instruction information for retaining the current component configuration, the instruction information for adjusting the current component installation position or angle, and the instruction information for replacing with the corresponding standard component type, to generate a matching optimization instruction including a component position identifier and an operation type identifier, wherein the component position identifier corresponds one-to-one to the spatial position mark in the on-site image data set.

[0188] The generated instructions for retaining the current component configuration, adjusting the current component installation position or angle, and replacing it with the corresponding standard component type are combined to generate matching optimization instructions. The fusion process organizes and merges these instructions according to the set rules to form a unified instruction set.

[0189] During the integration process, the integrity and accuracy of each indication must be ensured. For duplicate location identifiers, only the most recent indication is retained. Furthermore, the indications are sorted, for example, by location identifier order, to facilitate processing within the curtain wall construction management system.

[0190] The resulting matching optimization instructions contain the location and operation type identifiers for all components, and the location identifiers correspond one-to-one with the spatial location markers in the on-site image data set. This matching optimization instruction clearly and accurately informs the curtain wall construction management system of the treatment method and specific location of each component, facilitating component configuration adjustments.

[0191] The generated matching optimization instructions are fed back to the curtain wall construction management system. Upon receiving the matching optimization instructions, the system triggers the corresponding component configuration adjustment operations based on the component position and operation type identifiers in the instructions. For components that need to be retained, the system ensures their installation according to the original plan. For components that require adjustment, the system instructs construction personnel to adjust the installation position or angle. For components that need to be replaced, the system arranges for the selection of appropriate standard components from inventory for replacement. This approach achieves precise matching and optimized configuration of unitized curtain wall components, improving the quality and efficiency of curtain wall construction.

[0192] Furthermore, the training process of the above component matching model is as follows:

[0193] First, collect a large amount of image data of unitized curtain wall components as training data. This image data should cover a variety of standard components of different types and specifications, as well as some possible non-standard components, to ensure that the model can learn rich feature information. At the same time, annotate each image data with the corresponding standard component type label to supervise model training. During the data collection process, attention should be paid to data diversity and representativeness, such as collecting data from different curtain wall projects and different construction environments to improve the model's generalization ability.

[0194] The collected image data is then preprocessed, including grayscale conversion and noise filtering, similar to the preprocessing steps for the field image data set. Grayscale conversion converts color images to grayscale, reducing color interference; noise filtering removes noise and improves image quality. Furthermore, the image can be normalized to adjust pixel values ​​to an appropriate range, such as to the interval [0, 1], to facilitate model training convergence.

[0195] At the same time, a component matching model is constructed, which should include the necessary modules in the above step S130. Thus, the pre-processed image data can be input into the constructed component matching model for training. During the training process, the labeled standard component type labels are used as supervisory signals, and the difference between the prediction results of the component matching model and the true labels is measured by the loss function. Commonly used loss functions include cross entropy loss function, etc. Optimization algorithms such as stochastic gradient descent (SGD) and adaptive moment estimation (Adam) are used to update the parameters of the model so that the value of the loss function continues to decrease and the prediction performance of the component matching model continues to improve. The training process usually requires multiple rounds, and all the training data are input into the component matching model for training once in each round until the performance of the model reaches a satisfactory level.

[0196] Evaluate the trained component matching model using a portion of test data that was not used in training. Evaluation metrics can include precision, recall, and F1 value, which measure the performance of the component matching model on unseen data. If the evaluation results of the component matching model are unsatisfactory, adjust the model, such as adjusting its structure or adding more training data, and then retrain and reevaluate until the performance meets the requirements.

[0197] Figure 2 A schematic diagram illustrates exemplary hardware and software components of a unitized curtain wall component matching system 100 combined with AI recognition, which can implement the concepts of the present application, as provided in some embodiments of the present application. For example, the processor 120 can be used in the unitized curtain wall component matching system 100 combined with AI recognition and used to perform the functions of the present application.

[0198] The unitized curtain wall component matching system 100 combined with AI recognition can be a general-purpose server or a special-purpose server, both of which can be used to implement the unitized curtain wall component matching method combined with AI recognition of the present application. Although only one server is shown in this application, for convenience, the functions described in this application can be implemented in a distributed manner on multiple similar platforms to balance the processing load.

[0199] For example, the unitized curtain wall component matching system 100 combined with AI recognition may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and different forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the unitized curtain wall component matching system 100 combined with AI recognition may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The method of the present application can be implemented according to these program instructions. The unitized curtain wall component matching system 100 combined with AI recognition also includes an I / O interface 150 between the computer and other input and output devices.

[0200] For ease of explanation, only one processor is described in the unitized curtain wall component matching system 100 combined with AI recognition. However, it should be noted that the unitized curtain wall component matching system 100 combined with AI recognition in the present application may also include multiple processors, so the steps performed by one processor described in the present application may also be performed jointly or individually by multiple processors. For example, if the processor of the unitized curtain wall component matching system 100 combined with AI recognition executes step A and step B, it should be understood that step A and step B may also be executed jointly by two different processors or individually in one processor. For example, the first processor executes step A, the second processor executes step B, or the first processor and the second processor execute steps A and B together.

[0201] In addition, an embodiment of the present invention further provides a readable storage medium, in which computer-executable instructions are preset. When a processor executes the computer-executable instructions, the unitized curtain wall component matching method combined with AI recognition as described above is implemented.

[0202] It should be noted that in order to simplify the description of the present invention and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present invention, multiple features are sometimes combined into one embodiment, figure or description thereof.

Claims

1. A unitized curtain wall component matching method combined with AI recognition, characterized in that: The method comprises: Acquire a set of on-site image data of a unitized curtain wall to be matched, wherein the set of on-site image data includes a plurality of component local image units with spatial position marks; Performing image feature extraction processing on the on-site image data set to obtain surface texture features and structural contour features of the local image units of the component; Calling a pre-trained component matching model to perform a joint matching analysis on the surface texture features and the structural contour features to generate a standard component type matching result corresponding to the component local image unit; Determining matching consistency information between the local image unit of each component in the unitized curtain wall to be matched and the standard component library according to the standard component type matching result; generating a matching optimization instruction including a component replacement or adjustment instruction based on the matching consistency information, and feeding the matching optimization instruction back to the curtain wall construction management system to trigger a component configuration adjustment operation; The calling of the pre-trained component matching model to perform a joint matching analysis on the surface texture feature and the structural contour feature to generate a standard component type matching result corresponding to the component local image unit includes: Inputting the texture distribution law of the surface texture feature and the contour morphological feature of the structural contour feature into the feature alignment module of the component matching model, the feature alignment module performs coordinate alignment processing on the texture distribution law and the contour morphological feature based on pre-trained spatial mapping parameters, so that the key area of ​​the texture distribution corresponds to the boundary area of ​​the contour morphology in one-to-one spatial position, and generating a dual-feature alignment group with a positional correspondence relationship; Inputting the dual-feature alignment group into the context fusion layer of the component matching model, extracting the spatial position mark of the component local image unit through the context fusion layer, and obtaining the structural topology information corresponding to the spatial position mark from the standard component library, and generating context enhanced features containing the curtain wall structure position information by feature association encoding the structural topology information and the dual-feature alignment group; Inputting the context-enhanced features into the constraint filtering layer of the component matching model, extracting standard constraints corresponding to the installation position based on the structural topology information through the constraint filtering layer, and generating constraint-enhanced features focusing on key installation features by filtering redundant features in the context-enhanced features that do not meet the standard constraints; Inputting the constraint enhancement features into the feature aggregation layer of the component matching model, and performing cross-dimensional weighted fusion of texture distribution patterns, contour morphological features, and structural position information in the constraint enhancement features through the feature aggregation layer to generate a spatial enhancement feature vector that can comprehensively reflect the relationship between component features and installation scenarios; The classification output layer of the component matching model is called to perform standard component type confidence prediction processing on the spatial enhancement feature vector to generate a standard component type matching result including a standard component type confidence distribution map, wherein the standard component type confidence distribution map is used to represent the matching confidence corresponding to different standard component types.

2. The unitized curtain wall component matching method combined with AI recognition according to claim 1 is characterized in that: The performing image feature extraction processing on the on-site image data set to obtain surface texture features and structural contour features of the local image unit of the component includes: Performing an image preprocessing operation on the on-site image data set to obtain a preprocessed component local image unit, wherein the image preprocessing operation includes grayscale processing and noise filtering processing; Performing texture feature capture processing on the pre-processed local image unit of the component, and extracting texture distribution regularity characteristics of the local image unit of the component as surface texture features using a local binary pattern algorithm, wherein the surface texture features include a texture repetition period parameter and a texture contrast parameter; Performing contour edge detection processing on the preprocessed component local image unit, using an edge detection algorithm to identify contour boundary information of the component local image unit, and generating structural contour features based on the contour boundary information, wherein the structural contour features include a contour closure integrity parameter and a contour curvature change parameter; Performing correlation analysis on the surface texture features and the structural contour features across feature dimensions to obtain a set of correlation features with a unified feature space representation; Based on the importance evaluation results of different feature dimensions in the associated feature set, the feature fusion weight parameters of the surface texture feature and the structural contour feature in the joint matching analysis are dynamically adjusted.

3. The unitized curtain wall component matching method combined with AI recognition according to claim 2 is characterized in that: The step of performing texture feature capture processing on the pre-processed local image unit of the component and extracting the texture distribution regularity feature of the local image unit of the component as the surface texture feature by using a local binary pattern algorithm includes: Dividing the preprocessed component local image unit into a plurality of non-overlapping texture analysis window units, each texture analysis window unit containing a fixed number of pixels; For each texture analysis window unit, the grayscale value of the center pixel of the texture analysis window unit is used as a reference value, and the magnitude relationship between other pixels in the texture analysis window unit and the reference value is calculated to generate a binary code sequence; Converting the binary code sequence into a decimal value as a local binary pattern feature value of the texture analysis window unit; Counting the occurrence frequencies of local binary pattern eigenvalues ​​of all texture analysis window units and generating a histogram distribution of the local binary pattern eigenvalues; The texture repetition period parameter and texture contrast parameter of the local image unit of the component are determined based on the histogram distribution, the texture repetition period parameter is determined by the periodic fluctuation law of the eigenvalues ​​in the histogram distribution, and the texture contrast parameter is determined by the distribution concentration degree of the eigenvalues ​​in the histogram distribution.

4. The unitized curtain wall component matching method combined with AI recognition according to claim 2, characterized in that: The step of performing contour edge detection on the pre-processed component local image unit, identifying contour boundary information of the component local image unit using an edge detection algorithm, and generating a structural contour feature based on the contour boundary information includes: Performing smoothing filtering on the preprocessed local image unit of the component, and calculating the gradient magnitude and gradient direction of each pixel in the smoothed image, wherein the gradient magnitude is calculated by the convolution operation results in the horizontal and vertical directions, and the gradient direction is calculated by the relative relationship between the horizontal and vertical gradients; Performing non-maximum suppression processing on the gradient amplitude, retaining the local maximum pixel points in the gradient direction, and obtaining a set of candidate contour edge pixel points; Performing edge connection processing on the candidate contour edge pixel point set using a double threshold detection method to generate a continuous contour boundary line as the contour boundary information; Based on the continuous contour boundary line, the contour closure integrity parameter and the contour curvature change parameter of the local image unit of the component are determined as the structural contour feature. The contour closure integrity parameter is determined by the positional relationship between the starting point and the end point of the contour boundary line, and the contour curvature change parameter is determined by the change law of the curvature degree of each point on the contour boundary line.

5. The unitized curtain wall component matching method combined with AI recognition according to claim 1, characterized in that: The process of extracting the spatial position mark of the component local image unit through the context fusion layer, obtaining structural topology information corresponding to the spatial position mark from a standard component library, and generating context enhancement features containing the curtain wall structural position information by performing feature association encoding on the structural topology information and the dual-feature alignment group includes: Obtaining a spatial position mark of the local image unit of the component, wherein the spatial position mark is directly obtained from the spatial position mark in the on-site image data set and is used to indicate the specific installation coordinates of the component in the entire curtain wall; Querying structural topology information corresponding to the spatial position mark from a standard component library, the structural topology information including the horizontal connection level and vertical adjacent component types of the spatial position mark in the curtain wall; Performing feature association coding on the surface texture features in the dual-feature alignment group and the lateral connection levels in the structural topology information to generate hierarchical association features reflecting the adaptability of texture distribution and connection levels, wherein the feature association coding is performed by learning typical texture distribution patterns corresponding to different connection levels in historical installation data; Performing feature association coding on the structural contour features in the dual-feature alignment group and the longitudinally adjacent component types in the structural topology information to generate adjacency association features reflecting the matching between the contour morphology and the adjacent components, wherein the feature association coding is performed by learning typical contour morphology patterns corresponding to different adjacent component types in historical installation data; The hierarchical association features and adjacency association features are dynamically weighted by the attention mechanism module of the context fusion layer to generate context-enhanced features containing curtain wall structure position information. The weight allocation is determined based on the degree of influence of different connection levels and adjacent component types in historical installation data on component matching results.

6. The method for matching unitized curtain wall components combined with AI recognition according to claim 1, characterized in that: The determining, based on the standard component type matching result, matching consistency information between the local image unit of each component in the unitized curtain wall to be matched and the standard component library includes: parsing a standard component type confidence distribution diagram in the standard component type matching result, detecting a matching confidence peak point exceeding a preset confidence level and its corresponding standard component type identifier; Extracting the reference surface texture features and reference structural contour features of the corresponding standard component from the standard component library according to the standard component type identifier corresponding to the matching confidence peak point; Calculating a first feature similarity between a surface texture feature of a local image unit of the component and a reference surface texture feature, wherein the first feature similarity is determined by a matching degree between a texture repetition period parameter and a texture contrast parameter; Calculating a second feature similarity between a structural contour feature of the component local image unit and a reference structural contour feature, wherein the second feature similarity is determined by a matching degree of a contour closure integrity parameter and a contour curvature change parameter; Based on the first feature similarity and the second feature similarity, a weighted synthesis method is used to calculate the comprehensive matching consistency between the component local image unit and the standard component. The comprehensive matching consistency is used to represent the matching consistency information between the component local image unit and the standard component library.

7. The unitized curtain wall component matching method combined with AI recognition according to claim 6, characterized in that: The calculating the first feature similarity between the surface texture feature of the component local image unit and the reference surface texture feature includes: Extracting a texture repetition period parameter and a texture contrast parameter from the surface texture feature, and recording them as a first period parameter and a first contrast parameter respectively; Extracting a texture repetition period parameter and a texture contrast parameter from the reference surface texture feature, and recording them as a second period parameter and a second contrast parameter respectively; analyzing a first matching degree between the first period parameter and the second period parameter, the first matching degree being determined by similarity of period fluctuation patterns; analyzing a second matching degree between the first contrast parameter and the second contrast parameter, wherein the second matching degree is determined by similarity of distribution concentration; comprehensively determining a first feature similarity of the surface texture feature based on the first matching degree and the second matching degree; Furthermore, the calculating of the second feature similarity between the structural contour feature of the component local image unit and the reference structural contour feature comprises: Extracting a contour closure integrity parameter and a contour curvature change parameter from the structural contour feature, which are recorded as a first closure parameter and a first curvature parameter, respectively; Extracting a contour closure integrity parameter and a contour curvature variation parameter from the reference structure contour feature, and recording them as a second closure parameter and a second curvature parameter, respectively; analyzing a third degree of matching between the first closure parameter and the second closure parameter, the third degree of matching being determined by similarity between a start position and an end position of a contour boundary line; analyzing a fourth matching degree between the first curvature parameter and the second curvature parameter, wherein the fourth matching degree is determined by similarity in a changing pattern of curvature degrees of contour boundary lines; Based on the third matching degree and the fourth matching degree, a second feature similarity of the structural contour feature is comprehensively determined.

8. The method for matching unitized curtain wall components combined with AI recognition according to claim 1, characterized in that: The generating of the matching optimization instruction including the component replacement or adjustment instruction based on the matching consistency information includes: Determining a matching level corresponding to the component local image unit according to the comprehensive matching consistency in the matching consistency information, wherein the matching level includes complete matching, partial matching, and non-matching; For the component local image unit whose matching level is completely matched, generating instruction information for retaining the current component configuration; For the component local image unit whose matching level is partial matching, generating instruction information for adjusting the installation position or angle of the current component; For a component partial image unit whose matching level is mismatched, generating instruction information for replacing it with a corresponding standard component type; The instruction information for retaining the current component configuration, the instruction information for adjusting the current component installation position or angle, and the instruction information for replacing with the corresponding standard component type are integrated to generate a matching optimization instruction including a component position identifier and an operation type identifier, wherein the component position identifier corresponds one-to-one to the spatial position mark in the on-site image data set.

9. A unitized curtain wall component matching system combined with AI recognition, characterized in that: It includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the unit curtain wall component matching method combined with AI recognition as described in any one of claims 1 to 8.

Images