CN-121582196-B - Hidden frame glass curtain wall structural adhesive aging nondestructive testing method based on visual detection

CN121582196BCN 121582196 BCN121582196 BCN 121582196BCN-121582196-B

Abstract

The application provides a visual detection-based hidden frame glass curtain wall structural adhesive aging nondestructive detection method, which comprises the steps of obtaining hyperspectral image cubes on the surface of structural adhesive to be detected and end member spectrum characteristic information containing aging product components. And carrying out spectral decomposition on the hyperspectral image according to the characteristic information, and calculating the abundance value of each target chemical component at each pixel point to generate a material component abundance map. And clustering the abundance map to obtain an aging area cluster formed by pixel points with abnormal abundance of chemical components and adjacent space, and determining the position of the aging area cluster. And determining the regional component vector according to the component abundance vector of all the pixel points in the aging regional cluster, and inputting the regional component vector as an input vector into an aging degree evaluation model to obtain an aging degree index. And generating an aging state diagram of the structural adhesive based on the position information and the aging degree index of all the aging area clusters. The method can effectively identify the early aging of the structural adhesive and improve the aging evaluation accuracy.

Inventors

YANG TINGHAI
ZHAO XIN
JIANG ZHONGCHEN
HUO XIANGMING
Xu shuanglong
WANG SHAOHONG
LI MIAO
LUO WENFENG
JING QICHENG
ZHOU CHI
XIA JINLONG
JIANG HAIFENG
ZHAO PENG

Assignees

北京佑荣索福恩建筑咨询有限公司

Dates

Publication Date: 20260508
Application Date: 20251125

Claims (10)

1. The hidden frame glass curtain wall structural adhesive aging nondestructive testing method based on visual detection is characterized by comprising the following steps of: acquiring hyperspectral image cubes and end member spectrum characteristic information of the surface of the structural adhesive to be detected of the hidden frame glass curtain wall, wherein the end member spectrum characteristic information comprises spectrum characteristics of a plurality of target chemical components in the structural adhesive, and the plurality of target chemical components comprise at least one adhesive aging product component; according to the end member spectral characteristic information, spectral decomposition is carried out on the spectral signal of each pixel point of the hyperspectral image cube, the abundance value of each target chemical component in each pixel point is calculated, the component abundance vector of each pixel point is obtained, and a multichannel material component abundance diagram is generated; Carrying out spatial clustering on all pixel points in the material component abundance map through a spatial density clustering algorithm to obtain an aging area cluster formed by pixel points with abnormal chemical component abundance and adjacent spatial positions, and determining the position information of the aging area cluster on the surface of the structural adhesive to be detected; Determining a regional component vector of the aging region cluster according to the component abundance vectors of all pixel points in the aging region cluster, and inputting the regional component vector serving as an input vector into a trained aging degree evaluation model to obtain an aging degree index of the aging region cluster; and generating a structural adhesive aging state diagram of the structural adhesive to be detected based on the position information and the aging degree index of all the aging area clusters.
2. The method of claim 1, wherein the plurality of target chemical components further comprises at least one surface contaminant component, the method further comprising: Calculating the pollution degree value of each pixel point of the material component abundance diagram based on the abundance value of the channel corresponding to the surface pollutant component in the component abundance vector of each pixel point; calculating pollution regulation factors of each aging area cluster based on pollution degree values of all pixel points in the aging area cluster and a preset pollution threshold interval; after the area component vector is input into the trained aging degree evaluation model as an input vector to obtain the aging degree index of the aging area cluster, the method further comprises: Updating the ageing degree index according to the pollution regulation factor to obtain the updated ageing degree index.
3. The method according to claim 1, wherein the method further comprises: obtaining the fractal dimension of each aging region cluster by calculating the box counting dimension of the boundary contour of the aging region cluster; the step of inputting the regional component vector as an input vector into a trained aging degree evaluation model to obtain an aging degree index of the aging regional cluster, comprising the following steps: Constructing an input vector according to a preset sequence based on the regional component vector and the fractal dimension; And inputting the input vector into a trained aging degree evaluation model to obtain the aging degree index of the aging region cluster.
4. A method according to claim 3, wherein before said inputting said input vector into a trained aging assessment model to derive an aging index for said aging region cluster, said method comprises: Obtaining a training sample set, wherein the training sample set comprises a plurality of training samples, and each training sample comprises a historical input vector generated by historical hyperspectral data and historical end member spectral characteristic information and a corresponding real aging degree index; Respectively executing the steps of inputting the historical input vector in each training sample into a preset aging degree evaluation model to obtain a predicted aging degree index; Determining a loss function value of the aging degree evaluation model according to the real aging degree index label and the predicted aging degree index of each training sample; and under the condition that the loss function value does not meet the preset training stop condition, adjusting model parameters of the aging degree evaluation model to obtain an updated aging degree evaluation model, and returning to input the historical input vector into the aging degree evaluation model to obtain a predicted aging degree index until the loss function value meets the training stop condition to obtain the trained aging degree evaluation model.
5. The method of claim 1, wherein the generating a multi-channel material component abundance map by spectrally decomposing the spectral signal of each pixel of the hyperspectral image cube according to the end-member spectral feature information and calculating the abundance value of each target chemical component at each pixel to obtain a component abundance vector of each pixel comprises: Constructing a standard spectrum matrix based on the end member spectrum characteristic information, wherein each column of the standard spectrum matrix is a spectrum characteristic of the target chemical component, and the column sequence of the standard spectrum matrix is preset according to the appearance sequence of the target chemical component in the aging evolution process of the structural adhesive; the spectrum signal of each pixel point in the hyperspectral image cube is combined with the standard spectrum matrix, and a linear mixing equation taking the abundance value of each target chemical component as an unknown number is solved through a non-negative constraint least square method, so that an initial abundance vector is obtained; based on each initial abundance vector, normalizing the initial abundance vector by using a constraint obtained by summation to obtain the component abundance vector of each pixel point; Combining the component abundance vectors of all the pixel points, and mapping element values in each component abundance vector into one channel of the material component abundance map according to the column sequence of the standard spectrum matrix to generate the material component abundance map of the multiple channels.
6. The method according to claim 1, wherein the spatially clustering all the pixels in the material component abundance map by a spatial density clustering algorithm to obtain an aging region cluster composed of pixels with abnormal chemical component abundance and adjacent spatial positions, and determining the position information of the aging region cluster on the surface of the structural adhesive to be detected, includes: extracting an abundance value corresponding to each colloid aging product component in the component abundance vector of each pixel point in the material component abundance diagram to form an aging abundance vector of each pixel point; Traversing all the pixel points, and determining the pixel points, the numerical average value of the aging abundance vector of which is larger than a preset abundance threshold value and the number of the pixel points in a preset neighborhood radius of which is larger than a preset number threshold value, as core pixel points; starting from any unclassified core pixel point, creating an initial pixel point set, and adding the core pixel point into the pixel point set; Performing iterative expansion based on the pixel point set, adding pixels which are in the neighborhood radius of all the core pixels in the pixel point set and are not classified into the pixel point set in each iteration, repeating the iterative expansion until no new pixel point is added into the pixel point set in one complete iteration, and determining the pixel point set after the iterative expansion is completed as one aging area cluster; And determining the position information of the aging area clusters based on the space coordinates of the pixel points in each aging area cluster.
7. The method of claim 1, wherein said determining a regional component vector for the aged regional cluster from the component abundance vector for all pixels within the aged regional cluster comprises: vector average value calculation is carried out on the component abundance vectors of all pixel points in each aging area cluster, so that average component vectors of the aging area clusters are obtained; Extracting abundance values corresponding to each colloid aging product component from the component abundance vectors of each pixel point in each aging area cluster, and calculating the variance of the abundance values corresponding to each colloid aging product component to obtain aging abundance variances of each colloid aging product component; Combining the average component vector with all of the aging abundance variances to construct the regional component vectors of the aging regional clusters.
8. Hidden frame glass curtain wall structure adhesive aging nondestructive testing system based on visual detection, which is characterized by comprising: The device comprises an acquisition module, a detection module and a detection module, wherein the acquisition module is used for acquiring hyperspectral image cubes and end member spectral characteristic information of the surface of structural adhesive to be detected of a hidden frame glass curtain wall, the end member spectral characteristic information comprises spectral characteristics of a plurality of target chemical components in the structural adhesive, and the plurality of target chemical components comprise at least one adhesive aging product component; The generating module is used for carrying out spectrum decomposition on the spectrum signal of each pixel point of the hyperspectral image cube according to the end member spectrum characteristic information, calculating the abundance value of each target chemical component at each pixel point, obtaining the component abundance vector of each pixel point and generating a multichannel material component abundance diagram; The clustering module is used for carrying out spatial clustering on all pixel points in the material component abundance diagram through a spatial density clustering algorithm to obtain an aging area cluster formed by pixel points with abnormal chemical component abundance and adjacent spatial positions, and determining the position information of the aging area cluster on the surface of the structural adhesive to be detected; The evaluation module is used for determining an area component vector of the aging area cluster according to the component abundance vectors of all pixel points in the aging area cluster, and inputting the area component vector serving as an input vector into a trained aging degree evaluation model to obtain an aging degree index of the aging area cluster; the generation module is further configured to generate a structural adhesive aging state diagram of the structural adhesive to be detected based on the position information and the aging degree index of all the aging area clusters.
9. An electronic device, comprising: A memory for storing a computer program; A processor for implementing the steps of the visual inspection-based hidden frame glass curtain wall structural adhesive aging nondestructive inspection method according to any one of claims 1 to 7 when executing the computer program.
10. A computer readable storage medium, wherein a computer program is stored in the computer readable storage medium, and when the computer program is executed by a processor, the computer program can implement the hidden frame glass curtain wall structural adhesive aging nondestructive testing method based on visual detection as set forth in any one of claims 1 to 7.

Description

Hidden frame glass curtain wall structural adhesive aging nondestructive testing method based on visual detection Technical Field The application belongs to the technical field of computer vision, and particularly relates to a hidden frame glass curtain wall structural adhesive aging nondestructive testing method based on vision detection. Background The hidden frame glass curtain wall is used as a peripheral protection structure of a modern building, and the safety and durability of the curtain wall are directly affected by the adhesive property of structural adhesive. The structural adhesive is aged under the influence of environmental factors in the long-term service process, so that the adhesive property is reduced, and therefore, an effective nondestructive testing method is required to evaluate the state of the structural adhesive. In the prior art, a glass panel degumming identification method based on image processing is mainly adopted for detecting the structural colloid state. The method is characterized in that color images of structural adhesive areas are collected, color changes or texture features of the adhesive surfaces are analyzed through an image processing algorithm, and areas possibly with degumming or aging are identified. However, prior art detection methods based on apparent image features can only identify physical degumping of apparent appearance changes, and are insensitive to early intrinsic aging of structural adhesives. Meanwhile, the method is easily interfered by factors such as illumination conditions, surface pollution and the like, and the performance degradation degree of the colloid is difficult to accurately evaluate, so that the early aging identification capability is limited. Therefore, the technical problem that the early aging of the hidden frame glass curtain wall structural adhesive cannot be effectively identified and the aging degree evaluation is inaccurate exists in the prior art. Disclosure of Invention The application aims to provide a hidden frame glass curtain wall structural adhesive aging nondestructive testing method, a system, electronic equipment and a storage medium based on visual detection, which are used for solving the problems that the prior art cannot effectively identify the early aging of the hidden frame glass curtain wall structural adhesive and the aging degree evaluation is inaccurate. In order to solve the technical problems, in a first aspect, the application provides a hidden frame glass curtain wall structural adhesive aging nondestructive testing method based on visual detection, which comprises the following steps: Acquiring hyperspectral image cubes and end member spectral characteristic information of the surface of the structural adhesive to be detected of the hidden frame glass curtain wall, wherein the end member spectral characteristic information comprises spectral characteristics of a plurality of target chemical components in the structural adhesive, and the plurality of target chemical components comprise at least one adhesive aging product component; According to the end member spectral characteristic information, spectral decomposition is carried out on the spectral signal of each pixel point of the hyperspectral image cube, the abundance value of each target chemical component in each pixel point is calculated, the component abundance vector of each pixel point is obtained, and a multichannel material component abundance diagram is generated; carrying out spatial clustering on all pixel points in a material component abundance diagram through a spatial density clustering algorithm to obtain an aging area cluster formed by pixel points with abnormal chemical component abundance and adjacent spatial positions, and determining the position information of the aging area cluster on the surface of the structural adhesive to be detected; determining a regional component vector of the aging region cluster according to the component abundance vectors of all pixel points in the aging region cluster, and inputting the regional component vector serving as an input vector into a trained aging degree evaluation model to obtain an aging degree index of the aging region cluster; And generating a structural adhesive aging state diagram of the structural adhesive to be detected based on the position information and the aging degree index of all the aging area clusters. In one achievable embodiment, the plurality of target chemical components further comprises at least one surface contaminant component, the method further comprising: calculating the pollution degree value of each pixel point of the material component abundance diagram based on the abundance value of the channel corresponding to the surface pollutant component in the component abundance vector of each pixel point; Calculating pollution regulation factors of each aging area cluster based on pollution degree values of all pixel points in the aging area cluster and a preset