A curtain wall safety detection method and system

Abstract

The embodiment of the application discloses a curtain wall safety detection method and system; after obtaining the original image set of the curtain wall, the original image in the original image set can be matched to obtain the matched image pair, and the matched image pair is weighted and fused to obtain the fused image; the semantic features of the fused image are extracted through a preset semantic extraction model, and the evaluation weight corresponding to the safety evaluation index of the curtain wall is determined, the safety evaluation index includes the structure information of the curtain wall, the material state of the curtain wall and the environment information of the environment where the curtain wall is located; the semantic features of the fused image are evaluated based on the fuzzy relation matrix to obtain the evaluation score of the curtain wall, and the safety level of the curtain wall is determined according to the evaluation score and the evaluation threshold; therefore, the comprehensiveness, objectivity and accuracy of the curtain wall safety detection can be improved, and reliable quantitative basis is provided for curtain wall safety maintenance and risk control.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and more specifically to a method and system for detecting the safety of curtain walls. Background Technology

[0002] In recent years, with the acceleration of urbanization, building curtain walls have been widely used as the external envelope of high-rise buildings. However, their safety hazards have become increasingly prominent. Problems such as curtain wall aging, material failure, and structural deformation can easily lead to safety accidents such as falls and collapses, posing a serious threat to public safety.

[0003] In the process of researching and practicing current technologies, the inventors of this application have found that in related technologies, the safety inspection of curtain walls mainly relies on manual inspection. Manual inspection is limited by the height of high-rise buildings, posing a risk of not being able to reach the site, and the inspection efficiency is low and the coverage is not comprehensive. Moreover, safety inspection relies heavily on human experience to make judgments, and the evaluation results are highly subjective and inconsistent. Therefore, the evaluation results deviate significantly from the actual safety status, making it difficult to meet the actual needs of accurate safety inspection of curtain walls. Summary of the Invention

[0004] This application provides a method and system for curtain wall safety testing, which can improve the accuracy of curtain wall safety testing and provide a basis for curtain wall safety maintenance.

[0005] A method for testing the safety of a curtain wall includes: Obtain a set of original images of the curtain wall, the set of original images including at least one original image; The original images in the original image set are matched to obtain matched image pairs, and the matched image pairs are weighted and fused to obtain a fused image. The semantic features of the fused image are extracted by a preset semantic extraction model, and the evaluation weights corresponding to the safety evaluation indicators of the curtain wall are determined. The safety evaluation indicators include the structural information of the curtain wall, the material state of the curtain wall, and the environmental information of the environment in which the curtain wall is located. The semantic features of the fused image are evaluated based on the fuzzy relation matrix to obtain the evaluation score of the curtain wall. The fuzzy relation matrix is ​​constructed from the evaluation weights corresponding to the security evaluation indicators, the membership functions corresponding to the security evaluation indicators, and the adjustment functions. The evaluation score indicates the security score of the curtain wall. The safety level of the curtain wall is determined based on the assessment score and assessment threshold.

[0006] Accordingly, this application provides a curtain wall safety detection system, including: An image processing unit is configured to acquire an original image set, the original image set including at least one original image; match the original images in the original image set to obtain matched image pairs; and perform weighted fusion on the matched image pairs to obtain a fused image. The weight determination unit is used to extract the semantic features of the fused image through a preset semantic extraction model, and determine the evaluation weights corresponding to the safety evaluation indicators of the curtain wall. The safety evaluation indicators include the structural information of the curtain wall, the material state of the curtain wall, and the environmental information of the environment in which the curtain wall is located. The evaluation score determination unit is used to evaluate the semantic features of the fused image based on the fuzzy relation matrix to obtain the evaluation score of the curtain wall. The fuzzy relation matrix is ​​constructed from the evaluation weights corresponding to the security evaluation indicators, the membership functions corresponding to the security evaluation indicators, and the adjustment functions. The evaluation score indicates the security score of the curtain wall. The safety level determination unit is used to determine the safety level of the curtain wall based on the evaluation score and the evaluation threshold.

[0007] In this embodiment, after obtaining a set of original images of the curtain wall, including at least one original image, the original images in the set can be matched to obtain matched image pairs. These matched image pairs are then weighted and fused to obtain a fused image. Semantic features of the fused image are extracted using a preset semantic extraction model, and evaluation weights corresponding to the curtain wall's safety assessment indicators are determined. These safety assessment indicators include the curtain wall's structural information, material condition, and environmental information of the environment in which the curtain wall is located. The semantic features of the fused image are evaluated based on a fuzzy relation matrix to obtain an evaluation score for the curtain wall. Based on the evaluation score and an evaluation threshold, the safety level of the curtain wall is determined. This solution integrates complementary information from multiple image sources, enhancing the completeness and accuracy of curtain wall condition representation. This overcomes the limitations of traditional single-image detection, which suffers from incomplete information and blurred details. It enables comprehensive capture of surface defects and internal hidden dangers within the curtain wall. Furthermore, since the fuzzy relation matrix is ​​constructed from the evaluation weights, membership functions, and adjustment functions corresponding to the safety assessment indicators, it avoids the subjectivity and bias of manual judgment, achieving a quantitative assessment of the curtain wall's safety status. Therefore, it improves the comprehensiveness, objectivity, and accuracy of curtain wall safety inspection, providing a reliable quantitative basis for curtain wall safety maintenance and risk management. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 This is a schematic diagram of a scenario for a curtain wall safety detection method provided in an embodiment of this application; Figure 2 This is a flowchart illustrating a curtain wall safety testing method provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of a curtain wall safety detection system provided in an embodiment of this application; Figure 4 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0010] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0011] This application provides a method and system for detecting the safety of a curtain wall. The curtain wall safety detection system can be integrated into an electronic device, which can be a server or a terminal, etc.

[0012] The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery network (CDN), and big data and artificial intelligence platforms. The terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, etc., but is not limited to these. The terminal and server can be directly or indirectly connected via wired or wireless communication, which is not limited herein.

[0013] For example, see Figure 1Taking the integration of a curtain wall safety inspection system into an electronic device as an example, the electronic device can acquire the original image set of the curtain wall, match the original images in the original image set to obtain matched image pairs, and then perform weighted fusion on the matched image pairs to obtain a fused image. The semantic features of the fused image are extracted through a preset semantic extraction model, and the evaluation weights corresponding to the curtain wall's safety assessment indicators are determined. These safety assessment indicators include the curtain wall's structural information, material condition, and environmental information of the environment in which the curtain wall is located. The semantic features of the fused image are evaluated based on a fuzzy relation matrix to obtain the curtain wall's evaluation score. Based on the evaluation score and the evaluation threshold, the safety level of the curtain wall is determined. Therefore, the comprehensiveness, objectivity, and accuracy of curtain wall safety inspection can be improved, providing a reliable quantitative basis for curtain wall safety maintenance and risk management.

[0014] It is understood that, in the specific embodiments of this application, data such as the original image set of the curtain wall, curtain wall type information, and environmental impact parameters are involved. When the following embodiments of this application are applied to specific products or technologies, permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0015] The following sections provide detailed descriptions of each example. It should be noted that the order in which the embodiments are described is not intended to limit the preferred order of the embodiments.

[0016] This embodiment will be described from the perspective of a curtain wall safety detection system, which can be integrated into an electronic device, such as a server or a terminal. The terminal can include tablet computers, laptops, personal computers (PCs), wearable devices, virtual reality devices, or other smart devices capable of performing safety detection.

[0017] like Figure 2 As shown, the specific process of this curtain wall safety testing method is as follows: 101. Obtain the original image set of the curtain wall, wherein the original image set includes at least one original image.

[0018] In this context, a curtain wall can be understood as a continuous wall structure installed on the main structure of a high-rise building, serving as an external enclosure and decoration, typically composed of panels (such as glass or stone) and a supporting structural system. The original image set can be understood as the collection of images of the curtain wall acquired for assessing its safety.

[0019] Here, the original image can be understood as an image acquired by different acquisition devices in different imaging modalities. For example, the original image can be a first visible light image of the curtain wall acquired by the first acquisition device, or a second visible light image of the curtain wall acquired by the second acquisition device, or an infrared thermal image of the curtain wall, and so on.

[0020] The first acquisition device is a fixed acquisition device, such as a fixed camera installed at the base of the building or on the surrounding ground. The first visible light image acquired by the first acquisition device can be understood as a visible light image collected daily at regular intervals by the fixed camera for routine monitoring. Its resolution is usually no less than 5 million pixels, and it can be used to record macroscopic defects on the curtain wall surface, such as cracks, deformation, and corrosion.

[0021] The second acquisition device is a mobile acquisition device, such as a drone equipped with various sensors (e.g., visible light cameras, infrared thermal imagers). The second visible light image of the curtain wall acquired by the second acquisition device can be understood as an image captured by a high-resolution visible light camera mounted on the drone, with a resolution of over 40 million pixels. Because the high-resolution visible light camera on the drone has flexible shooting angles and close-range observation capabilities, it can capture richer details on the curtain wall surface. Therefore, the second visible light image can be used to capture the fine texture and minute defects of the curtain wall. The infrared thermal image of the curtain wall acquired by the second acquisition device can be used to reflect the temperature distribution on the curtain wall surface, thereby inferring hidden defects such as bonding failures, voids, and thermal stress concentrations. Its resolution is typically no less than 640×512, and the temperature resolution is no higher than 0.05℃.

[0022] There are several ways to obtain the original image set of the curtain wall. For example, a timed task can be set to control a fixed camera to automatically collect the first visible light image at a fixed time every day. Alternatively, a drone can be controlled to automatically scan the curtain wall according to a set forward overlap rate (e.g., ≥80%) and lateral overlap rate (e.g., ≥70%) through a preset drone flight path, simultaneously collecting the second visible light image and infrared thermal image. Alternatively, drones can be temporarily dispatched to supplement the collection of images in key areas when encountering severe weather or special inspection needs. Alternatively, real-time data streams from multiple fixed cameras and multiple drones can be received through edge computing nodes and aggregated to form the original image set, and so on.

[0023] It should be noted that the original image set may contain at least one first visible light image, at least one second visible light image, and at least one infrared thermal image. Furthermore, in terms of acquisition time, fixed cameras typically acquire data at a higher frequency (e.g., once daily) to form time-series data; while drones can perform full-cycle scans at a lower frequency (e.g., once weekly or monthly) as needed to obtain more comprehensive data. Thus, through this acquisition method, the original image set can achieve comprehensive, multi-dimensional coverage of the curtain wall surface and its internal condition.

[0024] In this application, a multi-source, multi-modal acquisition strategy of using a fixed camera and a drone to collaboratively acquire visible light and infrared images is adopted. This strategy ensures the timeliness of curtain wall safety inspection (fixed camera) and solves the problem of inspection of high-rise buildings that cannot be accessed (drone). It provides a rich and reliable data foundation for subsequent high-precision image fusion and curtain wall safety inspection and evaluation.

[0025] 102. Match the original images in the original image set to obtain matched image pairs, and perform weighted fusion on the matched image pairs to obtain fused images.

[0026] The matched image pair can be understood as a combination of images after spatial calibration. For example, registering and aligning a visible light image captured by a fixed camera with a visible light image captured by a drone, or registering and aligning a visible light image with an infrared thermal image, ensures that images from different sources have consistent geometric structure at the pixel level. The fused image can be understood as a composite image generated by effectively integrating complementary information (such as visible light texture details and infrared thermal radiation information) from the matched image pair through multi-scale transformation and weighting.

[0027] Before matching the original images in the original image set, the acquired original images can be preprocessed. There are several ways to preprocess the original images, including the following: (1) Grayscale processing: The acquired RGB color images (first visible light image I1, second visible light image I2, infrared thermal image T) can be converted into single-channel grayscale values ​​using the standard weighted average method, so as to convert them into grayscale images (first visible light image I1_g, second visible light image I2_g, infrared thermal image T_g) to reduce the amount of data and computational complexity.

[0028] (2) Image enhancement processing: For the grayscale image, the Retinex algorithm is used for enhancement processing to improve uneven lighting, enhance image contrast and detail visibility. Image enhancement processing can include multi-scale Retinex enhancement, color restoration processing and simple white balance processing.

[0029] Multi-scale Retinex enhancement involves applying Gaussian filtering at multiple scales to grayscale images (first visible light image I1_g, second visible light image I2_g, and infrared thermal image T_g) to estimate the illumination components at different scales. This allows for the separation and enhancement of the reflection components, which contain detailed information, from the original image, resulting in a multi-scale enhanced intermediate image. By employing multi-scale filtering, both global brightness information and local detail features can be preserved simultaneously, avoiding detail loss or over-enhancement issues caused by single-scale enhancement. After obtaining the multi-scale enhanced intermediate image, color restoration processing can be performed. For example, a color restoration factor can be introduced to correct the color of the multi-scale enhanced image, restoring its natural colors and avoiding color distortion that may result from multi-scale processing. Subsequently, a simple white balance adjustment can be applied to the color-restored image. This can be done by mapping the brightest parts of the image to white and the darkest parts to black, and linearly stretching the intermediate pixel values ​​to expand the dynamic range of the image, enhance the overall contrast, and make the brightness distribution more uniform and the visual effect clearer. Thus, by sequentially performing multi-scale Retinex enhancement, color restoration processing, and simple white balance processing, the quality of the image under complex lighting conditions can be significantly improved, resulting in enhanced images (enhanced first visible light image I1_e, enhanced second visible light image I2_e, and enhanced infrared thermal image T_e).

[0030] (3) Image denoising: Gaussian filtering can be used to smooth and denoise the enhanced first visible light image I1_e, the enhanced second visible light image I2_e, and the enhanced infrared thermal image T_e. By setting an appropriate Gaussian kernel standard deviation, a balance can be achieved between denoising intensity and detail preservation, thereby obtaining the denoised images (denoised first visible light image I1_d, denoised second visible light image I2_d, and denoised infrared thermal image T_d) to eliminate noise that may be introduced or amplified during image acquisition and enhancement.

[0031] In this application, through the sequential execution of the above-mentioned grayscale processing, image enhancement processing, and image denoising processing, the image quality of the original image is significantly improved, the image details are more prominent, the problem of uneven illumination is effectively improved, and the noise is effectively suppressed, thereby providing a foundation for subsequent image registration of the curtain wall.

[0032] Specifically, the method of matching the original images in the original image set to obtain matched image pairs may include the following steps S201-S204: S201. Use a feature extraction algorithm to extract features from the first visible light image to obtain the first multi-directional features corresponding to the first visible light image.

[0033] S202. Use a feature extraction algorithm to extract features from the second visible light image to obtain the second multi-directional features corresponding to the second visible light image.

[0034] The feature extraction algorithm can be understood as an algorithm that identifies and extracts representative, stable, and distinguishable local structures (such as edges, corners, and textures) from an image. In this application, the feature extraction algorithm can specifically employ an improved Shearlet-SURF algorithm. This algorithm first uses Shearlet transform to perform multi-scale and multi-directional decomposition of the image to capture structural information of the image in different directions. The first multi-directional feature can be understood as the set of feature points extracted at different scales and in different directions after performing Shearlet transform on the first visible light image. The second multi-directional feature can be understood as the set of feature points extracted at different scales and in different directions after performing Shearlet transform on the second visible light image.

[0035] Specifically, the method of extracting features from the first visible light image using a feature extraction algorithm to obtain the first multi-directional features corresponding to the first visible light image may include: performing multi-scale and multi-directional decomposition on the preprocessed first visible light image using Shearlet transform, wherein each scale corresponds to a Shearlet basis function in a different direction to decompose the first visible light image into multiple scales and directions; extracting direction and intensity information of edge features at each scale; analyzing the Shearlet transform coefficients at each scale to extract direction and intensity information of edge features to form feature descriptors; and selecting coefficient points with significant response values ​​from the transform coefficients as Shearlet feature points to form the first multi-directional feature set K1.

[0036] The method of extracting features from the second visible light image using a feature extraction algorithm to obtain the second multi-directional features corresponding to the second visible light image can specifically include: performing multi-scale and multi-directional decomposition on the preprocessed second visible light image using Shearlet transform, wherein each scale corresponds to a Shearlet basis function in a different direction to decompose the second visible light image into multiple scales and directions; extracting direction and intensity information of edge features at each scale; analyzing the Shearlet transform coefficients at each scale to extract direction and intensity information of edge features to form feature descriptors; and selecting coefficient points with significant response values ​​from the transform coefficients as Shearlet feature points to form the second multi-directional feature set K2.

[0037] As an example, the Shearlet transform can be used to decompose an image into multiple scales and directions, as shown in the following formula (A1): in, This can be understood as a scale of 2. j Shearlet basis functions with direction k and position l. It can be understood as either the first visible light image or the second visible light image. This can be understood as the coefficients obtained after transformation. It's also understandable that multiple decomposition scales can be set; for example, the scale parameter j can take values ​​of 1, 2, 4, etc., and each scale corresponds to a different number of directions; for example, the direction k can take values ​​of 4, 8, 16, etc.

[0038] It should be noted that Shearlet transform, as a multi-scale geometric analysis method, can capture anisotropic features in images, such as edges and contours, more effectively than traditional wavelet transform. Therefore, it is particularly suitable for feature extraction of curtain wall images with rich texture and structural information. By setting different scale and orientation parameters, the precision of feature extraction can be flexibly adjusted to adapt to curtain wall images of different resolutions and complexities. Furthermore, although the first and second visible light images originate from different acquisition devices and have different resolutions, viewing angles, and lighting conditions, by pre-setting the feature extraction algorithm and parameter settings, the first and second visible light images can be mapped to the same feature space for comparison, thereby achieving cross-viewpoint and cross-device image matching.

[0039] S203. Based on the first multi-directional features and the second multi-directional features, the first visible light image and the second visible light image are matched to obtain a reference matching pair.

[0040] The reference matching pair can be understood as an image pair formed by spatial matching of the first visible light image and the second visible light image. The spatial positions and viewing angles of the two visible light images in the reference matching pair are matched, corresponding to the same area of ​​the curtain wall.

[0041] The method of matching a first visible light image and a second visible light image based on the first multi-directional features and the second multi-directional features to obtain a reference matching pair can specifically include: matching feature points using the SURF algorithm based on the first multi-directional features (first multi-directional feature set K1) and the second multi-directional features (second multi-directional feature set K2) extracted by Shearlet transform. For example, for feature points detected by Shearlet transform, the SURF algorithm is used to calculate the approximate value of its Hessian matrix to accurately locate the feature points, and the principal direction is calculated for each accurately located feature point to improve the rotation invariance of the feature. Then, SURF descriptors of the surrounding area are extracted with the feature point as the center, thereby obtaining a SURF feature point set and its corresponding descriptors. After that, the fast nearest neighbor search algorithm is used for feature point matching, and the search efficiency is improved by constructing a kd-tree index to initially obtain matching point pairs. Finally, the initially obtained matching point pairs are filtered, for example, by using the distance ratio method to retain valid matching pairs with a distance ratio less than a preset threshold, thereby obtaining a preliminary feature point matching list, and then obtaining a reference matching pair.

[0042] The expression for the Hessian matrix and the SURF descriptor can be represented by formula (A2): ; (A2) in, This can be understood as the second derivative of the image in the x-direction; This can be understood as the second derivative of the image in the xy direction; It can be understood as the second derivative of the image in the y-direction; D can be understood as the SURF descriptor, which contains the orientation and intensity information of the region around the feature point, so as to obtain the SURF feature point set F1 and F2, and the corresponding descriptors D1 and D2.

[0043] It should be noted that the Shearlet transform acts as an initial screening tool for feature points in this process, quickly locating regions with significant structural information in the image through multi-scale and multi-directional decomposition; while the SURF algorithm performs precise localization and feature description on these candidate regions. The combination of the two ensures both feature richness and improves computational efficiency.

[0044] S204. Match and align the reference matching pair with the infrared thermal image to obtain a matched image pair.

[0045] The process of matching and aligning reference matching pairs with infrared thermal images to obtain matched image pairs can specifically include: employing a random sampling consensus algorithm to calculate the affine transformation matrix by randomly sampling multiple sets of feature point pairs, evaluating the proportion of interior points in all matching point pairs under this affine transformation matrix, and iteratively optimizing and retaining the transformation matrix with the highest proportion of interior points, thereby eliminating mismatched points and obtaining an optimized feature point matching list and a preliminary transformation matrix H0. Then, based on the optimized matching list, a coarse registration transformation matrix is ​​calculated. For example, the coordinates of the matching feature points can be extracted to construct coordinate matrices P and Q, and singular value decomposition can be used to solve the affine transformation matrix H0, thus obtaining the coarse registration result. Next, a fine registration transformation matrix is ​​calculated. The coarse registration matrix H0 is applied to the second visible light image for preliminary transformation, and then an improved iterative nearest-point algorithm is used for fine registration. In the iterative nearest-point algorithm, the transformation matrix is ​​successively optimized to find the best matching relationship between the two sets of point clouds until the registration error converges to within a preset threshold (e.g., 0.5 pixels), thus obtaining the final fine registration transformation matrix H. Then, the final transformation matrix H can be applied to the infrared thermal image to precisely align it with the first visible light image, forming a matched image pair that includes the first visible light image, the registered second visible light image, and the registered infrared thermal image.

[0046] After obtaining the matched image pairs, a weighted fusion process can be performed on them to obtain the fused image. Specifically, the weighted fusion process can include: performing wavelet decomposition on the matched image pairs to obtain the signal quality and gradient information of the matching images; then, determining the fusion weights based on the signal quality and gradient information of the matching images, and performing weighted fusion based on these weights to obtain the fused image.

[0047] The signal quality of the matched image can be understood as a quantitative indicator measuring image sharpness and noise level, typically represented by the signal-to-noise ratio (SNR). A higher SNR indicates richer effective information and less noise interference in the image. The gradient information of the matched image can be understood as a quantitative indicator reflecting the edge strength and texture richness of the image, typically represented by the gradient magnitude. A larger gradient magnitude indicates sharper edges and richer texture, and also means that the image contains more high-frequency detail information in that region. The image fusion weight can be understood as a quantitative indicator reflecting the degree of drastic change in pixel values ​​in the image, typically represented by the gradient magnitude. A larger gradient magnitude indicates sharper edges and richer texture.

[0048] The method of performing wavelet decomposition on the matched image pair to obtain the signal quality and gradient information of the matched image in the matched image pair may specifically include the following steps S211-S213: S211. Wavelet basis functions are used to decompose the matching image into multiple levels to extract low-frequency and high-frequency information of the matching image. The low-frequency information indicates the overall contour information of the matching image, and the high-frequency information indicates the edge information and detail information of the matching image.

[0049] In this context, wavelet basis functions can be understood as basis functions used for wavelet decomposition. They possess properties such as finite support and orthogonality, enabling the decomposition of image signals into different frequency bands and scales. In this application, Daubechies wavelet basis functions, such as the db4 wavelet, can be used. Low-frequency information can be understood as the low-frequency coefficients obtained after wavelet decomposition, corresponding to regions in the image with gentle gray-level changes, reflecting the overall structure and brightness distribution of the image. Overall contour information can be understood as the main geometric structures, shapes, and brightness distributions of the image, representing macroscopic features. High-frequency information can be understood as the high-frequency coefficients obtained after wavelet decomposition, corresponding to regions in the image with dramatic gray-level changes, reflecting the edges and texture details of the image. Edge information can be understood as the boundaries between different regions in the image, such as the seams of curtain wall panels or the outlines of cracks, typically manifested as step changes in pixel values. Detail information can be understood as the fine texture structure in the image, such as the fine lines and micro-cracks on the material surface, typically manifested as high-frequency fluctuations in pixel values.

[0050] The method of using wavelet basis functions to decompose the matching image at multiple levels and extract low-frequency and high-frequency information can be specifically as follows: Select the db4 orthogonal wavelet basis function as the core decomposition basis function; set the wavelet decomposition level to 3 or more according to the resolution and detail feature complexity of the matching image; input the matching image into the wavelet decomposition model; and decompose the matching image level by level through the convolution operation between the wavelet basis function and the image pixels. Each level of decomposition decomposes the current image component into 1 low-frequency component and 3 high-frequency components (including horizontal, vertical, and diagonal directions); after completing the decomposition of the preset number of levels, extract the low-frequency component obtained from the last level of decomposition as the low-frequency information of the matching image, and integrate all the high-frequency components obtained from each level of decomposition as the high-frequency information of the matching image.

[0051] It should be noted that multi-level wavelet decomposition can separate image information according to frequency, which facilitates the use of different fusion strategies for different frequency components, thereby better preserving the macroscopic structure and microscopic details of the image.

[0052] S212. Calculate the pixel mean and noise standard deviation of the effective region in the matched image, and calculate the signal quality of the matched image based on the pixel mean and noise standard deviation.

[0053] In this context, the effective region can be understood as the area in the image representing the main structure of the curtain wall. It typically needs to avoid irrelevant areas such as the background and sky to ensure that the calculated signal quality accurately reflects the condition of the curtain wall itself. The pixel mean of the effective region can be understood as the arithmetic mean of the grayscale values ​​of all pixels within the selected effective region, reflecting the average brightness level of that area. The noise standard deviation can be understood as the degree of fluctuation in pixel grayscale values ​​in flat or background areas of the image, used to estimate the intensity of noise in the image. The signal quality of the matched image can be understood as using the signal-to-noise ratio (SNR) as a metric, that is, the ratio of signal energy to noise energy in the effective region, usually expressed in decibels. A higher SNR indicates better image quality.

[0054] Specifically, the method for calculating the pixel mean and noise standard deviation of the effective region in the matching image can include: segmenting the matching image into regions using an image segmentation algorithm, identifying and extracting the main body region of the curtain wall in the image based on the contour features of the curtain wall, and using this as the effective region of the matching image; traversing all pixels within the effective region, counting the total number of pixels and calculating the sum of the gray values ​​of all pixels, and using the ratio of the sum to the total number of pixels as the pixel mean of the effective region; based on the pixel mean, calculating the sum of squares of the differences between the gray value of each pixel within the effective region and the pixel mean, and taking the square root of the ratio of the sum of squares of the differences to the total number of pixels to obtain the noise standard deviation of the effective region of the matching image.

[0055] One method for calculating the signal quality of a matched image based on pixel mean and noise standard deviation can be as follows: The root mean square signal-to-noise ratio (SNR) formula is used as the basis for signal quality calculation. The square of the effective pixel mean is used as the numerator, and the square of the noise standard deviation is used as the denominator. The ratio of these two values ​​is calculated, and the logarithm is taken to obtain the SNR in decibels (dB). This SNR value represents the signal quality of the matched image. As an example, the signal quality of the matched image can be calculated using the following formula (A3): (A3) in, This can be understood as signal quality. This can be understood as the average pixel value of the effective area. This can be understood as the noise standard deviation.

[0056] It should be noted that the extraction of the effective area must ensure completeness and accuracy to avoid deviations in the calculation of pixel mean and noise standard deviation due to omission of the main curtain wall area or selection of multiple background interference areas; the extraction rules for the effective area of ​​infrared thermal images and visible light images should be consistent to ensure the fairness and comparability of signal quality calculation.

[0057] S213. Calculate the gradient magnitude of the matching image to obtain the gradient information of the matching image. The gradient information indicates the edge sharpness of the matching image.

[0058] In this context, the gradient magnitude of the matched image can be understood as the composite value of the gray-level gradient values ​​of each pixel in the matched image in the horizontal and vertical directions. It quantitatively represents the magnitude of gray-level change at a single pixel; a larger gradient magnitude indicates a more drastic gray-level change at that pixel. The gradient information of the matched image can be understood as a quantized feature obtained by integrating the gradient magnitudes of all pixels within the effective region of the matched image, represented by the statistical values ​​(mean and maximum) of the pixel gradient magnitudes. The edge sharpness of the matched image can be understood as the degree to which features such as curtain wall edges, defect contours, and detailed textures are discernible in the image.

[0059] The method of calculating the gradient magnitude of the matching image to obtain its gradient information can specifically include: convolving the image with both horizontal and vertical Sobel operators to obtain the horizontal and vertical gradient components. Then, for each pixel, the square root of the sum of the squares of its horizontal and vertical gradient components is calculated to obtain the gradient magnitude at that point. After calculating for all pixels, the gradient magnitude matrix of the entire image is obtained, which represents the gradient information of the matching image. Regions with larger gradient magnitudes correspond to edges and textured areas of the image, while regions with smaller gradient magnitudes correspond to flat areas. As an example, the gradient information of the matching image can be calculated using the following formula (A4): (A4) in, This can be understood as matching the gradient information of the image. This can be understood as the gradient value in the horizontal direction. This can be understood as the gradient value in the vertical direction.

[0060] It should be noted that gradient calculation is performed only on the valid regions of the matched images to avoid invalid gradient values ​​in the background regions interfering with the calculation results. For all matched images in the matched image pair, the same gradient calculation operator and calculation rules are used to ensure the comparability of gradient information.

[0061] The method of performing wavelet decomposition on the matched image pair to obtain the signal quality and gradient information of the matched image in the matched image pair may specifically include the following steps S214-S215: S214. Based on the signal quality, calculate the low-frequency fusion weights corresponding to the matching images for the low-frequency information of the matching images.

[0062] The low-frequency fusion weights corresponding to the matched images can be understood as the weight coefficients assigned to each image when fusing the low-frequency coefficients of each matched image. These weights reflect the degree to which the low-frequency information of each image contributes to the final fused image.

[0063] Specifically, for the low-frequency information of the matched images, the method of calculating the low-frequency fusion weights corresponding to the matched images based on signal quality can include: obtaining the signal-to-noise ratio (SNR) of each matched image as a quantitative indicator of signal quality, and determining the weight ratio of each image based on the SNR magnitude. Images with higher SNRs have more reliable low-frequency information and should be assigned higher fusion weights. For example, a normalized weighted average method can be used for calculation. That is, for the registered first visible light image, second visible light image, and infrared thermal image, their SNR values ​​are divided by the sum of the SNRs of the three images. The result is the low-frequency fusion weight corresponding to each image, ensuring that images with high SNRs dominate in low-frequency fusion, thereby making the overall contour information of the fused image clearer and more reliable.

[0064] S215. Based on the high-frequency information of the matching image, calculate the high-frequency fusion weight corresponding to the matching image by combining signal quality and gradient information.

[0065] The high-frequency fusion weight can be understood as the weight coefficient assigned to each image when fusing the high-frequency coefficients of the matched images. This weight reflects the contribution of edge and detail information of each image to the final fused image.

[0066] The method of calculating the high-frequency fusion weights for matching images by combining signal quality and gradient information can specifically include: obtaining the signal-to-noise ratio (SNR) of each matching image as a signal quality indicator and the gradient magnitude of each matching image as a gradient information indicator, and then calculating the high-frequency fusion weights for each matching image by combining the signal quality and gradient information. For example, the SNR of each image can be multiplied by its gradient magnitude to obtain a comprehensive evaluation value, which reflects both the SNR level and edge sharpness of the image. Then, the comprehensive evaluation value of each image is divided by the sum of the comprehensive evaluation values ​​of the three images, and the result is the high-frequency fusion weight for each image. In this way, both the overall image quality (SNR) and the richness of image detail (gradient information) are considered, enabling the priority preservation of image regions that are both clear and contain rich details during the fusion process.

[0067] S216. Based on the low-frequency fusion weights corresponding to the low-frequency information of the matching images and the high-frequency fusion weights corresponding to the high-frequency information of the matching images, the inverse wavelet transform is used to fuse the matched image pairs to obtain the fused image.

[0068] The method of fusing matched image pairs using inverse wavelet transform to obtain the fused image can specifically include: multiplying the low-frequency coefficients of the registered first visible light image, second visible light image, and infrared thermal image by their respective low-frequency fusion weights, and then summing the multiplications to obtain the fused low-frequency coefficients. Next, based on the calculated high-frequency fusion weights, the high-frequency coefficients in each direction and at each scale are weighted and summed. For example, for the high-frequency coefficients in the horizontal, vertical, and diagonal directions, the high-frequency coefficients corresponding to the three images are multiplied by their respective high-frequency fusion weights and then summed to obtain the fused high-frequency coefficients in each direction. If multi-level decomposition is required, the above weighted fusion operation is performed on the high-frequency coefficients of each layer. Finally, the fused low-frequency coefficients and the high-frequency coefficients of each layer and direction are combined, and image reconstruction is performed using inverse wavelet transform. Through layer-by-layer reconstruction, the final fused visible-infrared composite image is obtained.

[0069] It should be noted that the inverse wavelet transform reconstruction process recombines the weighted multi-resolution components into a complete image, ensuring that the fused image retains both clear outlines and rich details, thereby achieving high-quality image fusion.

[0070] 103. Extract the semantic features of the fused image through a preset semantic extraction model, and determine the evaluation weights corresponding to the safety evaluation indicators of the curtain wall. The safety evaluation indicators include the structural information of the curtain wall, the material state of the curtain wall, and the environmental information of the environment in which the curtain wall is located.

[0071] The pre-set semantic extraction model can be understood as a pre-trained deep learning neural network model used to automatically extract high-level semantic features from images. In this application, a lightweight convolutional neural network, such as MobileNetV4, can be used. The semantic features of the fused image can be understood as high-dimensional feature vectors extracted from the fused image by the pre-set semantic extraction model, which can characterize the safety status of the curtain wall. Safety assessment indicators can include the structural information of the curtain wall, the material status of the curtain wall, and the environmental information of the environment in which the curtain wall is located.

[0072] The structural information of the curtain wall can be understood as indicators reflecting the safety status of the supporting structure and connection parts. For example, this may include specific indicators such as structural deformation, loose connection nodes, corrosion of the supporting structure, and inter-story displacement. The material condition of the curtain wall can be understood as indicators reflecting the aging and damage of the curtain wall panel materials and adhesive materials. For example, this may include specific indicators such as glass cracks, structural adhesive failure, sealant aging, and hollowing of insulation materials. The environmental information of the environment in which the curtain wall is located can be understood as external environmental factors affecting the safety of the curtain wall. For example, this may include specific indicators such as wind load level, temperature changes, humidity changes, and ultraviolet radiation intensity. The evaluation weights corresponding to the safety assessment indicators can be understood as quantitative coefficients measuring the importance of each safety indicator to the overall safety status of the curtain wall. The larger the weight, the greater the contribution of the indicator to the final safety assessment result. In this application, the evaluation weights are not fixed but dynamically generated based on the semantic features of the fused image, thereby achieving adaptive evaluation for different curtain walls and different working conditions.

[0073] There are several ways to extract semantic features from the fused image using a preset semantic extraction model and determine the evaluation weights corresponding to the safety assessment indicators of the curtain wall. Specifically, one approach is to construct a hierarchical model comprising a target layer, a criterion layer, and an indicator layer. The criterion layer includes three dimensions: structural information, material state, and environmental information of the curtain wall. Each criterion layer has multiple specific low-level indicators. Next, the fused image is input into the preset semantic extraction model, and the model's forward propagation extracts high-level semantic feature vectors that characterize the safety status of the curtain wall. Then, the extracted semantic feature vectors are input into a weight generation network composed of fully connected layers. After linear transformation and Softmax normalization, dynamic weights corresponding to each safety assessment indicator are obtained. Finally, the stability of the generated dynamic weights is verified. For example, the degree of weight fluctuation is judged by calculating the cosine similarity with the initial weights. If necessary, an exponentially weighted moving average method is used for smoothing to ensure the reliability of the evaluation results.

[0074] 104. The semantic features of the fused image are evaluated based on the fuzzy relation matrix to obtain the evaluation score of the curtain wall.

[0075] The fuzzy relation matrix can be understood as a two-dimensional matrix used to quantify the degree of membership of each safety assessment indicator with respect to different safety levels. The fuzzy relation matrix is ​​constructed from the assessment weights corresponding to the safety assessment indicators, the membership functions corresponding to the safety assessment indicators, and the adjustment function. The assessment score indicates the safety score of the curtain wall. In the fuzzy relation matrix, the rows correspond to different safety assessment indicators, and the columns correspond to preset safety levels (e.g., first safety level, second safety level, danger level). Each element in the matrix represents the membership degree of a certain safety assessment indicator to a certain safety level. The assessment score of the curtain wall can be understood as a numerical value calculated through fuzzy comprehensive evaluation to quantify the overall safety level of the curtain wall. The higher the score, the better the safety status of the curtain wall; the lower the score, the greater the safety hazard.

[0076] The membership functions corresponding to the security assessment indicators can include trapezoidal membership functions, Gaussian membership functions, and triangular membership functions. Different types of membership functions are adapted to assessment indicators with different distribution characteristics and are used to calculate the degree of membership of an indicator to a certain security level. The adjustment function can be understood as a correction function set based on the actual scenario of curtain wall inspection. It is used to compensate for errors in membership results caused by factors such as image feature extraction deviations and environmental interference, thereby improving the accuracy of membership calculation.

[0077] Optionally, in some embodiments, before evaluating the semantic features of the fused image based on the fuzzy relation matrix to obtain the evaluation score of the curtain wall, the following steps (S401-S404) are further included: S401. Based on the structural information, material condition, and environmental information of the curtain wall, and corresponding safety assessment indicators, set reference membership functions, including trapezoidal membership functions, Gaussian membership functions, and triangular membership functions.

[0078] The reference membership function can be understood as a pre-defined basic function form for each security assessment indicator, used to convert the indicator measurement value into membership degree. It can be understood that different types of indicators are suitable for different forms of membership function due to their different physical characteristics and changing patterns.

[0079] The trapezoidal membership function can be understood as a piecewise linear function with a trapezoidal shape. It is suitable for indicators with a defined threshold range, such as crack length and corrosion area. This function is typically controlled by four parameters, corresponding to the starting points of the lower and upper bases of the trapezoid. The Gaussian membership function can be understood as a function based on a normal distribution, with a bell-shaped shape. It is suitable for indicators influenced by random factors and exhibiting statistical distribution characteristics, such as bond strength and temperature changes. This function is controlled by two parameters: mean and standard deviation. The triangular membership function can be understood as a piecewise linear function with a triangular shape. It is suitable for indicators with optimal values ​​or ideal ranges, such as structural deformation and interlayer displacement. This function is typically controlled by three parameters, corresponding to the left, right, and left vertices of the triangle.

[0080] Specifically, setting reference membership functions based on the structural information, material condition, and environmental information of the curtain wall for each safety assessment indicator can include: sorting out the numerical range, safety threshold, and distribution characteristics of each safety assessment indicator, clarifying the correlation logic between the indicator and the safety level, and matching corresponding types of reference membership functions for indicators with different characteristics. For example, indicators with clear threshold ranges, such as the proportion of cracked areas and ultraviolet intensity, can be matched with trapezoidal membership functions; indicators with normal distribution, such as bonding strength, can be matched with Gaussian membership functions; and indicators with linear changes, such as structural deformation degree and inter-story displacement, can be matched with triangular membership functions.

[0081] It should be noted that different forms of membership functions have different mathematical properties and applicable ranges. Choosing the appropriate function type is the key to accurately quantifying the state of the indicator.

[0082] S402. Based on the characteristics of the security assessment index contained in the fused image, adjust the parameters of the reference membership function to obtain the target membership function corresponding to the security assessment index.

[0083] The parameters of the reference membership function can be understood as specific values ​​controlling the shape of the function, such as the four inflection points of a trapezoidal function, the mean and standard deviation of a Gaussian function, and the three vertices of a triangular function. The target membership function can be understood as a membership function that, after parameter adjustment, can adapt to the current fused image features, and is used to accurately convert the index measurement values ​​extracted from the current image into membership degrees.

[0084] There are several ways to adjust the parameters of the reference membership function to obtain the target membership function corresponding to the security assessment index based on the features of the security assessment index contained in the fused image. These methods can be as follows: (1) When the material state includes the crack region, the parameters of the trapezoidal membership function are adjusted based on the proportion of the crack region in the fused image to obtain the target membership function corresponding to the material state.

[0085] In this context, the cracked area can be understood as the pixel region in the fused image that represents cracks on the surface of glass or other panel materials, identified through edge detection, morphological processing, and other methods. The cracked area proportion can be understood as the ratio of the total number of pixels in the cracked area to the total number of pixels in the entire image or curtain wall area, used to characterize the overall severity of the cracks.

[0086] The process involves detecting and extracting crack regions from the fused image, calculating the proportion of cracked regions, and then dynamically adjusting the four parameters of the trapezoidal membership function based on this proportion to obtain the target membership function corresponding to the material state. For example, when the proportion of cracked regions is high, the threshold representing the safe zone in the trapezoidal function can be appropriately lowered, while the sensitivity representing the dangerous zone can be increased, thus enabling the membership function to more accurately reflect the current severity of the cracks. It should be noted that crack length is a crucial indicator for assessing curtain wall safety, but the visual representation of cracks in different images may vary due to factors such as lighting and angle. By dynamically adjusting the trapezoidal membership function parameters based on the proportion of cracked regions, the assessment model can adaptively adapt to crack characteristics under different image conditions, improving the robustness of the assessment.

[0087] (2) When the material state includes the bonding strength, the parameters of the Gaussian membership function are calculated based on the stress field distribution corresponding to the bonding strength in the fused image, and the target membership function corresponding to the material state is obtained.

[0088] Bond strength can be understood as a mechanical performance indicator of adhesive materials such as structural adhesives for curtain walls, reflecting their ability to resist external forces such as shear and tension. A decrease in bond strength usually indicates that the structural adhesive has aged or failed. The stress field distribution corresponding to bond strength can be understood as the temperature field distribution information extracted from infrared thermograms. Since stress concentration areas are often accompanied by temperature anomalies, the state of bond strength can be indirectly inferred through the distribution characteristics of the temperature field.

[0089] Specifically, stress field distribution data corresponding to the bonding area can be extracted from infrared thermograms. The mean and standard deviation of this distribution are calculated, and then used as parameters of a Gaussian membership function to describe the deviation of the current bond strength from the ideal state, thus obtaining the target membership function corresponding to the material state. It should be noted that bond strength is difficult to observe directly through visible light images, while infrared thermograms can sensitively reflect the internal stress state of the material. By dynamically calculating the parameters of the Gaussian membership function based on the stress field distribution, a quantitative assessment of latent defects can be achieved.

[0090] (3) When the structural information includes structural deformation and inter-layer displacement, the parameters of the triangle membership function are adjusted based on the degree of structural deformation and the amount of inter-layer displacement in the fused image to obtain the target membership function corresponding to the structural information.

[0091] Structural deformation can be understood as the geometrical change of the curtain wall supporting structure under load, such as bending and twisting. The degree of structural deformation can be understood as the quantified value of the structural deformation angle or displacement obtained through image analysis. Inter-story displacement can be understood as the relative horizontal displacement of the curtain wall between adjacent floors. The amount of inter-story displacement can be understood as the relative inter-story displacement value calculated through methods such as image feature point tracking or edge detection.

[0092] Specifically, the angular values ​​of structural deformation and the displacement of inter-story drift can be extracted from the fused image. Based on these measured values, the parameters of the three vertices of the triangle membership function are dynamically adjusted, allowing the membership function to more accurately reflect the severity of the current structural deformation. It should be noted that structural deformation and inter-story drift are key indicators of curtain wall structural safety, and their safety thresholds may vary depending on factors such as building height and structural form. By dynamically adjusting the function parameters based on measured values, adaptive assessments for different buildings and working conditions can be achieved.

[0093] (4) When the environmental information includes ultraviolet intensity, the parameters of the trapezoidal membership function are adjusted based on the ultraviolet intensity in the fused image to obtain the target membership function corresponding to the environmental information.

[0094] Among them, ultraviolet intensity can be understood as the intensity level of the ultraviolet part of solar radiation, which is the main environmental factor causing the aging of organic materials such as sealants and structural adhesives.

[0095] Specifically, the current ultraviolet (UV) intensity value can be extracted from the fused image or obtained from an external data source, and the parameters of the trapezoidal membership function can be dynamically adjusted based on this value. For example, when the UV intensity is high, the sensitivity to material aging risks can be appropriately increased. It should be noted that although environmental factors do not directly manifest as damage to the curtain wall, they have a significant impact on the long-term durability of the materials. By incorporating environmental information into the assessment system and dynamically adjusting its membership function, a more comprehensive assessment of the curtain wall's safety risks can be achieved.

[0096] S403. Correct the membership results obtained by the target membership function through the adjustment function.

[0097] The adjustment function can be understood as an auxiliary function that corrects the initial membership results. For example, it can adjust the membership by weighting or shifting based on external factors such as lighting conditions and weather conditions during image acquisition, thereby improving the robustness of the evaluation results. The membership results can be understood as the numerical values ​​of the degree of membership of each security evaluation index to different security levels, calculated through the target membership function.

[0098] The method of correcting the membership results calculated by adjusting the target membership function can specifically include: obtaining environmental parameters of the current image acquisition, such as light intensity, weather conditions, and shooting angle; determining correction coefficients based on these parameters; assigning correction weights to each security assessment indicator based on the correction coefficients; setting the correction weights according to the detection accuracy of the indicator—the higher the detection accuracy (e.g., structural deformation), the closer the correction weight is to 1; the lower the detection accuracy (e.g., ultraviolet intensity), the lower the correction weight is appropriately reduced; secondly, multiplying the initial membership results of each indicator by the corresponding correction weight to obtain the preliminary correction result; finally, normalizing the preliminary correction result to ensure that the sum of the membership degrees of each security level after correction is 1, thus obtaining the final corrected membership result.

[0099] It should be noted that the core function of the adjustment function is to reduce the interference of indicators with low detection accuracy on the evaluation results. The correction process must strictly follow the normalization constraints to avoid disrupting the fuzzy logic of membership. The setting of the correction weight can be dynamically adjusted in combination with the accuracy parameters of the on-site testing equipment to ensure the rationality of the correction.

[0100] S404. Associate the corrected membership results with the evaluation weights corresponding to the security evaluation indicators, and integrate them in a preset order to form a fuzzy relation matrix. The fuzzy relation matrix contains information on the membership degree of the security evaluation indicators to different security levels.

[0101] The corrected membership results can be understood as the final membership values ​​of each safety assessment indicator belonging to different safety levels after adjustment and normalization using an adjustment function. The preset order can be understood as the pre-determined arrangement of the safety assessment indicators, such as the classification order according to structural information, material state, and environmental information, and the specific order of indicators within each category. The fuzzy relation matrix can be understood as a two-dimensional matrix composed of the corrected membership results. The number of rows in the matrix equals the number of safety assessment indicators, and the number of columns equals the number of safety levels. Each element in the matrix represents the degree of membership of the safety assessment indicator to the corresponding safety level. Membership degree information can be understood as the numerical values ​​of each element in the fuzzy relation matrix, typically ranging from 0 to 1. A larger value indicates a higher degree of membership of the indicator to that safety level.

[0102] The method of associating the corrected membership results with the evaluation weights corresponding to the security assessment indicators and integrating them into a fuzzy relation matrix in a preset order can specifically include: arranging the corrected membership results of each indicator into a row according to a preset indicator order, forming a row vector containing the membership degree of the indicator to each security level. Then, stacking all the indicator row vectors from top to bottom in the same indicator order to form a complete two-dimensional matrix, which is the fuzzy relation matrix. The calculated evaluation weights of each security assessment indicator can then be associated with the indicator row vectors in the fuzzy relation matrix.

[0103] After obtaining the fuzzy relation matrix, the semantic features of the fused image can be evaluated based on the fuzzy relation matrix to obtain the evaluation score of the curtain wall. Specifically, this can include: determining the membership data corresponding to the semantic features of the fused image based on the semantic features of the fused image and the security evaluation index in the fuzzy relation matrix; then, combining the evaluation weights corresponding to the security evaluation index, weighted summing of the membership data to obtain a reference score; finally, based on the reference score, and combined with preset quantization rules, obtaining the evaluation score of the curtain wall, which is positively correlated with the safety level of the curtain wall.

[0104] In this context, membership data can be understood as membership values ​​extracted from the fuzzy relation matrix that correspond to the states of each indicator in the currently fused image. Reference scores can be understood as intermediate results obtained by weighted summation of the membership data. Preset quantization rules can be understood as rules for converting the reference score vector into a single numerical value. For example, each safety level can be assigned a quantization score (e.g., safe = 1.0, sub-safe = 0.5, dangerous = 0.0), and then each component in the reference score vector is multiplied by its corresponding quantization score and summed to obtain the final evaluation score.

[0105] Optionally, the method of determining the membership data corresponding to the semantic features of the fused image based on the security assessment indicators in the fuzzy relation matrix can specifically include: parsing the measured values ​​of each security assessment indicator from the semantic features of the fused image, and then calculating the membership data of each indicator from the target membership function corresponding to the indicator based on the measured value of each indicator; or, directly predicting the membership of each indicator to different security levels end-to-end through a deep learning model, and using the prediction results as membership data.

[0106] Optionally, a weighted summation of membership data, combining the assessment weights corresponding to the safety assessment indicators, can be used to obtain a reference score. Specifically, this can include: multiplying the membership row vector of each indicator by its weight to obtain the indicator's contribution to the reference score; then summing the contribution vectors of all indicators to obtain the final reference score vector. Each element in this vector represents the comprehensive degree to which the curtain wall belongs to the corresponding safety level.

[0107] Optionally, the evaluation score of the curtain wall can be obtained by converting the reference score into a pre-defined quantification rule. Specifically, this can include: pre-setting a quantification score for each safety level, for example, 1.0 point for the first safety level (safe), 0.5 points for the second safety level (sub-safe), and 0.0 points for the hazardous level; then multiplying each component in the reference score vector by its corresponding quantification score, and summing the products to obtain the final evaluation score. This evaluation score is between 0 and 1, with a higher score indicating a safer curtain wall.

[0108] It should be noted that by using the fuzzy comprehensive evaluation method, multiple qualitative or semi-quantitative safety indicators are combined into a single quantitative evaluation score, which not only preserves the original information of each indicator but also achieves a quantitative description of the overall safety status of the curtain wall.

[0109] 105. Determine the safety level of the curtain wall based on the assessment score and assessment threshold.

[0110] The assessment threshold can be understood as a critical score used to distinguish different safety levels. The assessment threshold can include a first assessment threshold and a second assessment threshold, with the first assessment threshold being greater than the second assessment threshold. The first assessment threshold can be used to define the boundary between safety and sub-safety, while the second assessment threshold can be used to define the boundary between sub-safety and danger. Safety levels can include three levels: a first safety level (representing safety), a second safety level (representing sub-safety), and a danger level (representing danger), which can correspond to different safety states and subsequent handling measures, respectively.

[0111] The first safety level can be understood as the curtain wall being in a safe state, with all indicators within the normal range and no obvious safety hazards. Subsequent inspections can be carried out according to the usual schedule. The second safety level can be understood as the curtain wall being in a sub-safe state, with certain safety hazards or abnormal indicators, requiring closer attention and a shorter inspection cycle. The danger level can be understood as the curtain wall being in a dangerous state, with serious safety hazards and the possibility of a safety accident at any time, requiring immediate countermeasures.

[0112] The method for determining the safety level of a curtain wall based on assessment scores and thresholds is as follows: First, obtain curtain wall type information and environmental impact parameters, including wind load-related parameters. Then, based on the curtain wall type information and environmental impact parameters, calculate a first assessment threshold and a second assessment threshold. If the assessment score is greater than or equal to the first assessment threshold, the curtain wall is determined to be at the first safety level. If the assessment score is greater than or equal to the second assessment threshold but less than the first assessment threshold, the curtain wall is determined to be at the second safety level. If the assessment score is less than the second assessment threshold, the curtain wall is determined to be at a dangerous level, and a secondary inspection process is triggered.

[0113] The curtain wall type information can be understood as classification information reflecting the structural form and stress characteristics of the curtain wall. Different types of curtain walls have different sensitivities and tolerances to safety risks. For example, it can include types such as exposed frame curtain walls, concealed frame curtain walls, and all-glass curtain walls, each with a different safety factor. Environmental impact parameters can be understood as quantitative parameters reflecting the external environmental conditions of the curtain wall, including various factors such as wind load, temperature, humidity, and ultraviolet radiation, which affect the actual stress state and aging rate of the curtain wall. Wind load-related parameters can be understood as key parameters reflecting the wind pressure acting on the curtain wall, which can be calculated based on factors such as the basic wind pressure of the building's location, building height, and shape coefficient.

[0114] Optionally, information on the curtain wall type and environmental impact parameters can be obtained, including wind load-related parameters. Specifically, this can include: reading the curtain wall type information of the currently monitored building from the system configuration file, for example, through manual input or import from the building information model; obtaining real-time meteorological data such as wind speed and direction at the building's location by connecting to a meteorological data service interface, and calculating wind load-related parameters by combining this data with the building's height and shape coefficient; or, collecting real-time wind speed data using wind speed sensors installed on the building's roof as the basis for wind load calculation.

[0115] Optionally, there are multiple ways to calculate the first and second evaluation thresholds based on curtain wall type information and environmental impact parameters. For example, the type coefficient can be determined according to the curtain wall type: 0.8 for exposed frame curtain walls, 0.9 for concealed frame curtain walls, and 1.0 for all-glass curtain walls. Alternatively, the wind load level coefficient can be determined according to wind load-related parameters. A preset dynamic threshold calculation formula can be used to combine the basic threshold with the type coefficient and the wind load level coefficient to obtain the first and second evaluation thresholds applicable to the current testing scenario. For example, the first evaluation threshold can be based on a basic value of 0.7, with the weighted terms of the type coefficient and the wind load level coefficient added; the second evaluation threshold can be based on a basic value of 0.4, with the weighted term of the type coefficient subtracted and the weighted term of the wind load level coefficient added, and so on.

[0116] Optionally, if the assessment score is less than the second assessment threshold, the curtain wall's safety level will be determined as hazardous, triggering a secondary inspection process. Specifically, this may include: automatically marking the current curtain wall's safety level as hazardous when the assessment score is below the second assessment threshold, and immediately triggering a secondary inspection process. The secondary inspection process may include: controlling a drone to perform a more detailed scan of the hazardous area to obtain higher-resolution image data; retrieving historical inspection data for the area for comparative analysis; pushing the inspection results and alarm information to the management personnel's mobile terminals or the monitoring center; and, if necessary, suggesting immediate on-site manual verification to further confirm the authenticity of the hazardous status and avoid false alarms due to errors in a single inspection.

[0117] Optionally, in some implementations, time series analysis can be performed on the curtain wall safety level results within a continuous testing cycle to calculate the safety level trend, which changes over time. If the safety level trend indicates that the safety level change exceeds a preset fluctuation threshold, the curtain wall safety status is determined to be fluctuating, and a safety risk alarm is issued. If the safety level trend indicates that the safety level remains stable and is at the first safety level, the regular testing frequency is maintained according to a preset cycle. If the safety level trend indicates that the safety level remains stable and is at the second safety level, the testing cycle is shortened and continuous monitoring is carried out, and a safety risk warning is issued.

[0118] The safety level trend can be understood as the pattern of safety status evolution over time obtained by analyzing the safety level results of multiple monitoring cycles, such as continuous decline, continuous increase, or stable fluctuation. The preset fluctuation threshold can be understood as a critical value used to judge whether the safety level change is significant; when the change in safety level within adjacent cycles or a period of time exceeds this threshold, the safety status is considered to have fluctuated significantly. The curtain wall safety status can be understood as a dynamic description of the overall safety status of the curtain wall after comprehensively considering the current safety level and historical trends; for example, it may include stable, fluctuating, or continuously deteriorating states. A safety risk alarm can be understood as an alert issued when the safety status shows abnormal fluctuations or continuous deterioration, prompting management personnel to pay immediate attention and take appropriate measures. A safety risk warning can be understood as a reminder issued when the safety status is in a sub-safe and stable state, prompting management personnel to strengthen monitoring but not to take immediate emergency measures.

[0119] Optionally, time series analysis can be performed on the curtain wall safety level results within consecutive inspection cycles to calculate the safety level trend. Specifically, this can include: extracting the evaluation scores and safety level results of the curtain wall for the most recent N consecutive inspection cycles (N≥6), and constructing a time series dataset in the order of inspection time; secondly, performing linear fitting on the evaluation score time series to obtain the slope k of the fitted curve. The slope represents the rate of change of the evaluation score, where > indicates an increase in score and improvement in safety status, < indicates a decrease in score and deterioration in safety status, and k≈0 indicates stable score and no significant change in safety status; finally, combining the fitted slope and the frequency of safety level changes, a quantitative value of the safety level trend is calculated. If the absolute value of the slope of the score change is >0.5, or if the safety level changes 2 or more times within consecutive inspection cycles, the safety level trend is determined to be a fluctuating trend; otherwise, it is a stable trend.

[0120] Optionally, if the safety level trend indicates that the change in safety level exceeds a preset fluctuation threshold, the system determines that the curtain wall's safety status is fluctuating and issues a safety risk alarm. Specifically, this can include: when the quantitative value of the safety level trend exceeds the preset fluctuation threshold (e.g., the absolute value of the fitting slope > 1.0, or a single leap in safety level), the system automatically determines that the curtain wall's safety status is fluctuating; immediately generates a safety risk alarm message, including details of the curtain wall's safety level fluctuation, the changing trend of core risk indicators, and the abnormal detection cycle; pushes the alarm to curtain wall maintenance and management personnel, building owners, and other relevant entities via SMS, email, and background pop-ups; simultaneously, marks the curtain wall as a high-risk fluctuation object in the system, automatically plans on-site verification routes, and reminds relevant personnel to conduct timely on-site inspections to investigate the causes of abnormal safety level fluctuations.

[0121] Optionally, if the safety level trend indicates that the safety level remains stable and is at the first safety level, the system maintains the regular inspection frequency according to the preset cycle. Specifically, this may include: when the safety level trend is stable and the safety level is the first safety level for N consecutive inspection cycles, the system determines that the curtain wall is in a stable and safe state; the preset regular inspection frequency remains unchanged, with fixed acquisition equipment performing image acquisition once a day and mobile acquisition equipment performing full curtain wall scanning inspection once a quarter; at the same time, the curtain wall is included in the regular monitoring log and is only subject to special review during the annual building safety inspection, without the need to increase the inspection frequency.

[0122] Optionally, if the safety level trend indicates that the safety level remains stable and is at the second safety level, the detection cycle will be shortened and continuous monitoring will be implemented, along with a safety risk warning. Specifically, this may include: when the safety level trend is stable and the safety level is at the second safety level for N consecutive detection cycles, the system determines the curtain wall's safety status as stable sub-safe; pushing safety risk warning information to management personnel, including the current safety status of the curtain wall, core minor risk indicators, and notifications of detection cycle adjustments; secondly, adjusting the detection frequency, increasing the image acquisition frequency of fixed acquisition equipment to twice daily (e.g., once in the morning and once in the evening), and shortening the full curtain wall scanning frequency of mobile acquisition equipment to once a month, focusing on monitoring core risk indicator areas; finally, establishing a special monitoring log in the system to record the assessment score and risk indicator changes for each detection, and immediately escalating to a safety risk alarm if a deteriorating trend in risk indicators is detected.

[0123] As can be seen from the above, in this embodiment of the application, after obtaining a set of original images of the curtain wall including at least one original image, the original images in the set can be matched to obtain matched image pairs, and the matched image pairs can be weighted and fused to obtain a fused image. Semantic features of the fused image are extracted using a preset semantic extraction model, and the evaluation weights corresponding to the safety evaluation indicators of the curtain wall are determined. The safety evaluation indicators include the structural information of the curtain wall, the material state of the curtain wall, and the environmental information of the environment in which the curtain wall is located. The semantic features of the fused image are evaluated based on a fuzzy relation matrix to obtain an evaluation score for the curtain wall. Based on the evaluation score and the evaluation threshold, the safety assessment of the curtain wall is determined. Safety Level: This solution integrates complementary information from multiple image sources, enhancing the completeness and accuracy of curtain wall condition representation. This overcomes the limitations of traditional single-image detection, which suffers from incomplete information and blurred details. It enables comprehensive capture of surface defects and internal hidden dangers within the curtain wall. Furthermore, since the fuzzy relation matrix is ​​constructed from the evaluation weights, membership functions, and adjustment functions corresponding to the safety assessment indicators, it avoids the subjectivity and bias of manual experience-based judgments, achieving a quantitative assessment of the curtain wall's safety status. Therefore, it improves the comprehensiveness, objectivity, and accuracy of curtain wall safety detection, providing a reliable quantitative basis for curtain wall safety maintenance and risk management.

[0124] To better implement the above methods, this application also provides a curtain wall safety detection system, which can be integrated into electronic devices, such as servers or terminals. The terminal may include tablet computers, laptops, and / or personal computers.

[0125] For example, such as Figure 3As shown, the curtain wall safety detection system may include an image processing unit 301, a weight determination unit 302, an evaluation score determination unit 303, and a safety level determination unit 304, as follows: (1) Image processing unit 301 is used to acquire an original image set, the original image set including at least one original image; to match the original images in the original image set to obtain a matched image pair, and to perform weighted fusion on the matched image pair to obtain a fused image.

[0126] (2) Weight determination unit 302 is used to extract the semantic features of the fused image through a preset semantic extraction model and determine the evaluation weights corresponding to the safety evaluation indicators of the curtain wall. The safety evaluation indicators include the structural information of the curtain wall, the material state of the curtain wall and the environmental information of the environment in which the curtain wall is located.

[0127] (3) Evaluation score determination unit 303 is used to evaluate the semantic features of the fused image based on the fuzzy relation matrix to obtain the evaluation score of the curtain wall. The fuzzy relation matrix is ​​constructed by the evaluation weights corresponding to the security evaluation indicators, the membership functions corresponding to the security evaluation indicators, and the adjustment functions. The evaluation score indicates the security score of the curtain wall.

[0128] (4) Safety level determination unit 304 is used to determine the safety level of the curtain wall based on the evaluation score and the evaluation threshold.

[0129] In practice, each of the above units can be implemented as an independent entity or can be arbitrarily combined to be implemented as the same or several entities. For the specific implementation of each of the above units, please refer to the previous method embodiments, which will not be repeated here.

[0130] This application also provides an electronic device, such as... Figure 4 As shown, it illustrates a structural schematic diagram of the electronic device involved in the embodiments of this application, specifically: The electronic device may include components such as a processor 401 with one or more processing cores, a memory 402 with one or more computer-readable storage media, a power supply 403, and an input unit 404. Those skilled in the art will understand that... Figure 4 The electronic device structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements. Wherein: The processor 401 is the control center of the electronic device, connecting various parts of the device via various interfaces and lines. It executes software programs and / or modules stored in the memory 402, and calls data stored in the memory 402, to perform various functions and process data. Optionally, the processor 401 may include one or more processing cores; preferably, the processor 401 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into the processor 401.

[0131] The memory 402 can be used to store software programs and modules. The processor 401 executes various functional applications and data processing by running the software programs and modules stored in the memory 402. The memory 402 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the electronic device, etc. In addition, the memory 402 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory 402 may also include a memory controller to provide the processor 401 with access to the memory 402.

[0132] The electronic device also includes a power supply 403 that supplies power to the various components. Preferably, the power supply 403 can be logically connected to the processor 401 through a power management system, thereby enabling functions such as charging, discharging, and power consumption management through the power management system. The power supply 403 may also include one or more DC or AC power supplies, recharging systems, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components.

[0133] The electronic device may also include an input unit 404, which can be used to receive input digital or character information, and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control.

[0134] Although not shown, the electronic device may also include a display unit, etc., which will not be described in detail here. Specifically, in this embodiment, the processor 401 in the electronic device loads the executable files corresponding to the processes of one or more applications into the memory 402 according to the following instructions, and the processor 401 runs the applications stored in the memory 402 to realize various functions.

[0135] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by instructions, or by instructions controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0136] Therefore, embodiments of this application provide a computer-readable storage medium storing a plurality of instructions that can be loaded by a processor to execute the steps in any of the curtain wall safety detection methods provided in embodiments of this application.

[0137] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.

[0138] The computer-readable storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0139] Since the instructions stored in the computer-readable storage medium can execute the steps in any of the curtain wall safety detection methods provided in the embodiments of this application, the beneficial effects that any of the curtain wall safety detection methods provided in the embodiments of this application can achieve can be realized. For details, please refer to the previous embodiments, which will not be repeated here.

[0140] According to one aspect of this application, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the methods provided in the various alternative implementations of the above-described curtain wall security detection.

[0141] The above provides a detailed description of a curtain wall safety detection method and system provided in the embodiments of this application. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for detecting the safety of curtain walls, characterized in that, include: Obtain a set of original images of the curtain wall, the set of original images including at least one original image; The original images in the original image set are matched to obtain matched image pairs, and the matched image pairs are weighted and fused to obtain a fused image. The semantic features of the fused image are extracted by a preset semantic extraction model, and the evaluation weights corresponding to the safety evaluation indicators of the curtain wall are determined. The safety evaluation indicators include the structural information of the curtain wall, the material state of the curtain wall, and the environmental information of the environment in which the curtain wall is located. The semantic features of the fused image are evaluated based on the fuzzy relation matrix to obtain the evaluation score of the curtain wall. The fuzzy relation matrix is ​​constructed from the evaluation weights corresponding to the security evaluation indicators, the membership functions corresponding to the security evaluation indicators, and the adjustment functions. The evaluation score indicates the security score of the curtain wall. The safety level of the curtain wall is determined based on the assessment score and assessment threshold.

2. The curtain wall safety testing method according to claim 1, characterized in that, The original image is a first visible light image of the curtain wall acquired by a first acquisition device, a second visible light image of the curtain wall acquired by a second acquisition device, or an infrared thermal image of the curtain wall. The first acquisition device is a fixed acquisition device, and the second acquisition device is a mobile acquisition device. Matching the original images in the original image set to obtain matched image pairs includes: A feature extraction algorithm is used to extract features from the first visible light image to obtain the first multi-directional features corresponding to the first visible light image; A feature extraction algorithm is used to extract features from the second visible light image to obtain the second multi-directional features corresponding to the second visible light image; Based on the first multi-directional features and the second multi-directional features, the first visible light image and the second visible light image are matched to obtain a reference matching pair; The reference matching pair is matched and aligned with the infrared thermal image to obtain a matched image pair.

3. The curtain wall safety testing method according to claim 1, characterized in that, The step of weighted fusion of the matched image pairs to obtain the fused image includes: Wavelet decomposition is performed on the matched image pair to obtain the signal quality and gradient information of the matched image in the matched image pair; The fusion weights of the images are determined based on the signal quality and gradient information of the matched images in the matched image pairs, and weighted fusion is performed based on the fusion weights to obtain the fused image.

4. The curtain wall safety testing method according to claim 3, characterized in that, Wavelet decomposition is performed on the matched image pair to obtain the signal quality and gradient information of the matched image in the matched image pair, including: The matching image is decomposed into multiple levels using wavelet basis functions to extract low-frequency and high-frequency information. The low-frequency information indicates the overall contour information of the matching image, and the high-frequency information indicates the edge and detail information of the matching image. The pixel mean and noise standard deviation of the effective region in the matched image are calculated, and the signal quality of the matched image is calculated based on the pixel mean and noise standard deviation. Calculate the gradient magnitude of the matching image to obtain the gradient information of the matching image, wherein the gradient information indicates the edge sharpness of the matching image; The process of determining the fusion weights of the images based on the signal quality and gradient information of the matched images in the matched image pair, and performing weighted fusion based on the fusion weights to obtain the fused image, includes: Based on the low-frequency information of the matched image, the low-frequency fusion weights corresponding to the matched image are calculated according to the signal quality. Based on the high-frequency information of the matched image, the high-frequency fusion weights corresponding to the matched image are calculated by combining signal quality and gradient information; Based on the low-frequency fusion weights corresponding to the low-frequency information of the matched images and the high-frequency fusion weights corresponding to the high-frequency information of the matched images, the matched image pairs are fused using inverse wavelet transform to obtain the fused image.

5. The curtain wall safety testing method according to claim 4, characterized in that, Before evaluating the semantic features of the fused image based on the fuzzy relation matrix to obtain the evaluation score of the curtain wall, the method further includes: Based on the structural information, material condition, and environmental information of the curtain wall, and corresponding safety assessment indicators, reference membership functions are set, including trapezoidal membership functions, Gaussian membership functions, and triangular membership functions. Based on the features of the security assessment indicators contained in the fused image, the parameters of the reference membership function are adjusted to obtain the target membership function corresponding to the security assessment indicator. The membership results obtained by the target membership function are corrected by adjusting the function. The corrected membership results are associated with the evaluation weights corresponding to the security evaluation indicators, and integrated in a preset order to form a fuzzy relation matrix. The fuzzy relation matrix contains information on the membership degree of the security evaluation indicators to different security levels.

6. The curtain wall safety testing method according to claim 5, characterized in that, The step of adjusting the parameters of the reference membership function based on the features of the security assessment index contained in the fused image to obtain the target membership function corresponding to the security assessment index includes: When the material state includes a cracked region, the parameters of the trapezoidal membership function are adjusted based on the proportion of the cracked region in the fused image to obtain the target membership function corresponding to the material state. When the material state includes the bonding strength, the parameters of the Gaussian membership function are calculated based on the stress field distribution corresponding to the bonding strength in the fused image, and the target membership function corresponding to the material state is obtained. When the structural information includes structural deformation and interlayer displacement, the parameters of the triangle membership function are adjusted based on the degree of structural deformation and the amount of interlayer displacement in the fused image to obtain the target membership function corresponding to the structural information. When the environmental information includes ultraviolet intensity, the parameters of the trapezoidal membership function are adjusted based on the ultraviolet intensity in the fused image to obtain the target membership function corresponding to the environmental information.

7. The curtain wall safety testing method according to claim 5, characterized in that, The evaluation of the semantic features of the fused image based on the fuzzy relation matrix to obtain the evaluation score of the curtain wall includes: Based on the semantic features of the fused image and the security evaluation index in the fuzzy relation matrix, the membership data corresponding to the semantic features of the fused image are determined. By combining the evaluation weights corresponding to the security evaluation indicators, the membership data are weighted and summed to obtain a reference score; The evaluation score of the curtain wall is obtained by converting the reference score into a preset quantification rule. The evaluation score is positively correlated with the safety level of the curtain wall.

8. The curtain wall safety testing method according to claim 1, characterized in that, The safety level of the curtain wall is determined based on the assessment score and the assessment threshold. The assessment threshold includes a first assessment threshold and a second assessment threshold, wherein the first assessment threshold is greater than the second assessment threshold, including: Obtain information on the curtain wall type and environmental impact parameters, including wind load-related parameters; Based on the curtain wall type information and environmental impact parameters, the first evaluation threshold and the second evaluation threshold are calculated. If the evaluation score is greater than or equal to the first evaluation threshold, then the safety level of the curtain wall is determined to be the first safety level; If the evaluation score is greater than or equal to the second evaluation threshold and less than the first evaluation threshold, then the safety level of the curtain wall is determined to be the second safety level. If the evaluation score is less than the second evaluation threshold, the safety level of the curtain wall is determined to be dangerous, and a secondary inspection process is triggered.

9. The curtain wall safety testing method according to claim 8, characterized in that, The method further includes: A time series analysis is performed on the curtain wall safety level results within a continuous detection period to calculate the safety level trend, which changes over time. If the safety level trend indicates that the change in safety level exceeds the preset fluctuation threshold, the safety status of the curtain wall is determined to be fluctuating, and a safety risk alarm is fed back. If the security level trend indicates that the security level remains stable and is at the first security level, then the normal detection frequency is maintained according to the preset cycle. If the security level trend indicates that the security level remains stable and is at the second security level, the detection cycle will be shortened and continuous monitoring will be carried out, and a security risk warning will be issued.

10. A curtain wall safety detection system, characterized in that, include: An image processing unit is configured to acquire an original image set, the original image set including at least one original image; match the original images in the original image set to obtain matched image pairs; and perform weighted fusion on the matched image pairs to obtain a fused image. The weight determination unit is used to extract the semantic features of the fused image through a preset semantic extraction model, and determine the evaluation weights corresponding to the safety evaluation indicators of the curtain wall. The safety evaluation indicators include the structural information of the curtain wall, the material state of the curtain wall, and the environmental information of the environment in which the curtain wall is located. The evaluation score determination unit is used to evaluate the semantic features of the fused image based on the fuzzy relation matrix to obtain the evaluation score of the curtain wall. The fuzzy relation matrix is ​​constructed from the evaluation weights corresponding to the security evaluation indicators, the membership functions corresponding to the security evaluation indicators, and the adjustment functions. The evaluation score indicates the security score of the curtain wall. The safety level determination unit is used to determine the safety level of the curtain wall based on the evaluation score and the evaluation threshold.

Images