External wall panel falling monitoring system and method based on visual image recognition

Abstract

The invention relates to the technical field of image recognition, in particular to an exterior wall panel falling monitoring system and method based on visual image recognition, and the system comprises an image integration module, a pixel potential energy calculation module, a tension gradient analysis module, a falling risk recognition module and a monitoring and early warning module. According to the method, the spatial calibration relation between the images is constructed through continuous frame gray extraction and texture direction angle measurement and calculation, so that the surface texture keeps pixel-level matching in a dynamic scene, the response sensitivity of microstructure deformation is improved through combined calculation of the gray change rate and the reflection intensity deviation, and pixel potential energy distribution is established to quantify the local stress state; the continuous tracking of the strain trend is realized, the joint judgment of the potential energy deviation of adjacent frames and the direction consistency deviation can accurately position a potential crack or loose area, misjudgment caused by external illumination, reflection and view angle difference is avoided, and the early recognition and trend evaluation of the multi-period outer wall falling risk are realized. And the stability of the monitoring process and the result credibility are obviously improved.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a system and method for monitoring the detachment of exterior wall panels based on visual image recognition. Background Technology

[0002] Image recognition technology is a branch of computer vision, primarily researching the use of computers to analyze and recognize acquired image information, thereby achieving automatic judgment and classification of target objects, scenes, or states. The core aspects of this technology include image acquisition, feature extraction, pattern recognition, and target detection. Its system content covers the entire process from data input from image acquisition devices to algorithmic analysis and processing of image content, and it is widely used in scenarios such as security monitoring, industrial inspection, medical diagnosis, and intelligent transportation.

[0003] Traditional exterior wall panel detachment monitoring systems acquire images of the wall surface using cameras installed on the building facade. These images are then analyzed using fixed image comparison or feature recognition methods to determine if there is a risk of detachment or displacement. Traditional systems typically involve periodically acquiring images and performing manual comparisons, or using image analysis methods based on edge detection and texture changes to identify surface anomalies, thereby monitoring the structural condition of the wall.

[0004] Traditional external wall panel monitoring relies on static image comparison, which lacks processing for pixel consistency between multiple frames. Image matching errors increase with changes in shooting position and lighting, making it difficult to maintain stable surface texture features. Manual comparison relies on experience judgment, which is insufficient for responding to early stress accumulation or local deformation. Edge or texture differences are easily affected by reflection interference in complex wall environments, causing early crack signals to be masked by noise. The monitoring results have low accuracy in identifying minute deformations, making it impossible to continuously track and dynamically warn of wall panel detachment trends, thus affecting the timeliness and accuracy of risk assessment. Summary of the Invention

[0005] To address the shortcomings of traditional external wall panel monitoring methods that rely on static image comparison, lack processing for pixel consistency across multiple images, and suffer from image matching errors that increase with shooting location and lighting conditions, making it difficult to maintain stable surface texture features, manual comparison relies on experience-based judgment and is insufficient in responding to early stress accumulation or local deformation, and edge or texture differences are easily affected by reflection interference in complex wall environments, causing early crack signals to be masked by noise, resulting in low accuracy in identifying minute deformations and an inability to continuously track and dynamically warn of wall panel detachment trends, thus affecting the timeliness and accuracy of risk assessment, this invention provides an external wall panel detachment monitoring system and method based on visual image recognition. The technical solution is as follows: On the one hand, a visual image recognition-based monitoring system for monitoring the detachment of exterior wall panels is provided. This system includes: The image integration module acquires a continuous frame image sequence of the exterior wall surface through a camera device, performs grayscale value extraction, measures the texture direction angle, detects the light reflection intensity, uses a bilinear interpolation algorithm to spatially calibrate the image sequence, generates a standardized pixel parameter set, and transmits it to the pixel potential energy calculation module. The pixel potential energy calculation module calls the standardized pixel parameter set, calculates the gray level change rate parameter, performs local gradient direction calculation, analyzes the reflection intensity deviation and calculates the pixel potential energy value, constructs the pixel potential energy distribution, and transmits it to the tension gradient analysis module. The tension gradient analysis module calls the pixel potential energy distribution, calculates the potential energy deviation of pixels at the same position in adjacent frames, compares the deviation result with a preset dynamic threshold, marks stress anomaly points when the threshold is exceeded, generates an abnormal area distribution map, and transmits it to the detachment risk identification module. The detachment risk identification module receives the abnormal area distribution map, calculates the directional consistency parameter, uses the least squares method to process the directional relationship between adjacent points, compares the deviation of the current directional consistency parameter with the previous cycle, and when it exceeds the preset trigger threshold, generates the wall panel detachment risk level result and transmits it to the monitoring and early warning module.

[0006] As a further embodiment of the present invention, the standardized pixel parameter set includes grayscale value, texture direction angle and light reflection intensity; the pixel potential energy distribution includes grayscale gradient change rate, local gradient direction distribution and reflected light intensity deviation value; the abnormal area distribution map includes stress anomaly point coordinates, pixel stress deviation amplitude and abnormal area boundary contour; and the wall panel detachment risk level result includes directional consistency coefficient, periodic change deviation value and risk level threshold coefficient.

[0007] As a further aspect of the present invention, the image integration module includes: The image acquisition submodule acquires a continuous frame image sequence of the exterior wall surface through a camera device, detects the pixel array time index of the frame image, judges abnormal frames based on the fluctuation trend of inter-frame displacement difference, calculates the inter-frame pixel displacement difference and adjusts the frame sequence index, and generates the frame sequence index matrix value. The grayscale and texture extraction submodule extracts pixel grayscale data and counts grayscale differences based on the frame sequence index matrix value, calculates local pixel gradient direction and determines direction angle, detects brightness change reflection areas and calculates reflection intensity, and generates grayscale texture direction intensity set. The spatial calibration submodule compares the pixel coordinate differences between adjacent frames based on the grayscale texture direction intensity set, calculates the spatial position offset, executes a bilinear interpolation algorithm, adjusts the spatial mapping relationship between grayscale and direction data, reconstructs pixel spatial parameters, and generates a standardized pixel parameter set.

[0008] As a further aspect of the present invention, the pixel potential energy calculation module includes: The grayscale change rate extraction submodule calls the standardized pixel parameter set to extract grayscale value data, calculates the grayscale difference between adjacent pixels, constructs a grayscale change matrix based on the coordinate difference and grayscale difference, performs discrete difference and operation on the matrix, and generates grayscale change rate parameter values. The local gradient direction calculation submodule calls the gray-level change rate parameter value, calculates the horizontal and vertical gray-level difference ratio based on the row and column coordinates of the pixel in two-dimensional space, performs arctangent operation to obtain the direction angle, and performs a weighted operation on the direction angle and gray-level change rate to generate a set of direction distribution coefficients. The reflection intensity deviation calculation submodule performs differential comparison on the reflection intensity signal of the pixel unit according to the directional distribution coefficient set, calculates the difference amplitude between the signal and the illumination reference value, performs regional integration based on the difference amplitude, and generates pixel potential energy distribution.

[0009] As a further aspect of the present invention, the tension gradient analysis module includes: The pixel potential energy deviation calculation submodule calls the pixel potential energy distribution, extracts the gray value and pixel spatial parameters of the pixels in each frame and converts them into pixel potential energy values, constructs a pixel potential energy dataset, performs inter-frame difference and normalization correction on pixels at the same position, and generates a pixel potential energy deviation value sequence. The dynamic threshold determination submodule calculates the deviation amplitude based on the pixel potential energy deviation value sequence, compares the deviation amplitude with a preset dynamic threshold, filters out pixels whose deviation amplitude exceeds the dynamic threshold, and generates stress anomaly marker data. The abnormal region generation submodule performs connectivity clustering calculations on spatially adjacent pixels based on the stress anomaly marker data and the abnormal pixel coordinate information. It calculates the pixel concentration and centroid parameters of the cluster units, divides the regions based on the continuity of the cluster unit boundaries, and generates an abnormal region distribution map.

[0010] As a further aspect of the present invention, the step of performing inter-frame difference and normalization correction on pixels at the same position refers to the pixel potential energy difference value of the same pixel in adjacent frames, and the Min-Max normalization method is used to perform numerical mapping processing on all pixel potential energy differences. The dynamic threshold is a dynamic adjustment coefficient determined based on the root mean square deviation of the full-frame pixel potential energy deviation value sequence and the statistical distribution results of the original frame.

[0011] As a further aspect of the present invention, the shedding risk identification module includes: The orientation data extraction submodule obtains the coordinates and orientation vectors of pixels in the distribution map of the abnormal area, analyzes the orientation vector group of adjacent points, calculates the orientation angle and performs smooth adjustment, and calculates the difference relationship based on the adjusted orientation data to obtain the orientation difference sequence value. The orientation consistency calculation submodule calls the orientation difference sequence value, performs least squares fitting on the orientation change rate data, calculates the orientation offset sequence based on the residual distribution, and obtains the orientation consistency parameter value after normalization by the stability function. The risk level determination submodule calculates the deviation rate based on the direction consistency parameter value and the parameter value of the previous cycle, compares the deviation rate with the set detachment risk trigger threshold, and determines the high-risk range when the deviation rate is higher than the detachment risk trigger threshold, and outputs the wall panel detachment risk level result. The threshold for triggering the risk of detachment is based on statistical analysis of wall panel detachment monitoring data. By analyzing the relationship between the directional consistency parameter value and the wall panel crack propagation rate, the threshold for triggering the risk of detachment is set as the lower limit of the directional consistency parameter deviation rate distribution range.

[0012] As a further embodiment of the present invention, the monitoring and early warning module receives the risk level result of the wall panel falling off, compares the risk level with a preset safety threshold, extracts the risk area coordinates when it is greater than the safety threshold, marks the risk area coordinates in the monitoring interface coordinate system, and triggers the external wall panel falling off safety monitoring alarm signal. The safety monitoring alarm signal for the detachment of exterior wall panels includes risk area coordinates, alarm trigger status, and warning signal identifier.

[0013] As a further aspect of the present invention, the monitoring and early warning module includes: The risk level receiving submodule acquires the risk level results of the wall panel detachment, detects the correspondence between the risk level parameters and the time label, calculates the fluctuation range of the risk level at the monitoring point, analyzes the trend of change, and generates statistical values ​​of the risk level. The threshold comparison and determination submodule calls a preset safety threshold based on the risk level statistics, compares the difference between the risk level statistics of the monitoring point and the safety threshold, calculates the threshold exceedance ratio and the number of triggers, and obtains the threshold exceedance risk determination result. The preset safety threshold is derived from the statistical analysis of the correspondence between the risk level statistics in long-term monitoring data and confirmed wall panel detachment events, by extracting the steady-state distribution range of the risk level statistics in multiple monitoring cycles before the detachment occurs. The risk area marking alarm submodule calls the above-threshold risk judgment result, extracts the coordinate index value of the above-threshold area, constructs the risk area coordinate matrix and maps it to the monitoring interface, marks the high-risk area, and generates a safety monitoring alarm signal for the fall off the exterior wall panel.

[0014] On the other hand, the method for monitoring the detachment of exterior wall panels based on visual image recognition, which is executed based on the aforementioned monitoring system for the detachment of exterior wall panels based on visual image recognition, includes the following steps: S1: Acquire a continuous frame image sequence of the exterior wall surface through a camera device, extract gray values, measure texture direction angle, detect light reflection intensity, and use a bilinear interpolation algorithm to spatially calibrate the image sequence to generate a standardized pixel parameter set; S2: Call the standardized pixel parameter set, calculate the gray level change rate parameter, perform local gradient direction operation, analyze the reflection intensity deviation and calculate the pixel potential energy value, and construct the pixel potential energy distribution; S3: Call the pixel potential energy distribution, calculate the potential energy deviation of pixels at the same position in adjacent frames, compare the deviation result with a preset dynamic threshold, mark stress anomaly points when the threshold is exceeded, and generate an anomaly area distribution map. S4: Receive the abnormal area distribution map, calculate the direction consistency parameter, use the least squares method to process the direction relationship between adjacent points, compare the deviation of the current direction consistency parameter with the previous cycle, and when it exceeds the preset trigger threshold, generate the wall panel detachment risk level result. S5: Receive the risk level result of the wall panel falling off, compare the risk level with the preset safety threshold, and extract the coordinates of the risk area when it is greater than the safety threshold. Mark the coordinates of the risk area in the coordinate system of the monitoring interface and trigger the safety monitoring alarm signal for the falling off of the external wall panel.

[0015] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: By extracting grayscale values ​​from consecutive frames and measuring texture direction angles, a spatial calibration relationship between images is constructed, enabling surface textures to maintain pixel-level matching in dynamic scenes. The joint calculation of grayscale change rate and reflection intensity deviation improves the response sensitivity to deformation of fine structures. The establishment of pixel potential energy distribution presents the local stress state in a quantitative form, enabling continuous tracking of strain trends. The joint judgment of potential energy deviation and direction consistency deviation between adjacent frames allows potential cracks or loose areas to be accurately located in the early stages, avoiding misjudgments caused by differences in external lighting or viewing angles. Thus, in multi-time period monitoring, early identification and trend assessment of the risk of exterior wall detachment are achieved, improving the stability of monitoring and the reliability of results. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, 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 the accompanying drawings without creative effort.

[0017] Figure 1 This is a system schematic diagram of the present invention; Figure 2 This is a schematic diagram of the system framework of the present invention; Figure 3 This is a flowchart of the image integration module in this invention; Figure 4 This is a flowchart of the pixel potential energy calculation module in this invention; Figure 5 This is a flowchart of the tension gradient analysis module in this invention; Figure 6 This is a flowchart of the shedding risk identification module in this invention; Figure 7 This is a flowchart of the monitoring and early warning module in this invention; Figure 8 This is a flowchart of the method of the present invention. Detailed Implementation

[0018] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0019] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0020] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0021] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0022] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0023] This invention provides an external wall panel detachment monitoring system based on visual image recognition, such as... Figure 1-2 The diagram shown illustrates a visual image recognition-based external wall panel detachment monitoring system. This system includes: The image integration module acquires a continuous frame image sequence of the exterior wall surface through a camera device, performs grayscale value extraction, measures the texture direction angle, detects the light reflection intensity, uses a bilinear interpolation algorithm to spatially calibrate the image sequence, generates a standardized pixel parameter set, and transmits it to the pixel potential energy calculation module. The pixel potential energy calculation module calls the standardized pixel parameter set, calculates the gray level change rate parameter, performs local gradient direction calculation, analyzes the reflection intensity deviation and calculates the pixel potential energy value, constructs the pixel potential energy distribution, and transmits it to the tension gradient analysis module. The tension gradient analysis module calls the pixel potential energy distribution, calculates the potential energy deviation of pixels at the same position in adjacent frames, compares the deviation result with a preset dynamic threshold, marks stress anomaly points when the threshold is exceeded, generates an anomaly area distribution map, and transmits it to the detachment risk identification module. The detachment risk identification module receives the abnormal area distribution map, calculates the directional consistency parameter, uses the least squares method to process the directional relationship between adjacent points, compares the deviation of the current directional consistency parameter with the previous cycle, and when it exceeds the preset trigger threshold, generates the wall panel detachment risk level result and transmits it to the monitoring and early warning module. The monitoring and early warning module receives the risk level results of wall panel detachment, compares the risk level with the preset safety threshold, and extracts the coordinates of the risk area when it is greater than the safety threshold. The coordinates of the risk area are marked in the coordinate system of the monitoring interface, triggering the external wall panel detachment safety monitoring alarm signal. The standardized pixel parameter set includes grayscale value, texture orientation angle, and light reflection intensity; the pixel potential energy distribution includes grayscale gradient change rate, local gradient direction distribution, and reflected light intensity deviation value; the abnormal area distribution map includes stress anomaly point coordinates, pixel stress deviation amplitude, and abnormal area boundary contour; the wall panel detachment risk level result includes directional consistency coefficient, periodic variation deviation value, and risk level threshold coefficient; and the external wall panel detachment safety monitoring alarm signal includes risk area coordinate information, alarm triggering status, and warning signal identifier.

[0024] Specifically, such as Figure 2 , 3 As shown, the image integration module includes: The image acquisition submodule acquires a continuous frame image sequence of the exterior wall surface through a camera device, detects the pixel array time index of the frame image, judges abnormal frames based on the fluctuation trend of inter-frame displacement difference, calculates the inter-frame pixel displacement difference and adjusts the frame sequence index, and generates the frame sequence index matrix value. The process involves acquiring a sequence of consecutive frames of images of the exterior wall surface using a camera device. Specifically, this is done by utilizing a drone mounted on a device with a resolution of [resolution value missing]. Pixels, capture frame rate per second A frame of industrial camera, along the exterior wall of the building. The camera ascends vertically at a constant speed to capture images, then detects the pixel array time index of the frame image, assigning a timestamp accurate to milliseconds to each captured image. For example, the first frame image in the sequence... Assigned a time index The second frame image The index is Similarly, abnormal frames are then identified based on the fluctuation trend of inter-frame displacement differences, starting with adjacent frames. and Identify and match between Given a fixed set of feature points, calculate the average displacement vector of these feature points in the image coordinate system. For example, from a frame to frame average displacement vector Calculated as Pixels, and from frames to frame displacement vector Due to the influence of sudden crosswinds, the calculation is as follows: Pixels, and then calculate the inter-frame displacement difference, this difference value Defined as a continuous displacement vector and The magnitude of the vector difference between them, i.e. The trend of fluctuation is determined by comparing the calculated difference value with a preset "fluctuation threshold". The comparison is complete; this threshold is based on... Statistical analysis was performed on the inter-frame displacement difference values ​​under stable flight conditions, and the maximum observed value was taken. of The setting is based on multiples, that is Due to the calculated Greater than this threshold Therefore, frame Frames identified as abnormal are those that are determined to be abnormal during the steps of calculating the inter-frame pixel displacement difference and adjusting the frame sequence index. The index will be derived from the original frame sequence index. The abnormal frame index was removed from the middle frame. The specific processing procedure is shown in Table 1 below. This table shows the monitoring of inter-frame displacement and the judgment process of abnormal frames. Finally, a new frame sequence index matrix value without abnormal frame index is generated. .

[0025] Table 1: Examples of Inter-Frame Displacement Monitoring and Abnormal Frame Judgment Frame pairs Displacement vector (pixels) Displacement difference Is it abnormal? F1-F2 (0.2，-98.9) - no F2-F3 (0.1，-99.1) 0.22 no F3-F4 (12.5，-101.3) 12.59 yes F4-F5 (0.3，-99.5) 12.42 no As shown in Table 1, the table lists the displacement vectors of consecutive frame pairs, the difference values ​​calculated based on the displacement vectors, and the judgment results after comparing the difference values ​​with the threshold.

[0026] The grayscale and texture extraction submodule extracts pixel grayscale data based on the frame sequence index matrix value and counts grayscale differences, calculates local pixel gradient directions and determines direction angles, detects brightness change reflection areas and calculates reflection intensity, and generates a grayscale texture direction intensity set. Based on frame sequence index matrix values Extracting pixel grayscale data and calculating grayscale differences is performed for each valid frame image in the sequence. For example, in frame... Select coordinates Centered For each pixel region, extract its grayscale data matrix, with values ​​as follows: first row [83, 90, 95], second row [88, 92, 98], third row [91, 94, 100], where the value range is... to The grayscale difference statistics are performed on the center pixel. It calculates the average of the absolute differences between the gray values ​​of the pixel and its eight neighboring pixels. This value serves as a parameter describing the texture complexity of that point. Next, the local pixel gradient direction is calculated and the angle is determined. Specifically, the grayscale change rate of the center pixel in the horizontal and vertical directions is calculated respectively. and Horizontal rate of change Based on the grayscale value of its right neighbor Subtract the gray value of the left neighbor Obtain, for Vertical rate of change Based on the grayscale values ​​of its neighbors below Subtract the grayscale value of the neighbor above. Obtain, for Angle of gradient direction By calculating the arctangent function The direction and angle are determined based on a preset angle range, where... to interval and to The interval is defined as the "horizontal direction". to The interval is defined as the "vertical direction," and the rest are the "diagonal direction," therefore The direction is determined to be horizontal. Subsequently, the reflective area with brightness changes is detected and the reflection intensity is calculated. A brightness threshold is set for this process. The threshold setting is based on the reference to Histogram analysis was performed on 100 exterior wall images under standard lighting conditions to determine the cumulative distribution function. The corresponding grayscale value is used as the threshold and set to . If the grayscale value of any pixel in the image is higher than 1, then... It is then marked as part of the reflective region, for example, another coordinate in the image. The pixel grayscale value at that location is If the average grayscale value of its surrounding connected pixel blocks is 0, then the pixel is marked. Then the value That is, the reflection intensity of this region, for the center pixel in the example. Its grayscale value is lower than Therefore, its reflection intensity is recorded as Finally, we will target the pixels. The calculated parameters are integrated to generate a grayscale texture direction intensity set {grayscale value:} Gray-scale difference: Gradient direction: Reflection intensity: }

[0027] The spatial calibration submodule calculates the spatial position offset and performs a bilinear interpolation algorithm based on the difference in pixel coordinates between adjacent frames according to the grayscale texture direction intensity set, adjusts the spatial mapping relationship between grayscale and direction data, reconstructs pixel spatial parameters, and generates a standardized pixel parameter set. Select the generated frame sequence that has had abnormal frames removed, such as frame... and frame In the frame coordinates A blemish with specific grayscale and texture features was identified, and its parameter set is {grayscale value:}. Gray-scale difference: ...}, and then in the frame Search for feature points with the most similar parameter set in the coordinates. The match was successful, thus the pixel coordinate difference vector of this feature point is obtained. This process is repeated between two frames. Then, spatial position offsets are calculated and bilinear interpolation is performed using these distinct feature points. The average offset vector is obtained by averaging the components of the coordinate difference vector. Assume the calculated average offset vector is... This vector represents the camera's position in the captured frame. and frame The relative movement between them, in order to move the frames With frames Alignment requires frames When performing geometric transformations and bilinear interpolation, for any target pixel coordinate in the output image, for example... First, based on the calculated spatial offset, its position in the source image frame is calculated in reverse. The corresponding coordinates in, i.e. Because of the coordinates Quantity At pixel and Interpolation is required between frames, assuming frames The grayscale values ​​of the four pixels surrounding this position are respectively , , , First along Linear interpolation is performed in the direction to obtain the point. grayscale value ,point grayscale value Due to the target coordinates The components are integers Therefore, there is no need to proceed. Directional interpolation, final point The grayscale value is This step adjusts the spatial mapping relationship between grayscale and orientation data for the frame. The same transformation is applied to each pixel and its associated grayscale, texture, and orientation data to reconstruct the pixel space parameters, resulting in a transformed frame. At the pixel level, compared to frames Alignment generates a standardized set of pixel parameters.

[0028] Specifically, such as Figure 2 , 4 As shown, the pixel potential energy calculation module includes: The grayscale change rate extraction submodule calls the standardized pixel parameter set to extract grayscale value data, calculates the grayscale difference between adjacent pixels, constructs a grayscale change matrix based on the coordinate difference and grayscale difference, performs discrete difference and operation on the matrix, and generates grayscale change rate parameter values. The generated normalized pixel parameter set is invoked, specifically from the aligned frame. Choose one Analyze the pixel region, the coordinates of the center point of this region are: The grayscale matrix extracted from the standardized parameter set is as follows: first row [52, 50, 48], second row [54, 49, 47], third row [55, 51, 49]. Then, the grayscale difference between adjacent pixels is calculated, with the center pixel as the center pixel. For example, its grayscale value is Its right adjacent pixel grayscale value The horizontal grayscale difference is The adjacent pixels below it grayscale value The vertical grayscale difference is Next, a grayscale variation matrix is ​​constructed based on the coordinate difference and grayscale difference for each pixel. The coordinate difference is in both the horizontal and vertical directions. Using individual pixels and the calculated grayscale difference, a vector representing local grayscale changes is constructed. For the whole Repeating this process for each pixel within the region yields a matrix composed of vectors. Then, a discrete difference operation is performed on this matrix, applied to the grayscale change vector of each pixel. For example, for each pixel... vector First, calculate the square of each component, that is... and Then add the two squared values ​​together to get Finally, take its arithmetic square root. This result is the grayscale change rate of that pixel. This calculation process is applied to every pixel in the image to generate the grayscale change rate parameter value.

[0029] The local gradient direction calculation submodule calls the gray-level change rate parameter value, calculates the horizontal and vertical gray-level difference ratio based on the row and column coordinates of the pixel in two-dimensional space, performs arctangent operation to obtain the direction angle, and performs a weighted operation on the direction angle and gray-level change rate to generate a set of direction distribution coefficients. The generated grayscale change rate parameter value is called, specifically the pixel value calculated in the previous process. Horizontal grayscale difference Vertical grayscale difference and grayscale change rate The ratio of horizontal to vertical grayscale difference is calculated based on the row and column coordinates of pixels in two-dimensional space. The vertical grayscale difference... Divide by horizontal grayscale difference To obtain the ratio Next, the arctangent operation is performed to obtain the direction angle, and the quotient is then calculated. Perform arctangent calculation to obtain the direction angle. To map it to to The complete interval, adjusted angle is Then, a weighted calculation is performed on the orientation angle and the grayscale change rate. This calculation involves two weights. and The setting of these two weights is based on analysis. The features of known wall crack sample images are analyzed, and the contribution of their orientation angle and grayscale change rate to crack identification is statistically determined. (The last part, "Setting...", appears to be an error and doesn't translate directly.) for , for Before the calculation, the orientation angle and grayscale change rate need to be normalized. The orientation angle is normalized by dividing its value by... ,Right now The normalization of grayscale change rate is referenced to the image of this type under similar lighting conditions. The maximum rate of change observed in each sample To proceed, that is The specific process of weighted operation is as follows: normalized direction angle Multiply by its weight get Normalized grayscale change rate Multiply by its weight get The sum of the two is: This value is the orientation distribution coefficient of the pixel. Table 2 below shows the calculation process of the orientation distribution coefficients of several pixels. Finally, a set of orientation distribution coefficients composed of the calculation results of the pixels is generated.

[0030] Table 2: Example Table for Calculating Directional Distribution Coefficient Pixel coordinates Horizontal grayscale difference Vertical grayscale difference Gray scale change rate Direction and angle (degrees) Distribution coefficient (499，800) -2 2 2.828 135 0.1840 (499，801) -2 2 2.828 135 0.1840 (500，800) -4 -2 4.472 206.6 0.2831 (500，801) -2 2 2.828 135 0.1840 As shown in Table 2, this table lists the grayscale difference, rate of change, orientation angle, and orientation distribution coefficient values ​​obtained by weighted calculation for the selected pixels.

[0031] The reflection intensity deviation calculation submodule performs differential comparison on the reflection intensity signal of the pixel unit according to the direction distribution coefficient set, calculates the difference amplitude between the signal and the illumination reference value, performs regional integration based on the difference amplitude, and generates pixel potential energy distribution. First, extract a pixel parameter set from the standardized pixel parameter set. Reflectance intensity data for the region, for example, the center point is located at The region has a reflection intensity signal matrix as follows: first row [241, 245, 248], second row [240, 247, 250], third row [238, 246, 249]. Differential matching is used to calculate the center pixel. The average absolute value of the intensity differences between it and its eight neighboring pixels, i.e. Next, the difference between the signal and the illumination reference value is calculated. It selects a block of the image. The average pixel reflection intensity of a background wall area that is free of obvious flaws and has uniform lighting is calculated and set as [value]. For the center pixel, the difference is The difference ranges in The range between these two values ​​is defined as "low". The range is "medium", higher than For a "high" range, therefore The magnitude is moderate. Then, region integration is performed based on the difference magnitude. This process involves multiplying the difference magnitude of each pixel within the specified region by its corresponding directional distribution coefficient, and then summing the multiplications. Assuming the center point... and its right-hand point The difference magnitudes are respectively and The corresponding directional distribution coefficients are respectively and The integral of these two points is Extend this operation to the pixels of the entire target region and sum the product terms, for example, one The total integral value of the region is This generates the pixel potential energy distribution.

[0032] Specifically, such as Figure 2 , 5 As shown, the tension gradient analysis module includes: The pixel potential energy deviation calculation submodule calls the pixel potential energy distribution, extracts the grayscale value and pixel spatial parameters of the pixels in each frame and converts them into pixel potential energy values, constructs a pixel potential energy dataset, performs inter-frame difference and normalization correction on pixels at the same position, and generates a pixel potential energy deviation value sequence. From the generated pixel potential energy distribution, the total integral value of a certain region is used as the potential energy value of the center pixel of that region. For example, for an aligned frame... and The coordinates are The pixels, which are in the frame The pixel potential energy value calculated in the middle is As time goes by, in frames In the middle, due to possible minute changes inside the wall, the pixel potential energy value at that point becomes Next, a pixel potential energy dataset is constructed. This dataset is a matrix with the same size as the image, where each element represents the potential energy value of the corresponding pixel coordinates. Then, inter-frame differencing is performed on pixels at the same location, followed by normalization correction. Inter-frame differencing is accomplished by calculating the difference in potential energy values ​​of the same pixel across different frames. The normalization correction compares the difference with a preset maximum possible potential energy difference. For comparison, this maximum value was obtained through statistical analysis of the potential energy changes in a wall from intact to showing obvious cracks in a large amount of raw data, and was set as [value missing]. The normalized pixel potential energy deviation value is This calculation is applied to the pixels between two comparison frames, as shown in Table 3 below. This table records the potential energy values ​​of some pixels and their deviation calculation results, and finally generates a sequence of pixel potential energy deviation values ​​composed of the normalized difference results of the pixels.

[0033] Table 3: Example Table for Calculating Pixel Potential Energy Deviation Pixel coordinates frame F'5 potential value frame F'7 potential value Potential energy difference Potential energy deviation value (400，600) 85.2 93.7 8.5 0.0567 (400，601) 86.1 98.3 12.2 0.0813 (401，600) 84.9 95.0 10.1 0.0673 (401，601) 85.5 110.5 25.0 0.1667 As shown in Table 3, the table details the changes in potential energy values ​​of several adjacent pixels in frames acquired at different times, and provides the calculated potential energy difference and the normalized potential energy deviation.

[0034] The dynamic threshold determination submodule calculates the deviation amplitude based on the pixel potential energy deviation value sequence, compares the deviation amplitude with the preset dynamic threshold, filters out pixels with deviation amplitudes exceeding the dynamic threshold, and generates stress anomaly marker data. Based on the generated sequence of pixel potential energy deviation values, the deviation amplitude is calculated. This amplitude is the absolute value of the calculated pixel potential energy deviation value. For example, for a pixel... Its deviation range is Next, the deviation amplitude is compared with a preset dynamic threshold. The calculation process of the dynamic threshold first requires obtaining the root mean square deviation of the full frame pixel potential energy deviation value sequence. This value is obtained by comparing the image... Potential energy deviation value per pixel Sum the squares of the numbers and divide by the total number of pixels. Then take the square root to obtain the root mean square deviation, assuming it is calculated over the entire image. for Determining the dynamic threshold also requires a dynamic adjustment coefficient. The baseline value of this coefficient It is based on the The standard deviation of the pixel potential energy distribution in the original exterior wall image without obvious defects was obtained through statistical analysis and set as . Then, fine-tune the settings based on the overall lighting conditions of the current frame, comparing the average grayscale value of the entire frame with the standard lighting grayscale value. By comparison, a lighting adjustment factor is obtained: ; If the average grayscale of the current frame for ,but: ; The final dynamic adjustment coefficient is: ; Dynamic threshold It is the product of the root mean square deviation and the dynamic adjustment coefficient, i.e. Then, pixels with deviations exceeding a dynamic threshold are filtered out. The deviation of each pixel is compared with this threshold. Its deviation range Not greater than the threshold Therefore, it is not marked, and it is assumed to be another pixel. The deviation amplitude is calculated as follows Then because Greater than The pixel is marked as an anomaly, generating stress anomaly marker data.

[0035] The abnormal region generation submodule performs connectivity clustering calculations on spatially adjacent pixels based on stress anomaly marker data and abnormal pixel coordinate information. It calculates the pixel concentration and centroid parameters of the cluster units, divides the regions based on the continuity of the cluster unit boundaries, and generates an abnormal region distribution map. Based on the generated stress anomaly marker data, this data is a binary image where the value is... The pixels represent outliers, and the values ​​are... The normal points are represented by the abnormal pixels. Then, based on the coordinate information of the abnormal pixels, connectivity clustering is performed on the spatially adjacent pixels. This calculation starts from a stress anomaly marker point, for example, with coordinates of... Starting with a pixel, examine its eight neighboring pixels. If a neighboring pixel is also marked as an anomaly, it is grouped into the same cluster. This neighborhood check process is repeated for newly added pixels until all connected anomalies are grouped into a single set. For example, this ultimately forms a cluster containing... Clustering of abnormal pixels Next, the pixel concentration and centroid parameters of the cluster units are calculated. To calculate the pixel concentration, the cluster is first determined. The smallest bounding rectangle, assuming it covers the area from... arrive ,from arrive The area of ​​the rectangle is . The pixel concentration is calculated by dividing the number of outlier pixels in the cluster by the area of ​​the rectangle. The concentration is higher than Defined as "high" concentration, lower than The concentration is "low", therefore For high concentration, the location centroid parameter is calculated by clustering. 1 pixel coordinates and The average value of the coordinates is obtained; assuming the calculation result is... Then, regions are divided based on the continuity of cluster unit boundaries, and the clusters are formed by tracing the constituent clusters. The outermost pixels form a boundary point sequence. Calculate the average distance between adjacent points in the sequence. If the average distance is less than... If diagonal connections are allowed, the boundary is considered continuous, and the cluster forms an independent anomalous region, ultimately generating an anomalous region distribution map.

[0036] Specifically, such as Figure 2 , 6 As shown, the detachment risk identification module includes: The orientation data extraction submodule obtains the coordinates and orientation vectors of pixels in the distribution map of abnormal areas, analyzes the orientation vector groups of adjacent points, calculates the orientation angle and performs smooth adjustment, and calculates the difference relationship based on the adjusted orientation data to obtain the orientation difference sequence value. From the generated clusters of Among the anomalous pixels, extract a feature along its outer boundary contour. An ordered sequence of pixels It retrieves the orientation angle of each pixel from the standardized pixel parameter set and converts it into a two-dimensional orientation vector. For example, the first three points in the sequence... The directions and angles are respectively Their corresponding direction vectors are as follows: ; ; ; Next, the direction vector groups of adjacent points are analyzed, the direction angles are calculated, and smooth adjustments are performed. First, the angles between adjacent vectors are calculated, such as... and The included angle between them is obtained by calculating the inverse cosine of their dot product, i.e. Smoothing adjustment applies a window of size [size missing] to the original direction and angle sequence. Moving average calculation, points The adjusted angle is to Add the angle values ​​and divide by ,Right now Applying this process to the entire sequence yields a new, smoothed angle sequence; for example, the angles of the first four points become... Then, based on the adjusted direction data, the difference relationship is calculated. This step involves calculating the difference between adjacent angles in the smoothed angle sequence. The first difference is... The second difference is The third difference is By iterating through the entire smooth angle sequence, we obtain the direction difference sequence values.

[0037] The orientation consistency calculation submodule calls the orientation difference sequence value, performs least squares fitting on the orientation change rate data, calculates the orientation offset sequence based on the residual distribution, and obtains the orientation consistency parameter value after normalization by the stability function. Call the calculated continuous angle difference, for example, a... The sequence of points is The unit is degrees. Next, least squares fitting is performed on the directional change rate data. This process is for... Data points (in The index of the point , Find the best-fitting straight line for the corresponding direction difference value. By calculating data points The sum of them is , The sum of them is , The sum of them is , The sum of them is Substitute into the calculation slope and intercept The calculation yields the slope. and intercept The fitted straight line is Then, the direction offset sequence is calculated based on the residual distribution, where the residual is the actual value at each point. The predicted value of the fitted line at that point The difference between them, for example, for the first Each point has an actual value of [number]. The predicted value is The residual is The residuals at these points are calculated in this way to obtain the direction offset sequence, as shown in Table 4. After normalization using the stability function, the absolute value of each residual is negative and used as the natural exponent. powers, for example, for residuals ,calculate This step maps large offsets to near... The value, a small offset maps to a value close to The value, finally this The normalized values ​​are summed and averaged. Let the calculated average be... This yields the value of the direction consistency parameter.

[0038] Table 4: Example of Directional Consistency Calculation Process Serial Number Actual difference Fitted Predictions Directional offset value (residual) 1 3.3 2.41 0.89 2 1.7 2.35 -0.65 3 1.5 2.29 -0.79 4 2.0 2.22 -0.22 5 1.8 2.16 -0.36 As shown in Table 4, this table illustrates the process of linearly fitting the directional difference sequence values ​​and calculating the offset of each data point relative to the fitted line.

[0039] The risk level determination submodule calculates the deviation rate based on the direction consistency parameter value and the parameter value of the previous cycle, compares the deviation rate with the set detachment risk trigger threshold, and determines the high-risk range when the deviation rate is higher than the detachment risk trigger threshold, and outputs the wall panel detachment risk level result. Based on the direction consistency parameter value The deviation rate is calculated based on the parameter values ​​from the previous period. Assuming the directional consistency parameter value for the same anomaly area acquired in the previous monitoring period (e.g., one week ago) is... The deviation rate is calculated by subtracting the previous period's value from the current value and then dividing by the previous period's value. Next, the deviation rate is compared with the set detachment risk trigger threshold. The detachment risk trigger threshold is set based on the... Statistical analysis of monitoring data from the entire process of a real wall panel, from the appearance of micro-cracks to eventual detachment, revealed that during the stage where the risk of detachment significantly increased, the periodic deviation rate of the directional consistency parameter was distributed as follows: arrive Within the given range, the lower limit of the distribution range, i.e., the end with the smaller absolute value of the rate of change, is used to set the detachment risk trigger threshold. In the numerical comparison, the calculated deviation rate will be... With threshold The comparison is performed, and when the deviation rate is lower than a certain threshold (i.e., the absolute value of the negative value is greater than the absolute value of the trigger threshold), it is judged as high risk. Less than Therefore, this abnormal area enters a high-risk zone; if the deviation rate is within... arrive If the deviation rate is between these values, it is classified as "medium risk"; if the deviation rate is greater than [a certain value], it is classified as "medium risk". (i.e., the absolute value of the negative change is less than) If the risk level is low, the wall panel detachment risk level is determined to be "high risk" based on the final calculation results.

[0040] Specifically, such as Figure 2 , 7 As shown, the monitoring and early warning module includes: The risk level receiving submodule acquires the risk level results of wall panel detachment, detects the correspondence between risk level parameters and time labels, calculates the fluctuation range of risk level at monitoring points, analyzes the trend of change, and generates risk level statistics. Get the abnormal area The output determines the "high risk" level, while simultaneously checking the correlation between the risk level parameters and time tags. This process quantifies the risk level and sets "low risk" as... "Medium risk" is "High risk" is And retrieve the monitoring point (centroid coordinates) Data from the past four monitoring periods (once a week) are used to form a time series. For example, the risk levels for the first to fourth weeks are "low", "low", "medium", and "high" respectively, which are then quantified into a sequence. Next, the fluctuation range of the risk level at the monitoring point is calculated by calculating the absolute value of the difference between the risk levels of adjacent periods. The fluctuation in the second week is... The third week is The fourth week is The analysis then examines the changing trends, which is achieved by calculating a risk level sequence. With time period A linear fit with varying slopes is achieved by using the period as the independent variable. The grade value is used as the dependent variable. The slope value was calculated to be... This value is greater than Determined as a "significant growth" trend, to The values ​​in between are classified as "slow growth," while the rest are classified as "stable or declining." Finally, the current risk level value, the average volatility of the last three fluctuations, and the trend slope are integrated to generate a comprehensive risk level statistical value. This statistical value is calculated using a weighted summation, with weights based on the original data. The analysis settings for each risk evolution case, and the weight of the current level value. Set as Weight of average fluctuation amplitude Set as The weight of the trend slope Set as The current level value is The average fluctuation range of the last three times was The trend slope is Substituting the values ​​into the calculation yields the statistical value of the risk level. This generates statistical values ​​for risk levels.

[0041] The threshold comparison and judgment submodule calls the preset safety threshold based on the risk level statistics, compares the difference between the risk level statistics of the monitoring point and the safety threshold, calculates the threshold exceedance ratio and the number of triggers, and obtains the threshold exceedance risk judgment result. Based on risk level statistics Invoke the preset security threshold. The preset security threshold is set based on... For example, a retrospective analysis of the original monitoring data of a confirmed wall panel detachment event revealed that, in the last three monitoring periods before the actual detachment, the statistical values ​​of its risk level were stably distributed as follows: to Within the interval, select the lower limit of that interval. As a preset safety threshold, the difference between the statistical value of the risk level at the monitoring point and the safety threshold is then compared, and the current statistical value is calculated. With safety threshold The difference is obtained. Then, the over-threshold ratio and the number of triggers are calculated. The over-threshold ratio is the difference value divided by the safety threshold, i.e. , or expressed as This ratio is in to The range is defined as "slightly exceeding the threshold". to It is classified as "moderate overthreshold", higher than the threshold. The condition is "severely exceeding the threshold," and the number of triggers is counted in the most recent consecutive events. Within a monitoring period, the risk level statistical value exceeds the safety threshold. The total number of times, assuming in the previous Throughout the cycle, the risk level statistics for this monitoring point did not exceed [a certain value]. If the current period exceeds this limit for the first time, the trigger count is recorded as follows. By integrating the out-of-threshold situation, out-of-threshold ratio, and number of triggers, the out-of-threshold risk assessment result is obtained.

[0042] The risk area marking alarm submodule calls the risk exceeding the threshold judgment result, extracts the coordinate index value of the exceeding threshold area, constructs the risk area coordinate matrix and maps it to the monitoring interface, marks the high-risk area, and generates the external wall panel falling off safety monitoring alarm signal. Retrieve the threshold risk assessment result, which is a result containing "Threshold exceeded: Yes" and "Threshold exceeded ratio: "Trigger count: " The dataset is used to extract the coordinate index values ​​of the out-of-threshold regions, which are the abnormal regions. The bounding box coordinate range, its The coordinate range is to , The coordinate range is to and the centroid coordinates calculated in the previous steps Next, a risk area coordinate matrix is ​​constructed and mapped to the monitoring interface. This matrix constitutes the abnormal areas. All The coordinate set of pixels. During mapping, at the corresponding positions on the panoramic view of the building facade taken and stitched by the drone, the data of this coordinate matrix is used to render the covered area as a semi-transparent red highlight layer. Subsequently, high-risk areas are marked. At the center position of the red highlight layer, that is, the center-of-gravity coordinates automatically generate a visual label containing the words "High Risk". And when the monitoring personnel hover the mouse cursor over this label, the system will pop up a detailed information box, as shown in Table 5 below. The key quantitative parameters of this risk area are listed in this form. Finally, an alarm signal for the safety monitoring of the external wall panel detachment is generated. This signal is set to two parallel actions. One is on the main software interface of the monitoring center, triggering a visual flashing effect with a frequency of and lasting seconds, accompanied by a beeping sound. The other is to send a text alarm message to the mobile phone number of the preset property safety supervisor through the communication interface integrated in the system. The content of this message is: "

External Wall Safety Alarm

[0043] Table 5: Alarm Information Table for Risk Areas Alarm parameters Parameter value Area Code C1 center coordinates (512，724) Risk level high Risk statistics 2.074 Safety threshold 2.0 Overthreshold ratio 3.7% Trigger time 2025-10-28-06:58 [[ID=I12]] As shown in Table 5, this table pops up on the monitoring interface when the alarm is triggered, clearly showing the key quantitative indicators of the high-risk area to the operator.

[0044] Please refer to Figure 8 , the method for monitoring the detachment of external wall panels based on visual image recognition is executed based on the above-mentioned monitoring system for the detachment of external wall panels based on visual image recognition, including the following steps: S1: Obtain a sequence of consecutive frame images of the external wall surface through a camera device, perform grayscale value extraction, measure the texture direction angle, detect the light reflection intensity, and perform spatial calibration on the image sequence using the bilinear interpolation algorithm to generate a standardized pixel parameter set; S2: Call the standardized pixel parameter set, calculate the grayscale change rate parameter, perform local gradient direction operations, analyze the reflection intensity deviation and calculate the pixel potential energy value, and construct a pixel potential energy distribution; S3: Call the pixel potential energy distribution, calculate the pixel potential energy deviation of the same-position pixels in adjacent frames, compare the deviation result with a preset dynamic threshold, and mark stress abnormal points when exceeding the threshold to generate a distribution map of abnormal areas; S4: Receive the abnormal area distribution map, calculate the direction consistency parameter, use the least squares method to process the direction relationship between adjacent points, compare the deviation of the current direction consistency parameter with the previous cycle, and generate the wall panel detachment risk level result when the deviation exceeds the preset trigger threshold. S5: Receive the risk level result of wall panel detachment, compare the risk level with the preset safety threshold, and extract the risk area coordinates when the risk level is greater than the safety threshold. Mark the risk area coordinates in the monitoring interface coordinate system and trigger the external wall panel detachment safety monitoring alarm signal.

[0045] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A monitoring system for exterior wall panel detachment based on visual image recognition, characterized in that, The system includes: The image integration module acquires a continuous frame image sequence of the exterior wall surface through a camera device, performs grayscale value extraction, measures the texture direction angle, detects the light reflection intensity, uses a bilinear interpolation algorithm to spatially calibrate the image sequence, generates a standardized pixel parameter set, and transmits it to the pixel potential energy calculation module. The pixel potential energy calculation module calls the standardized pixel parameter set, calculates the gray level change rate parameter, performs local gradient direction calculation, analyzes the reflection intensity deviation and calculates the pixel potential energy value, constructs the pixel potential energy distribution, and transmits it to the tension gradient analysis module. The tension gradient analysis module calls the pixel potential energy distribution, calculates the potential energy deviation of pixels at the same position in adjacent frames, compares the deviation result with a preset dynamic threshold, marks stress anomaly points when the threshold is exceeded, generates an abnormal area distribution map, and transmits it to the detachment risk identification module. The detachment risk identification module receives the abnormal area distribution map, calculates the directional consistency parameter, uses the least squares method to process the directional relationship between adjacent points, compares the deviation of the current directional consistency parameter with the previous cycle, and when it exceeds the preset trigger threshold, generates the wall panel detachment risk level result and transmits it to the monitoring and early warning module.

2. The external wall panel detachment monitoring system based on visual image recognition according to claim 1, characterized in that, The standardized pixel parameter set includes grayscale value, texture orientation angle, and light reflection intensity; the pixel potential energy distribution includes grayscale gradient change rate, local gradient direction distribution, and reflected light intensity deviation value; the abnormal area distribution map includes stress anomaly point coordinates, pixel stress deviation amplitude, and abnormal area boundary contour; and the wall panel detachment risk level result includes directional consistency coefficient, periodic variation deviation value, and risk level threshold coefficient.

3. The external wall panel detachment monitoring system based on visual image recognition according to claim 1, characterized in that, The image integration module includes: The image acquisition submodule acquires a continuous frame image sequence of the exterior wall surface through a camera device, detects the pixel array time index of the frame image, judges abnormal frames based on the fluctuation trend of inter-frame displacement difference, calculates the inter-frame pixel displacement difference and adjusts the frame sequence index, and generates the frame sequence index matrix value. The grayscale and texture extraction submodule extracts pixel grayscale data and counts grayscale differences based on the frame sequence index matrix value, calculates local pixel gradient direction and determines direction angle, detects brightness change reflection areas and calculates reflection intensity, and generates grayscale texture direction intensity set. The spatial calibration submodule compares the pixel coordinate differences between adjacent frames based on the grayscale texture direction intensity set, calculates the spatial position offset, executes a bilinear interpolation algorithm, adjusts the spatial mapping relationship between grayscale and direction data, reconstructs pixel spatial parameters, and generates a standardized pixel parameter set.

4. The external wall panel detachment monitoring system based on visual image recognition according to claim 1, characterized in that, The pixel potential energy calculation module includes: The grayscale change rate extraction submodule calls the standardized pixel parameter set to extract grayscale value data, calculates the grayscale difference between adjacent pixels, constructs a grayscale change matrix based on the coordinate difference and grayscale difference, performs discrete difference and operation on the matrix, and generates grayscale change rate parameter values. The local gradient direction calculation submodule calls the gray-level change rate parameter value, calculates the horizontal and vertical gray-level difference ratio based on the row and column coordinates of the pixel in two-dimensional space, performs arctangent operation to obtain the direction angle, and performs a weighted operation on the direction angle and gray-level change rate to generate a set of direction distribution coefficients. The reflection intensity deviation calculation submodule performs differential comparison on the reflection intensity signal of the pixel unit according to the set of directional distribution coefficients, calculates the difference amplitude between the signal and the illumination reference value, performs regional integration based on the difference amplitude, and generates the pixel potential energy distribution.

5. The external wall panel detachment monitoring system based on visual image recognition according to claim 1, characterized in that, The tension gradient analysis module includes: The pixel potential energy deviation calculation submodule calls the pixel potential energy distribution, extracts the gray value and pixel spatial parameters of the pixels in each frame and converts them into pixel potential energy values, constructs a pixel potential energy dataset, performs inter-frame difference and normalization correction on pixels at the same position, and generates a pixel potential energy deviation value sequence. The dynamic threshold determination submodule calculates the deviation amplitude based on the pixel potential energy deviation value sequence, compares the deviation amplitude with a preset dynamic threshold, filters out pixels whose deviation amplitude exceeds the dynamic threshold, and generates stress anomaly marker data. The abnormal region generation submodule performs connectivity clustering calculations on spatially adjacent pixels based on the stress anomaly marker data and the anomaly pixel coordinate information. It calculates the pixel concentration and centroid parameters of the cluster units, divides the regions based on the continuity of the cluster unit boundaries, and generates an abnormal region distribution map.

6. The external wall panel detachment monitoring system based on visual image recognition according to claim 5, characterized in that, The step of performing inter-frame difference and normalization correction on pixels at the same position refers to the pixel potential energy difference value of the same pixel between adjacent frames. The Min-Max normalization method is used to perform numerical mapping processing on all pixel potential energy difference values. The dynamic threshold is a dynamic adjustment coefficient determined based on the root mean square deviation of the full-frame pixel potential energy deviation value sequence and the statistical distribution results of the original frame.

7. The external wall panel detachment monitoring system based on visual image recognition according to claim 1, characterized in that, The shedding risk identification module includes: The orientation data extraction submodule obtains the coordinates and orientation vectors of pixels in the abnormal area distribution map, analyzes the orientation vector group of adjacent points, calculates the orientation angle and performs smooth adjustment, and calculates the difference relationship based on the adjusted orientation data to obtain the orientation difference sequence value. The orientation consistency calculation submodule calls the orientation difference sequence value, performs least squares fitting on the orientation change rate data, calculates the orientation offset sequence based on the residual distribution, and obtains the orientation consistency parameter value after normalization by the stability function. The risk level determination submodule calculates the deviation rate based on the direction consistency parameter value and the parameter value of the previous cycle, compares the deviation rate with the set detachment risk trigger threshold, and determines the high-risk range when the deviation rate is higher than the detachment risk trigger threshold, and outputs the wall panel detachment risk level result.

8. The external wall panel detachment monitoring system based on visual image recognition according to claim 1, characterized in that, The monitoring and early warning module receives the risk level result of the wall panel falling off, compares the risk level with the preset safety threshold, and extracts the risk area coordinates when the risk level is greater than the safety threshold. The risk area coordinates are marked in the monitoring interface coordinate system, and the external wall panel falling off safety monitoring alarm signal is triggered. The safety monitoring alarm signal for the detachment of exterior wall panels includes risk area coordinates, alarm trigger status, and warning signal identifier.

9. The external wall panel detachment monitoring system based on visual image recognition according to claim 8, characterized in that, The monitoring and early warning module includes: The risk level receiving submodule acquires the risk level results of the wall panel detachment, detects the correspondence between the risk level parameters and the time label, calculates the fluctuation range of the risk level at the monitoring point, analyzes the trend of change, and generates statistical values ​​of the risk level. The threshold comparison and determination submodule calls a preset safety threshold based on the risk level statistics, compares the difference between the risk level statistics of the monitoring point and the safety threshold, calculates the threshold exceedance ratio and the number of triggers, and obtains the threshold exceedance risk determination result. The risk area marking alarm submodule calls the above-threshold risk judgment result, extracts the coordinate index value of the above-threshold area, constructs the risk area coordinate matrix and maps it to the monitoring interface, marks the high-risk area, and generates an external wall panel detachment safety monitoring alarm signal.

10. A method for monitoring the detachment of exterior wall panels based on visual image recognition, characterized in that, The external wall panel detachment monitoring system based on visual image recognition according to any one of claims 1-9 includes the following steps: S1: Acquire a continuous frame image sequence of the exterior wall surface through a camera device, extract grayscale values, measure texture direction angle, detect light reflection intensity, and use a bilinear interpolation algorithm to spatially calibrate the image sequence to generate a standardized pixel parameter set; S2: Call the standardized pixel parameter set, calculate the gray level change rate parameter, perform local gradient direction operation, analyze the reflection intensity deviation and calculate the pixel potential energy value, and construct the pixel potential energy distribution; S3: Call the pixel potential energy distribution, calculate the potential energy deviation of pixels at the same position in adjacent frames, compare the deviation result with a preset dynamic threshold, mark stress anomaly points when the threshold is exceeded, and generate an anomaly area distribution map. S4: Receive the abnormal area distribution map, calculate the direction consistency parameter, use the least squares method to process the direction relationship between adjacent points, compare the deviation of the current direction consistency parameter with the previous cycle, and when it exceeds the preset trigger threshold, generate the wall panel detachment risk level result. S5: Receive the risk level result of the wall panel falling off, compare the risk level with the preset safety threshold, and extract the coordinates of the risk area when it is greater than the safety threshold. Mark the coordinates of the risk area in the coordinate system of the monitoring interface and trigger the safety monitoring alarm signal for the falling off of the external wall panel.

Images