An intelligent detection system and method for the bonding quality of the thermal insulation layer of a house building outer wall

Abstract

The application discloses a kind of housing construction outer wall insulation layer bonding quality intelligent detection system and method, belong to bonding quality intelligent detection technical field, comprising: image acquisition and pretreatment module, dynamic threshold image segmentation and defect determination module, sequence image panorama splicing and global integration module and defect grading quantitative evaluation module.The application obtains single frame and inspection sequence image by synchronously collecting visible light, infrared thermal imaging image and completing pretreatment.According to whether the thickness of insulation layer selects dynamic update segmentation threshold, whether local threshold is corrected is judged in combination with acoustic resonance frequency offset, and then defect detection is carried out.Sequence image overlapping area is fused and spliced to generate panoramic image, and the defect detection result is mapped to realize global detection.According to defect distribution density, multi-level grading of defect is completed, and finally the bonding quality detection of outer wall insulation is realized.

Description

Technical Field

[0001] This invention relates to the field of intelligent bonding quality detection technology, and in particular to an intelligent testing system and method for bonding quality of building exterior wall insulation layers. Background Technology

[0002] With the widespread promotion of building energy conservation projects, external wall insulation systems have become an indispensable part of building construction. The bonding quality between the insulation layer and the wall base directly determines the overall safety, durability, and service life of the external wall insulation system. If defects such as poor bonding, local hollowing, or large-area delamination occur, it will not only reduce the thermal insulation effect, but also easily lead to safety accidents such as insulation layer falling off and falling objects injuring people. Therefore, conducting comprehensive and accurate testing of the bonding quality of the external wall insulation layer is a key link in building completion acceptance and subsequent operation and maintenance.

[0003] Currently, image processing for detecting the bonding quality of exterior wall insulation layers in building construction still suffers from numerous key technical deficiencies, making it difficult to meet the high-precision, full-coverage, and intelligent detection requirements of engineering projects. In the image preprocessing stage, visible light images are easily affected by light fluctuations, building shadows, wall stains, dust, and debris obstruction, while infrared thermal imaging images are highly susceptible to interference from external factors such as ambient temperature, humidity, and wind, resulting in numerous noises and uneven grayscale distribution in the original images, severely masking defects such as bonding voids and weak adhesion. At the feature extraction level, existing image processing algorithms mostly use general texture, grayscale, and contour extraction logic, without specific optimization for the unique imaging characteristics of exterior wall insulation bonding failure. This makes it difficult to accurately capture the grayscale abrupt changes, texture differences, and thermal gradient changes between void areas and densely bonded areas. It also fails to effectively distinguish the imaging differences between gaps, surface cracks, material color differences in normal wall finish layers and internal bonding defects, frequently resulting in missed or incorrect feature extraction. Furthermore, its ability to identify hidden point-like debonding and shallow micro-voids is extremely poor. In the image segmentation and defect identification stages, existing technologies mostly rely on fixed thresholds to complete image segmentation and defect identification. They fail to incorporate dynamic threshold correction models based on key parameters such as insulation layer thickness and shooting distance. This results in low segmentation accuracy for images from different perspectives and scales, frequent false positives and false negatives, and an inability to accurately quantify the area and distribution of defects. Furthermore, fragmented single-frame detection results cannot generate comprehensive detection data for the entire wall surface, making it difficult to conduct overall quality assessments of building exterior walls. In summary, existing image processing technologies have weak anti-interference capabilities and insufficient overall intelligence and refinement, making them unsuitable for the routine high-precision inspection needs of large-area exterior wall insulation quality in high-rise buildings. Summary of the Invention

[0004] In view of this, embodiments of the present invention provide an intelligent detection system and method for the bonding quality of building exterior wall insulation layers. This system can simultaneously acquire visible light and infrared thermal imaging images and perform preprocessing to obtain single-frame and inspection sequence images. Based on the insulation layer thickness, it selects whether to dynamically update the segmentation threshold, and combines the acoustic resonance frequency offset to determine whether to correct local thresholds before conducting defect detection. Overlapping areas of the sequence images are fused and stitched to generate a panoramic image, and the defect detection results are mapped to achieve full-domain detection. Based on the defect distribution density, multi-level defect classification is performed, ultimately achieving the detection of the bonding quality of exterior wall insulation.

[0005] This invention provides an intelligent inspection system for the bonding quality of exterior wall insulation layers in building construction. The system includes: an image acquisition and preprocessing module, a dynamic threshold image segmentation and defect determination module, a sequence image panoramic stitching and full-domain integration module, and a defect grading and quantitative evaluation module. The image acquisition and preprocessing module simultaneously acquires and preprocesses visible light and infrared thermal imaging images of the exterior wall insulation area to be inspected, obtaining single-frame images and continuous inspection sequence images containing the insulation layer. The dynamic threshold image segmentation and defect determination module determines whether to perform intelligent adjustment of the dynamically updated image segmentation threshold based on the insulation layer thickness. If yes, adaptive and precise segmentation is performed after intelligent adjustment; otherwise, the process proceeds directly. The system employs an adaptive precision segmentation method. During this process, it acquires the acoustic resonance frequency offset and determines whether to perform local threshold correction driven by the audio frequency offset. If so, defect detection is performed after correction; otherwise, defect detection is performed directly. The sequence image panoramic stitching and global integration module receives continuous inspection sequence images and fuses them with overlapping areas. It stitches multiple single-frame images into a complete panoramic image of the building's exterior facade and maps the single-frame defect detection results to the panoramic image, achieving global detection of bonding defects in the building's exterior wall insulation layer. The defect grading and quantification evaluation module is used to classify the detected bonding defects into multiple levels of severity based on the image defect distribution density, thereby completing the quality inspection of the exterior wall insulation bonding.

[0006] This invention also provides an intelligent detection method for the bonding quality of building exterior wall insulation layers. The method is applied to an intelligent detection system for the bonding quality of building exterior wall insulation layers. The method includes: simultaneously acquiring visible light and infrared thermal imaging images of the building exterior wall insulation area to be inspected and preprocessing them to obtain single-frame images containing the insulation layer and continuous inspection sequence images; determining whether to perform intelligent adjustment of dynamically updated image segmentation thresholds based on the insulation layer thickness; if yes, performing adaptive precise segmentation after intelligent adjustment; otherwise, directly performing adaptive precise segmentation; during the adaptive precise segmentation process, acquiring the acoustic resonance frequency offset and determining whether to perform audio frequency offset-driven local threshold correction; if yes, performing defect detection after correction; otherwise, directly performing defect detection; receiving continuous inspection sequence images and fusing them with overlapping areas to stitch multiple single-frame images into a complete panoramic image of the building exterior wall facade; mapping the single-frame defect detection results to the panoramic image to achieve full-area detection of bonding defects in the building exterior wall insulation layer; and classifying the detected bonding defects into multiple severity levels based on the image defect distribution density to complete the external wall insulation bonding quality detection.

[0007] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0008] 1. This system employs simultaneous acquisition and preprocessing of images using visible light and infrared thermal imaging. It dynamically adjusts the segmentation threshold and local grayscale threshold based on both insulation layer thickness and acoustic resonance frequency offset, avoiding the limitations of traditional fixed-threshold detection methods. It can adapt to the imaging characteristics of different insulation structures and various bonding defect morphologies, effectively avoiding false detections and missed detections caused by interference from lighting, materials, and the environment. This significantly improves image segmentation accuracy and defect recognition accuracy, especially in accurately identifying hidden bonding defects such as shallow micro-voids and point-like weak adhesion.

[0009] 2. This system relies on feature point matching and pixel coordinate offset correction to complete the registration of sequential images and seamless fusion of overlapping areas, stitching together scattered single-frame images into a complete panoramic image of the exterior wall and simultaneously mapping defect detection results. This solves the shortcomings of traditional segmented detection data fragmentation and the inability to perform comprehensive analysis, achieving full-coverage localization and unified statistics of bonding defects in the insulation layer of the entire exterior wall. Simultaneously, it ensures the integrity of image details and the absence of misalignment or ghosting after stitching, providing reliable image data support for comprehensive quality analysis.

[0010] 3. This system identifies false defects and distinguishes defect types by the area of ​​the defect's connected components. It then combines this with defect distribution density to achieve multi-level severity grading, realizing a quantitative process for the entire defect process from identification and classification to risk rating. It not only accurately removes invalid false defects caused by image noise but also objectively reflects the degree of defect aggregation and safety hazard level in local and overall areas. The detection results are quantitative and intuitive, facilitating quality assessment, hazard rectification, and subsequent operation and maintenance management for engineers. Attached Figure Description

[0011] Figure 1 This application provides a schematic diagram of the structure of an intelligent detection system for the bonding quality of exterior wall insulation layers in buildings.

[0012] Figure 2 A flowchart illustrating the intelligent control of dynamically updated image segmentation thresholds for an intelligent detection system for the bonding quality of building exterior wall insulation layers, provided in this application embodiment;

[0013] Figure 3 A flowchart illustrating an intelligent detection method for the bonding quality of exterior wall insulation layers in building construction, provided in this application embodiment. Detailed Implementation

[0014] like Figure 1 The diagram shown is a structural schematic of an intelligent detection system for the bonding quality of exterior wall insulation layer in a building provided in an embodiment of this application. The system includes: an image acquisition and preprocessing module, a dynamic threshold image segmentation and defect determination module, a sequence image panoramic stitching and global integration module, and a defect grading and quantitative evaluation module.

[0015] As the first module of this system, the image acquisition and preprocessing module is used to simultaneously acquire visible light images and infrared thermal imaging images of the exterior wall insulation area of ​​the building to be inspected and preprocess them to obtain single-frame images and continuous inspection sequence images containing the exterior wall insulation layer.

[0016] In this embodiment, the module is equipped with a visible light imaging acquisition unit and an infrared thermal imaging acquisition unit, which can perform synchronous and parallel image acquisition operations on the exterior wall insulation facade area of ​​high-rise buildings and multi-story buildings. It can acquire static single-frame exterior wall insulation images to achieve fine imaging detection of fixed-point areas, and can also be equipped with a drone inspection platform to continuously cruise and shoot along the exterior wall facade to obtain continuous inspection sequence images covering the entire exterior wall. To address the issues of noise, grayscale imbalance, blurred details, and thermal field diffusion in original images caused by complex outdoor construction site conditions such as uneven lighting, dust interference, shadow occlusion, and fluctuating ambient temperature and humidity, the module performs specialized image preprocessing operations after acquiring the original visible light and infrared thermal imaging images. These operations include image noise reduction, distortion correction, grayscale equalization, detail enhancement, and invalid background removal. This effectively filters out image interference information from the outdoor environment, retains and highlights the surface texture, boundary contours, and latent imaging features corresponding to internal bonding defects of the external wall insulation layer. The final output is a clear, feature-complete, and low-interference single-frame image and a continuous inspection sequence of images of the external wall insulation layer. This provides high-quality and highly effective original image data support for subsequent back-end detection processes such as dynamic threshold image segmentation, defect feature extraction, panoramic image stitching, and defect classification assessment, thereby reducing the adverse effects of complex outdoor conditions on the intelligent detection accuracy of insulation layer bonding defects from the source.

[0017] As the second module of this system, the dynamic threshold image segmentation and defect determination module is used to determine whether to perform intelligent adjustment of the dynamically updated image segmentation threshold based on the thickness of the insulation layer. If yes, adaptive precise segmentation is performed after intelligent adjustment; otherwise, adaptive precise segmentation is performed directly. During the adaptive precise segmentation process, the acoustic resonance frequency offset is obtained and it is determined whether to perform local threshold correction driven by the audio frequency offset. If yes, defect detection is performed after correction; otherwise, defect detection is performed directly.

[0018] It should be understood that Figure 2 The specific process of the intelligent control flowchart for dynamically updated image segmentation threshold of the intelligent detection system for bonding quality of building exterior wall insulation layer provided in this application embodiment is as follows: First, the measured value of insulation layer thickness is obtained, and then compared with the lower critical boundary and the critical interval in sequence. If the thickness is less than or equal to the lower critical boundary, it is determined to be a low-thickness interval. The thickness deviation is calculated and the adjustment coefficient is obtained through the mapping relationship. The initial threshold benchmark value is adjusted downward. If the thickness is within the critical interval, it is determined to be a medium-thickness interval. The initial threshold benchmark value is kept unchanged. If the thickness is greater than or equal to the upper critical boundary, it is determined to be a high-thickness interval. The thickness deviation is calculated and the adjustment coefficient is obtained through the mapping relationship. The initial threshold benchmark value is adjusted upward. Finally, the target image segmentation threshold adapted to the current insulation layer thickness is output.

[0019] It needs to be explained that the specific process for determining whether to perform intelligent adjustment of dynamically updated image segmentation thresholds is as follows:

[0020] Multiple insulation layer thickness ranges are pre-divided and corresponding to preset initial segmentation threshold benchmarks. The measured insulation layer thickness data of the current detection area is obtained in real time. The thickness range to which the measured data belongs is determined, and the initial segmentation threshold benchmark of the corresponding range is retrieved. The multiple insulation layer thickness ranges include low-grade insulation layer thickness range, medium-grade insulation layer thickness range and high-grade insulation layer thickness range.

[0021] The insulation layer thickness is the overall vertical thickness of the insulation structure layer at the location of the exterior wall to be tested, from the surface of the base wall to the surface of the exterior finish layer. The initial segmentation threshold benchmark value is the standard pixel feature critical value that distinguishes between the densely bonded area and the bond defect area in the image under the corresponding insulation layer thickness range.

[0022] The thickness of the insulation layer is compared with the pre-divided multiple insulation layer thickness ranges. Based on the comparison results, the image segmentation threshold is dynamically adjusted. The image segmentation threshold is a pixel feature threshold used to determine whether an image pixel belongs to a dense bonding area or a bonding defect area. The pixel features include at least one of grayscale value and infrared thermal imaging temperature value.

[0023] A thickness deviation-segmentation threshold adjustment coefficient mapping relationship is pre-constructed. This mapping relationship is used to characterize the one-to-one mapping relationship between the thickness of the insulation layer and the median thickness of the corresponding thickness range and the segmentation threshold adjustment coefficient.

[0024] The specific process for dynamically adjusting the image segmentation threshold is as follows:

[0025] If the insulation layer thickness is less than or equal to the critical lower limit of the insulation layer thickness, the thickness range to which the measured data belongs is determined to be the low-grade insulation layer thickness range. The difference between the current insulation layer thickness and the median thickness of the low-grade insulation layer thickness range is calculated to obtain the thickness deviation. The thickness deviation is input into the thickness deviation-segmentation threshold adjustment coefficient mapping relationship, and the segmentation threshold adjustment coefficient is output. The corresponding preset initial segmentation threshold benchmark value and the segmentation threshold adjustment coefficient are multiplied to obtain the target image segmentation threshold.

[0026] If the thickness of the insulation layer is within the critical range of the insulation layer thickness, the thickness range to which the measured data belongs is determined to be the medium-grade insulation layer thickness range. The initial segmentation threshold benchmark value corresponding to the current medium-grade insulation layer thickness range remains unchanged. The critical range of insulation layer thickness represents the open interval formed by the critical lower boundary and the critical upper boundary of the insulation layer thickness.

[0027] If the insulation layer thickness is greater than or equal to the upper limit of the insulation layer thickness threshold, the thickness range to which the measured data belongs is determined to be the high-grade insulation layer thickness range. The difference between the current insulation layer thickness and the median thickness of the high-grade insulation layer thickness range is calculated to obtain the thickness deviation. The thickness deviation is input into the thickness deviation-segmentation threshold adjustment coefficient mapping relationship, and the segmentation threshold adjustment coefficient is output. The corresponding preset initial segmentation threshold benchmark value and the segmentation threshold adjustment coefficient are multiplied to obtain the target image segmentation threshold.

[0028] In this embodiment, the system addresses the shortcomings of traditional external wall insulation defect detection, which uses a fixed image segmentation threshold and cannot adapt to different insulation layer thicknesses, resulting in poor detection accuracy and frequent false positives and false negatives. It innovatively adopts a control mechanism that dynamically adjusts the image segmentation threshold based on the insulation layer thickness. In practice, the system first pre-divides the system into three gradient thickness ranges—low-grade, medium-grade, and high-grade—based on commonly used construction thickness specifications for external wall insulation in building construction. For each thickness range, an initial segmentation threshold benchmark value matching its imaging characteristics is pre-calibrated and stored. Specifically, the insulation layer thickness is defined as the overall vertical thickness of the insulation structure at the detection location of the external wall, from the outer surface of the base wall to the surface of the external wall finish layer, accurately reflecting the current detection depth. The system measures the physical specifications of the insulation structure. The initial segmentation threshold is a standard pixel feature threshold value specifically calibrated within the corresponding insulation layer thickness range, capable of stably distinguishing between densely bonded areas and defective areas in the image. The overall image segmentation threshold is used to accurately determine whether each pixel in the image belongs to a densely bonded area or a defective area. The corresponding pixel features specifically include at least one of image grayscale features and infrared thermal imaging temperature features, which can be adapted to dual-modal detection scenarios of visible light detection and infrared thermal imaging detection. At the same time, the system pre-constructs and solidifies the thickness deviation-segmentation threshold adjustment coefficient mapping relationship. This mapping relationship can uniquely characterize the one-to-one correspondence between the measured insulation layer thickness and the thickness deviation of the median thickness in the same thickness range and the segmentation threshold adjustment coefficient, providing data support for dynamic threshold correction.During actual testing, the system acquires real-time measured data of the insulation layer thickness corresponding to the current testing area. By comparing the measured insulation layer thickness with three preset thickness ranges, the system determines the target thickness range to which the current measured data belongs and automatically retrieves the initial segmentation threshold benchmark value corresponding to this target thickness range as the base threshold. Further, a differentiated dynamic threshold adjustment strategy is executed based on the thickness range type. Specifically, when the measured insulation layer thickness is less than or equal to the preset lower limit of the insulation layer thickness, the current testing area is determined to belong to the low-grade insulation layer thickness range. The system calculates the difference between the current measured insulation layer thickness and the median thickness of the low-grade insulation layer thickness range to obtain a precise thickness deviation. This thickness deviation is input into a preset thickness deviation-segmentation threshold adjustment coefficient mapping relationship, correspondingly outputting a segmentation threshold reduction coefficient adapted to the weak defect characteristics of thin insulation layers. By multiplying the retrieved initial segmentation threshold benchmark value with this segmentation threshold reduction coefficient, the final target image segmentation threshold adapted to thin insulation layer detection is obtained, thereby reducing the segmentation standard and improving shallow layer segmentation. The system possesses strong detection capabilities for micro-voids and weak adhesion defects. When the measured insulation layer thickness falls within the critical open interval formed by the lower and upper critical limits of the insulation layer thickness, the current detection area is determined to belong to the mid-range insulation layer thickness range of standard specifications. Within this range, the insulation layer imaging characteristics are stable and the defect discrimination is moderate, eliminating the need for threshold correction. The system directly maintains the initial segmentation threshold benchmark value corresponding to the mid-range insulation layer thickness range as constant and uses it as the final target image segmentation threshold, ensuring the stability and efficiency of detection under normal working conditions. When the measured insulation layer thickness falls within the critical open interval formed by the lower and upper critical limits of the insulation layer thickness, the system directly maintains the initial segmentation threshold benchmark value corresponding to the mid-range insulation layer thickness range as constant and uses it as the final target image segmentation threshold, ensuring the stability and efficiency of detection under normal working conditions. When the layer thickness is greater than or equal to the preset upper limit of the insulation layer thickness, the current detection area is determined to belong to the high-grade insulation layer thickness range. Similarly, the thickness deviation between the current measured insulation layer thickness and the median thickness of the high-grade insulation layer thickness range is calculated. After substituting the thickness deviation into the preset mapping relationship, the segmentation threshold adjustment coefficient is output. The target image segmentation threshold adapted to the thick insulation layer is obtained by multiplying the initial segmentation threshold benchmark value by the adjustment coefficient. By appropriately raising the segmentation threshold, the false defect detection problem caused by the surface texture difference and slight thermal field disturbance of the thick insulation layer is effectively avoided. Thus, the system realizes fully automatic and adaptive intelligent correction of the image segmentation threshold according to different insulation layer thickness conditions, completely solving the industry technical problems of traditional fixed thresholds being unable to adapt to the detection of multiple specifications of insulation layers, having poor versatility, and low accuracy.

[0029] It should be further explained that the specific process for determining whether to perform audio frequency offset-driven local threshold correction is as follows:

[0030] The real-time acoustic resonance frequency offset of the insulation area to be tested is obtained. The acoustic resonance frequency offset is the difference between the actual resonance frequency of the insulation layer to be tested and the reference resonance frequency of the preset standard dense bonding area. The reference resonance frequency of the preset standard dense bonding area is the standard reference frequency for determining whether the vibration of the insulation layer has shifted.

[0031] The real-time acoustic resonance frequency offset is compared with the preset offset threshold range. Based on the comparison result, the local grayscale threshold is dynamically adjusted. The local grayscale threshold is used to determine whether the corresponding local pixels belong to the abnormal area of ​​bonding defects.

[0032] A one-to-one mapping relationship between acoustic resonance frequency offset and local grayscale threshold correction coefficient is established in advance;

[0033] If the acoustic resonance frequency offset is within the preset offset threshold range, the current local grayscale threshold will remain unchanged. The preset offset threshold range represents the open interval formed by the lower limit of the preset offset threshold and the upper limit of the preset offset threshold.

[0034] The specific process for dynamically adjusting the local grayscale threshold is as follows:

[0035] If the acoustic resonance frequency offset is less than or equal to the preset offset threshold lower limit, the current acoustic resonance frequency offset is input into the corresponding mapping relationship, the local grayscale threshold downgrading factor is output, and the current local grayscale threshold is multiplied by the local grayscale threshold downgrading factor to obtain the target local grayscale threshold.

[0036] If the acoustic resonance frequency offset is greater than or equal to the preset offset threshold upper limit, the current acoustic resonance frequency offset is input into the corresponding mapping relationship, and the local grayscale threshold adjustment factor is output. The current local grayscale threshold and the local grayscale threshold adjustment factor are multiplied to obtain the target local grayscale threshold.

[0037] In this embodiment, to address the technical problems that a single image threshold segmentation method is difficult to adapt to different bonding defect levels in the external wall insulation layer, weak defect features are easily missed, and strong interference areas are easily misdetected, the system constructs an intelligent control mechanism based on real-time driving of local grayscale threshold adaptive correction by acoustic resonance frequency offset, thereby realizing deep coupling detection of acoustic vibration features and image grayscale features. In the specific implementation process, the system first pre-stores the reference resonance frequency of the calibrated standard dense bonded area. This reference resonance frequency is the natural vibration frequency of the structure obtained after applying a uniform standard excitation signal to a standard external wall insulation area with no voids, no debonding, and a dense and uniform bond. It serves as a benchmark for determining whether the vibration characteristics of the insulation area under test have deviated abnormally. During on-site testing, the system acquires the acoustic signal of the insulation area under test in real time and analyzes it to obtain the real-time acoustic resonance frequency offset. This acoustic resonance frequency offset is the difference between the actual resonance frequency generated by the external wall insulation layer under test after excitation and the reference resonance frequency of the preset standard dense bonded area. It can quantitatively characterize the degree of distortion of structural vibration characteristics caused by bonding defects such as cavities, debonding, and loose bonding inside the test area. The more severe the defect, the larger the absolute value of the frequency offset. The system pre-establishes a one-to-one quantization mapping relationship between acoustic resonance frequency offset and local grayscale threshold correction coefficient. At the same time, it presets an offset threshold open interval consisting of a lower offset threshold and an upper offset threshold to distinguish between normal bonding state and defective bonding state. Among them, the local grayscale threshold is a grayscale feature judgment critical value independently configured for the local test area of ​​the image. It is specifically used to accurately determine whether the pixels in the corresponding local area belong to the abnormal area of ​​bonding defect. It can realize local fine threshold control, which is different from the traditional global fixed grayscale threshold.During actual calibration and control, the system compares the real-time acoustic resonance frequency offset with the preset offset threshold range in real time. When the real-time acoustic resonance frequency offset is within the open range formed by the lower and upper limits of the preset offset threshold, it indicates that the vibration characteristics of the insulation area under test are basically consistent with the dense standard area, and there are no obvious bonding defects inside. The system maintains the local grayscale threshold of the current local area constant, without the need for threshold correction, ensuring the detection efficiency and judgment stability of the normal area. When the real-time acoustic resonance frequency offset is less than or equal to the lower limit of the preset offset threshold, it indicates that the insulation area under test has hidden weak defects such as shallow adhesion and micro-voids. The internal cavity volume is small, and the corresponding image grayscale anomaly features are weak and easily masked by environmental noise and wall texture. At this time, the system substitutes the current acoustic resonance frequency offset into the preset mapping relationship and matches and outputs the corresponding local grayscale threshold. The adjustment factor, obtained by multiplying the current initial local grayscale threshold with the adjustment factor, yields a target local grayscale threshold suitable for weak defect identification. By lowering the local grayscale judgment standard, the detection capability of shallow weak bonding defects is effectively improved, eliminating the problem of missed detection of micro-defects. When the real-time acoustic resonance frequency offset is greater than or equal to the upper limit of the preset offset threshold, it indicates that there are strong defects such as large-area debonding and severe hollowing in the insulation area under test. The internal cavity structure is significant, and the grayscale difference features of the image are too prominent. It is easy to misjudge normal texture color difference and slight lighting disturbance as defects. At this time, the system inputs the current acoustic resonance frequency offset into the mapping relationship, matches and outputs the corresponding local grayscale threshold adjustment factor, and multiplies the current local grayscale threshold with the adjustment factor to obtain the optimized target local grayscale threshold. By appropriately raising the local grayscale judgment standard, false defect interference is effectively filtered out, and the false detection probability of severe defect areas is significantly reduced. In summary, this embodiment corrects the local grayscale threshold of the image in real time by using acoustic vibration characteristics, and achieves adaptive dynamic adaptation of the image segmentation standard under different defect levels. It effectively makes up for the technical shortcomings of pure image processing methods, such as weak anti-interference, poor weak defect recognition ability, and easy misjudgment of strong defects, and significantly improves the accuracy and adaptability of the detection of bonding defects in the external wall insulation layer.

[0038] It should be understood that the specific process for defect detection is as follows:

[0039] The preprocessed image is segmented into pixels based on the adjusted image segmentation threshold, and all pixels belonging to the adhesion defects are initially marked.

[0040] The defective pixels are clustered and divided into independent defective connected regions. Spatially adjacent defective pixels with the same attributes are integrated into independent defective connected regions, and the area value corresponding to each independent defective connected region is calculated one by one.

[0041] Multi-level connected component area judgment thresholds are pre-set, and the area of ​​each defective connected component is compared with the corresponding judgment threshold in turn;

[0042] If the measured area of ​​the connected region of the defect is less than or equal to the threshold for eliminating the smallest pseudo-defect, then the connected region is determined to be a pseudo-defect region.

[0043] If the measured area of ​​the connected region is greater than the threshold for eliminating minimal false defects, the connected region is determined to be a valid defect region, and the measured area of ​​the connected region is compared with the preset micro-defect judgment threshold, specifically:

[0044] If the area of ​​the effective defect connected region is between the threshold for eliminating the smallest pseudo-defect and the threshold for determining the micro-defect, then the current connected region is determined to correspond to a point-like, dispersed shallow micro-bonding defect.

[0045] If the effective defect connected region area is greater than or equal to the micro-defect judgment threshold, the measured connected region area is compared with the preset large-area defect judgment threshold; if the measured connected region area is between the micro-defect judgment threshold and the large-area defect judgment threshold, it is judged as a local small-area bonding debonding or hollow defect.

[0046] If the measured area of ​​the connected domain is greater than or equal to the threshold for judging large-area defects, it is judged as a continuous large-area bonding failure defect.

[0047] Based on the comparison results, the identification and output of the bonding defects of the insulation layer are completed. The defect connected region area refers to the total number of pixels in the image that form the connected region of adjacent pixels that are identified as bonding defects.

[0048] In this embodiment, based on image preprocessing, dynamic control of insulation layer thickness threshold, and local grayscale threshold correction of acoustic resonance frequency offset, the system further completes refined bonding defect detection and genuine / fake differentiation through a multi-level defect connectivity area discrimination mechanism. This effectively solves the technical problems of traditional external wall insulation defect detection, such as difficulty in distinguishing noise pseudo-defects, inability to quantify defect scale, and inaccurate detection results caused by mixed judgment of micro-defects and large-area defects. In the specific implementation process, the system first relies on the final target image segmentation threshold and target local grayscale threshold obtained after thickness adaptive adjustment and audio frequency offset correction to perform global pixel segmentation processing on the pre-processed high-quality visible light image and infrared thermal imaging image of the exterior wall. Through precise pixel feature discrimination rules, it initially traverses and marks all bonding defect pixels in the image that meet the defect feature conditions, completing the initial screening and coarse marking of defect pixels. In order to avoid single-point misjudgment interference caused by scattered isolated pixels, random noise, and slight texture fluctuations of the wall surface, the system further performs spatial clustering processing on all initially marked defect pixels. Based on the principle of pixel spatial adjacency relationship and defect attribute consistency, discrete defect pixels that are spatially adjacent and have the same defect feature attributes are aggregated and integrated into several independent defect connected regions. The area of ​​the defect connected region is specifically defined as the total number of pixels in the connected region formed by all adjacent homogeneous pixels in the image that are judged by the system as bonding defects. It can intuitively and quantitatively reflect the actual distribution scale and disease range of a single bonding void, debonding, and loose bonding defect. The system is pre-configured and solidified with a multi-level connected domain area determination threshold system, which mainly includes a three-level gradient determination standard: a threshold for eliminating extremely small pseudo-defects, a threshold for determining micro-defects, and a threshold for determining large-area defects. This standard is used to identify connected domains of defects at different scales and to distinguish between pseudo-defect interference and real bonding defects.During the detection and judgment process, the system calculates the measured area of ​​each independent defect connected region within the image, and compares the measured area of ​​each connected region with preset thresholds at each level for progressive judgment: when the measured defect connected region area is less than or equal to the threshold for eliminating extremely small false defects, the system determines that this type of ultra-small area connected region is a false defect area caused by outdoor lighting disturbance, random image noise, slight color difference on the wall surface, and texture abrupt change, and directly eliminates it, not including it in the statistical results of real bonding defects, thus thoroughly filtering out invalid interference information from the data level; when the measured connected region area is greater than the threshold for eliminating extremely small false defects, the connected region is determined to be a real and valid defect area, and its measured area is further compared with the preset micro-defect judgment threshold; if the area of ​​the valid defect connected region is between the threshold for eliminating extremely small false defects and the micro-defect judgment threshold, it is said that... The system accurately identifies the defects in this area as point-like, scattered, shallow micro-adhesion defects, corresponding to minor localized loose adhesion and small hollow areas in the external wall insulation layer. If the effective defect connected area is greater than or equal to the micro-defect judgment threshold, a third-level comparison is performed between the measured connected area and the preset large-area defect judgment threshold. When the measured connected area is between the micro-defect judgment threshold and the large-area defect judgment threshold, the area is judged as a localized small-area adhesion detachment and localized hollow area defect, belonging to the category of conventional localized adhesion failure defects. When the measured connected area is greater than or equal to the large-area defect judgment threshold, the area is judged as a continuous large-area adhesion failure defect, indicating that there is a large-scale, continuous hollow and detachment problem in the insulation layer, posing a high risk of detachment. Through the above-mentioned multi-level connected domain area hierarchical comparison, step-by-step screening and classification process, the system can accurately eliminate image pseudo-defect interference, effectively distinguish between micro-defects, small-area defects and large-area contiguous defects, and achieve refined, quantitative and structured intelligent identification of bonding defects in building exterior wall insulation layers. Finally, it outputs accurate, reliable and quantifiable defect detection results, providing accurate pre-defect data support for subsequent panoramic defect mapping, defect distribution density calculation and defect severity classification.

[0049] As the third module of this system, the sequence image panoramic stitching and global integration module is used to receive continuous inspection sequence images, and to fuse them with overlapping areas. It stitches multiple single-frame images into a complete panoramic image of the building's exterior wall facade, and maps the single-frame defect detection results to the panoramic image, thereby realizing the global detection of bonding defects in the building's exterior wall insulation layer.

[0050] It needs to be explained that the specific process for merging overlapping areas is as follows:

[0051] Feature points in the overlapping area of ​​two adjacent consecutive inspection images are extracted, the matching similarity of all feature point pairs is calculated, and the matching similarity of each feature point pair is compared with the preset feature point matching threshold. If the matching similarity of each feature point pair is greater than or equal to the preset feature point matching threshold, it is retained and marked as a valid feature point.

[0052] The feature point matching threshold is a similarity threshold used to determine whether the feature point pairs extracted from overlapping regions of adjacent images are valid matching feature points.

[0053] If the matching similarity of each feature point pair is less than the preset feature point matching threshold, then it is discarded;

[0054] Based on the obtained effective feature point pairs, the horizontal and vertical offsets of pixel coordinates between adjacent images are solved to obtain the pixel coordinate offset. The pixel coordinate offset is the pixel deviation value in the horizontal and vertical dimensions of the pixel coordinates corresponding to the same physical position in the overlapping area of ​​two adjacent inspection sequence images.

[0055] The fusion processing with overlapping regions also includes:

[0056] Based on the pixel coordinate offset, the pixel coordinate translation correction and alignment transformation of the subsequent frame image are performed to eliminate pixel misalignment, ghosting and offset problems in the overlapping area of ​​adjacent images, and to achieve geometric registration of the overlapping area of ​​the preceding and following frames.

[0057] After completing the geometric registration of the overlapping area of ​​the images in the preceding and following frames, weighted fusion, brightness equalization and texture transition processing are performed on the pixels in the overlapping area to achieve seamless fusion of the overlapping areas of adjacent image sequences.

[0058] In this embodiment, the system addresses the problems of perspective deviation, pixel misalignment, scale inconsistency, and image overlap caused by flight jitter, camera position shift, slight changes in shooting angle, and fluctuations in acquisition distance during continuous inspection of exterior walls by UAVs. It constructs a high-precision image registration and seamless fusion mechanism for overlapping areas based on feature point matching and pixel coordinate offset correction, which effectively solves the defects of traditional image stitching, such as ghosting, misalignment, stitching seams, texture breaks, and fragmentation of full-domain detection data. In the specific implementation process, the system first receives a sequence of multiple frames of external wall insulation layer inspection images continuously collected by the UAV. Then, it performs overlapping area feature detection and feature point extraction operations on adjacent previous and subsequent frames in turn. It comprehensively extracts stable feature information such as corner points, edge features, and texture key points in the overlapping area of ​​the two images. The system then performs pairwise matching operations on the two sets of extracted feature points and calculates the feature matching similarity between all feature point pairs one by one to characterize the degree of correspondence between the two sets of feature points. The system has a fixed and configurable feature point matching threshold. This feature point matching threshold is a similarity threshold specifically used to determine whether feature point pairs in the overlapping area of ​​adjacent images have a valid matching relationship. It is the core judgment criterion for screening true corresponding features and eliminating mismatched interference features. After calculating the similarity of feature point pairs, the system compares the matching similarity of each pair with a preset feature point matching threshold. If the matching similarity of a feature point pair is greater than or equal to the preset threshold, the pair is considered to have high matching accuracy and corresponds to the same physical location on the wall, and is retained and marked as a valid feature point pair. If the matching similarity is less than the preset threshold, the pair is considered to be a mismatch, incorrect match, or invalid interference feature, and is directly removed to avoid mismatched features interfering with subsequent registration accuracy. Based on obtaining a sufficient number of accurate valid feature point pairs, the system further calculates the horizontal and vertical pixel offsets between adjacent images through the coordinate correspondence of the valid feature points, thus obtaining the complete pixel coordinate offset. The pixel coordinate offset specifically refers to the pixel deviation in the horizontal and vertical dimensions of the pixel coordinates corresponding to the same physical location on the wall within the overlapping area of ​​two adjacent inspection sequence images, which can accurately quantify the overall misalignment and offset magnitude between image frames. After obtaining the precise pixel coordinate offset, the system performs global pixel coordinate translation correction and precise alignment transformation on the subsequent frame image based on the offset parameter. By compensating for the pixel coordinates, the system corrects the positional deviation between the preceding and following frames, completely eliminating pixel misalignment, image ghosting, local offset, and geometric distortion problems in the overlapping areas of adjacent images. This achieves high-precision geometric registration of the overlapping areas of the preceding and following frames, ensuring the uniformity of the spatial coordinate system of multiple frames.Based on the accurate geometric registration of adjacent images, the system further performs refined fusion processing on pixels in overlapping areas. This includes weighted fusion of overlapping pixels, global brightness balancing, and smooth texture transition processing. An adaptive weight allocation method balances the pixel information ratio in overlapping areas of consecutive frames, and combined with brightness and color uniformity correction, it eliminates inter-frame illumination differences, color deviations, and brightness banding, resulting in natural texture transitions and complete pixel information in overlapping areas. Ultimately, this achieves seamless fusion of overlapping areas in adjacent inspection sequence images. This mechanism, through a combination of precise feature point selection, quantitative pixel offset correction, geometric registration, and weighted fusion, effectively solves the problems of low stitching accuracy, image distortion, and large defect location mapping deviations in complex outdoor working conditions. It ensures that the stitched panoramic exterior wall image has continuous texture, complete boundaries, and no stitching defects, providing high-precision and high-completeness panoramic image data support for subsequent full-area mapping, full-coverage detection, and overall quality assessment of defects in the entire exterior wall insulation layer.

[0059] As the fourth module of this system, the defect classification and quantitative evaluation module is used to classify the detected bonding defects into multiple levels of severity based on the defect distribution density in the image, so as to complete the bonding quality inspection of the external wall insulation.

[0060] It is important to understand that the specific process for classifying the severity of detected bonding defects into multiple levels is as follows:

[0061] The stitched panoramic inspection image of the exterior wall is pre-divided into several regular local evaluation sub-regions of uniform size. Based on each effective evaluation sub-region, the total area of ​​all effective bonding defects after filtering by connected components within the sub-region is counted. Based on the total area of ​​effective detection pixels in the sub-region, the defect distribution density of each sub-region is calculated in real time.

[0062] The defect distribution density is the ratio of the total area of ​​all effective defect connected regions to the total area of ​​the corresponding effective pixels in the local detection area of ​​the preset external wall insulation layer.

[0063] The system has a three-level gradient density determination mechanism with low, medium, and high density thresholds. The real-time calculated defect distribution density is compared step-by-step with each density threshold level.

[0064] If the defect distribution density is less than the low density threshold, it is classified as a first-level minor defect.

[0065] If the defect distribution density is between the low density threshold and the medium density threshold, it is classified as a second-level general defect level.

[0066] If the defect distribution density is between the medium density threshold and the high density threshold, it is classified into three levels of relatively severe defects.

[0067] If the defect distribution density is greater than or equal to the high density threshold, it is classified as a level four severe defect.

[0068] In this embodiment, to address the shortcomings of traditional external wall insulation defect detection methods—which can only identify individual defects, cannot quantify the degree of defect aggregation in a region, have strong subjectivity in defect level determination, and use a single, crude grading standard—the system establishes a multi-level quantitative defect severity grading mechanism based on the defect distribution density of panoramic images. This achieves a precise upgrade from single-point detection of external wall insulation bonding defects to a comprehensive, regionalized, gridded, and graded assessment. In the specific implementation process, the system first uses the aforementioned sequential image panoramic stitching technology to obtain a complete, seamless, and distortion-free panoramic detection image of the external wall insulation layer. Using this panoramic image as the basis for comprehensive assessment, the entire panoramic image is pre-divided into regular grids, uniformly dividing the complete external wall panoramic detection image into several local assessment sub-regions of uniform size, consistent specifications, and regular boundaries. Simultaneously, it automatically identifies and removes ineffective assessment sub-regions containing external wall backgrounds, obstructions, or invalid wall areas, retaining only the effective assessment sub-regions covering the actual external wall insulation layer for subsequent defect quantitative analysis. For each selected effective evaluation sub-region, the system calls upon the aforementioned connected component filtering results to accurately count all real and effective bond defect connected components within each sub-region after pseudo-defect removal and scale filtering. The total area of ​​all effective defect connected components within a single evaluation sub-region is calculated by summing them up. Simultaneously, the total area of ​​effective detection pixels on the insulation wall surface within the corresponding sub-region is also counted. Based on these two core data, the defect distribution density corresponding to each local evaluation sub-region is calculated in real time. Among them, the defect distribution density is the ratio of the total area of ​​all effective defect connected components within the preset local detection area of ​​the external wall insulation layer to the total area of ​​effective pixels in the corresponding detection area. This can accurately quantify the density and overall deterioration of bond defects in the local insulation area. Unlike the limitation that a single defect area can only reflect the size of a single defect, this can effectively reflect the overall degradation state of the bond quality of the area.To achieve refined graded assessment, the system pre-configures a three-tiered density judgment standard with low, medium, and high density thresholds, forming a progressive and interval-independent defect level judgment system. During the detection process, the defect distribution density of each sub-region calculated in real time is compared with the preset density thresholds at each level to complete the automated grading of defect severity: when the defect distribution density of a local sub-region is less than the preset low density threshold, it indicates that the number of bonding defects in the area is small, the distribution is sparse, the scope of the defect is minimal, and the overall bonding performance of the insulation layer is basically intact. The system classifies the defect status of this area as a first-level minor defect. When the defect distribution density of a local sub-region is between the low and medium density thresholds, it indicates that multiple scattered defects have appeared in the area. The bonding defects are generally present, with an increased frequency of defects and a decrease in local bonding stability, but no concentrated or continuous defects have yet formed. The system classifies this as a Level 2 general defect. When the defect distribution density in a local sub-region is between the medium and high density thresholds, it indicates that the bonding defects in that area are significantly clustered, with a high defect coverage ratio. Local voids and debonding problems are concentrated, the bonding reliability of the insulation layer decreases, and there are certain safety hazards. The system classifies this as a Level 3 relatively serious defect. When the defect distribution density in a local sub-region is greater than or equal to the preset high density threshold, it indicates that the defects in that area are densely clustered and the defects are spreading over a large area. The overall bonding failure of the insulation layer is serious, with an extremely high risk of detachment, water seepage, and safe detachment. The system classifies this as a Level 4 severe defect. This embodiment, through grid-based zoning statistics, density quantification calculation, and multi-level gradient threshold comparison, breaks away from the traditional crude mode of judging the severity of defects by relying on manual experience or single defect size. It achieves standardized, quantified, and refined intelligent classification of the severity of bonding defects in external wall insulation layers, and can accurately distinguish between four types of defects: minor, general, moderate, and severe. This provides comprehensive, objective, and quantitative graded data support for the quality acceptance of building external wall insulation, defect risk assessment, targeted operation and maintenance rectification, and subsequent quality traceability.

[0069] like Figure 3The diagram shows a flowchart of an intelligent detection method for the bonding quality of building exterior wall insulation layer provided in this application embodiment. The method includes: simultaneously acquiring visible light and infrared thermal imaging images of the building exterior wall insulation area to be inspected and preprocessing them to obtain single-frame images and continuous inspection sequence images containing the exterior wall insulation layer; determining whether to perform intelligent adjustment of dynamically updated image segmentation thresholds based on the insulation layer thickness; if yes, performing adaptive precise segmentation after intelligent adjustment; otherwise, directly performing adaptive precise segmentation; during the adaptive precise segmentation process, acquiring the acoustic resonance frequency offset and determining whether to perform audio frequency offset-driven local threshold correction; if yes, performing defect detection after correction; otherwise, directly performing defect detection; receiving continuous inspection sequence images and fusing them with overlapping areas to stitch multiple single-frame images into a complete panoramic image of the building exterior wall facade; mapping the single-frame defect detection results to the panoramic image to achieve full-area detection of bonding defects in the building exterior wall insulation layer; and classifying the detected bonding defects into multiple severity levels based on the image defect distribution density to complete the exterior wall insulation bonding quality detection.

[0070] In this embodiment, the intelligent detection system for the bonding quality of the building's exterior wall insulation layer adopts an integrated intelligent detection process in actual detection operations, which includes multimodal image acquisition, dual-parameter dynamic threshold control, panoramic image fusion registration, and quantitative grading evaluation. This process achieves automated, high-precision, and full-coverage detection of bonding defects in the exterior wall insulation layer, effectively overcoming the industry drawbacks of traditional manual tapping detection and single-image detection methods, which are characterized by strong subjectivity, high rates of missed and false detections, limited detection range, and inability to achieve full-area quantitative rating. In practice, the system first uses the image acquisition and preprocessing module to simultaneously acquire visible light and infrared thermal images of the exterior wall insulation area of ​​the building to be inspected. Relying on dual imaging units, it achieves simultaneous acquisition of visual texture features and temperature field features. At the same time, it performs professional preprocessing and optimization on the original images to address complex working conditions such as uneven lighting, air dust, environmental shadows, and temperature and humidity interference at the outdoor construction site. This includes image noise reduction, distortion correction, grayscale equalization, detail enhancement, and invalid background removal. Ultimately, it obtains clear, feature-complete, and interference-free single-frame images of the exterior wall insulation layer and continuous inspection sequence images collected by UAV patrols, providing high-quality basic image data for subsequent intelligent segmentation and defect detection. After completing image acquisition and preprocessing, the system initiates an adaptive threshold intelligent control and image segmentation process through the dynamic threshold image segmentation and defect judgment module. The system first acquires the insulation layer thickness parameters of the area to be detected in real time, and determines whether dynamic intelligent control of the image segmentation threshold needs to be performed based on the measured insulation layer thickness data. If threshold intelligent control is required, the system dynamically adjusts or lowers the benchmark threshold according to the preset multi-level insulation layer thickness range and the thickness deviation-segmentation threshold adjustment coefficient mapping relationship to obtain the target image segmentation threshold that is suitable for the current insulation specification before performing the image adaptive precise segmentation operation. If threshold control is not required, the preset benchmark segmentation threshold is directly called to complete the image adaptive precise segmentation. During the precise image segmentation process, the system simultaneously acquires acoustic vibration signals from the insulation area under test and analyzes the acoustic resonance frequency offset. It then determines whether a local grayscale threshold correction operation driven by the acoustic frequency offset is required. If local threshold correction is required, the corresponding threshold correction factor is matched based on the deviation amplitude of the acoustic resonance frequency offset to complete the adaptive upward or downward adjustment optimization of the local grayscale threshold. After obtaining the accurate target local grayscale threshold, refined defect detection is carried out. If local threshold correction is not required, the system directly completes the entire defect detection process based on the current threshold parameters, including defect pixel screening, connected component clustering, false defect removal, and real defect determination, to accurately identify various bonding failure defects in the external wall insulation layer, such as micro-voids, local loose adhesion, and large-area debonding.Meanwhile, the system continuously receives and collects continuous inspection sequence images through the sequence image panoramic stitching and global integration module. Through feature point matching and filtering, pixel coordinate offset calculation, image translation alignment correction and overlapping area weighted fusion, brightness equalization, and texture transition processing, it completes the seamless stitching and fusion of multiple fragmented single-frame images, integrating multiple local single-frame detection images into a complete panoramic image covering the entire building exterior wall facade. At the same time, it accurately maps the detection data such as defect detection coordinates, defect type, and defect area corresponding to all single-frame images to the panoramic image coordinate system, completely solving the problem of fragmented traditional single-frame detection data and the inability to achieve global overall analysis, and completing the global detection and positioning of bonding defects in the exterior wall insulation layer without blind spots. Finally, the system uses a defect grading and quantitative evaluation module to quantify and grade all bonding defects detected in the panoramic image. By dividing the panoramic image into gridded sub-regions, the system calculates the total area of ​​the effective defect connected domains and the area of ​​the effective detection pixels in each effective sub-region, calculates the defect distribution density in each region in real time, and completes step-by-step comparison and judgment based on preset three-level gradient thresholds of low density, medium density, and high density. According to the density of defect aggregation, the bonding defects in different regions are divided into four levels: Level 1 (minor), Level 2 (general), Level 3 (relatively severe), and Level 4 (severe), realizing a quantitative grading and evaluation of the severity of defects. Finally, the system completes the intelligent, precise, and quantitative inspection of the bonding quality of the exterior wall insulation layer of the entire building.

[0071] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An intelligent detection system for the bonding quality of exterior wall insulation layers in building construction, characterized in that, include: The module includes an image acquisition and preprocessing module, a dynamic threshold image segmentation and defect determination module, a sequence image panoramic stitching and global integration module, and a defect grading and quantitative evaluation module. The image acquisition and preprocessing module is used to simultaneously acquire visible light images and infrared thermal imaging images of the exterior wall insulation area of ​​the building to be inspected and preprocess them to obtain single-frame images and continuous inspection sequence images containing the exterior wall insulation layer. The dynamic threshold image segmentation and defect determination module is used to determine whether to perform dynamic update image segmentation threshold intelligent control based on the thickness of the insulation layer. If yes, adaptive precise segmentation is performed after intelligent control; otherwise, adaptive precise segmentation is performed directly. During the adaptive precise segmentation process, the acoustic resonance frequency offset is obtained and it is determined whether to perform audio frequency offset driven local threshold correction. If yes, defect detection is performed after correction; otherwise, defect detection is performed directly. The sequence image panoramic stitching and global integration module is used to receive continuous inspection sequence images and fuse them with overlapping areas to stitch multiple single-frame images into a complete panoramic image of the building's exterior wall facade. It also maps the single-frame defect detection results to the panoramic image to achieve global detection of bonding defects in the building's exterior wall insulation layer. The defect grading and quantification assessment module is used to classify the detected bonding defects into multiple levels of severity based on the defect distribution density in the image, so as to complete the bonding quality inspection of the external wall insulation. The specific procedure for classifying the severity of detected bonding defects into multiple levels is as follows: The stitched panoramic inspection image of the exterior wall is pre-divided into several regular local evaluation sub-regions of uniform size. Based on each effective evaluation sub-region, the total area of ​​all effective bonding defects after filtering by connected components within the sub-region is counted. Based on the total area of ​​effective detection pixels in the sub-region, the defect distribution density of each sub-region is calculated in real time. The defect distribution density is the ratio of the total area of ​​all effective defect connected regions to the total area of ​​the corresponding effective pixels in the local detection area of ​​the preset external wall insulation layer. The system has a three-level gradient density determination mechanism with low, medium, and high density thresholds. The real-time calculated defect distribution density is compared step-by-step with each density threshold level. If the defect distribution density is less than the low density threshold, it is classified as a first-level minor defect. If the defect distribution density is between the low density threshold and the medium density threshold, it is classified as a second-level general defect level. If the defect distribution density is between the medium density threshold and the high density threshold, it is classified into three levels of relatively severe defects. If the defect distribution density is greater than or equal to the high density threshold, it is classified as a level four severe defect.

2. The intelligent detection system for the bonding quality of building exterior wall insulation layer as described in claim 1, characterized in that: The specific process for determining whether to perform intelligent adjustment of dynamically updated image segmentation thresholds is as follows: Multiple insulation layer thickness ranges are pre-divided and corresponding to preset initial segmentation threshold benchmarks. The measured insulation layer thickness data of the current detection area is obtained in real time. The thickness range to which the measured data belongs is determined, and the initial segmentation threshold benchmark of the corresponding range is retrieved. The multiple insulation layer thickness ranges include low-grade insulation layer thickness range, medium-grade insulation layer thickness range and high-grade insulation layer thickness range. The insulation layer thickness is the overall vertical thickness of the insulation structure layer at the location of the exterior wall to be tested, from the surface of the base wall to the surface of the exterior finish layer. The initial segmentation threshold benchmark value is the standard pixel feature critical value that distinguishes between the densely bonded area and the bond defect area in the image under the corresponding insulation layer thickness range. The thickness of the insulation layer is compared with the pre-divided multiple insulation layer thickness ranges. Based on the comparison results, the image segmentation threshold is dynamically adjusted. The image segmentation threshold is a pixel feature critical value used to determine whether an image pixel belongs to a densely bonded area or a bond defect area. A thickness deviation-segmentation threshold adjustment coefficient mapping relationship is pre-constructed. This mapping relationship is used to characterize the one-to-one mapping relationship between the difference between the insulation layer thickness and the median thickness of the corresponding thickness interval and the segmentation threshold adjustment coefficient.

3. The intelligent detection system for bonding quality of building exterior wall insulation layer as described in claim 2, characterized in that: The specific process for dynamically adjusting the image segmentation threshold is as follows: If the insulation layer thickness is less than or equal to the critical lower limit of the insulation layer thickness, the thickness range to which the measured data belongs is determined to be the low-grade insulation layer thickness range. The difference between the current insulation layer thickness and the median thickness of the low-grade insulation layer thickness range is calculated to obtain the thickness deviation. The thickness deviation is input into the thickness deviation-segmentation threshold adjustment coefficient mapping relationship, and the segmentation threshold adjustment coefficient is output. The corresponding preset initial segmentation threshold benchmark value and the segmentation threshold adjustment coefficient are multiplied to obtain the target image segmentation threshold. If the thickness of the insulation layer is within the critical range of the insulation layer thickness, the thickness range to which the measured data belongs is determined to be the medium-grade insulation layer thickness range. The initial segmentation threshold benchmark value corresponding to the current medium-grade insulation layer thickness range remains unchanged. The critical range of the insulation layer thickness represents the open interval formed by the critical lower boundary and the critical upper boundary of the insulation layer thickness. If the insulation layer thickness is greater than or equal to the upper limit of the insulation layer thickness threshold, the thickness range to which the measured data belongs is determined to be the high-grade insulation layer thickness range. The difference between the current insulation layer thickness and the median thickness of the high-grade insulation layer thickness range is calculated to obtain the thickness deviation. The thickness deviation is input into the thickness deviation-segmentation threshold adjustment coefficient mapping relationship, and the segmentation threshold adjustment coefficient is output. The corresponding preset initial segmentation threshold benchmark value and the segmentation threshold adjustment coefficient are multiplied to obtain the target image segmentation threshold.

4. The intelligent detection system for the bonding quality of building exterior wall insulation layer as described in claim 1, characterized in that: The specific process for determining whether to perform audio frequency offset-driven local threshold correction is as follows: The real-time acoustic resonance frequency offset of the insulation area to be tested is obtained. The acoustic resonance frequency offset is the difference between the actual resonance frequency of the insulation layer to be tested and the reference resonance frequency of the preset standard dense bonding area. The reference resonance frequency of the preset standard dense bonding area is the standard reference frequency for determining whether the vibration of the insulation layer has shifted. The real-time acoustic resonance frequency offset is compared with the preset offset threshold range. Based on the comparison result, the local grayscale threshold is dynamically adjusted. The local grayscale threshold is used to determine whether the corresponding local pixel belongs to the bonding defect abnormal area. A one-to-one mapping relationship between acoustic resonance frequency offset and local grayscale threshold correction coefficient is established in advance; If the acoustic resonance frequency offset is within the preset offset threshold range, the current local grayscale threshold remains unchanged. The preset offset threshold range represents the open interval formed by the lower limit of the preset offset threshold and the upper limit of the preset offset threshold.

5. The intelligent detection system for bonding quality of building exterior wall insulation layer as described in claim 4, characterized in that: The specific process for dynamically adjusting the local grayscale threshold is as follows: If the acoustic resonance frequency offset is less than or equal to the preset offset threshold lower limit, the current acoustic resonance frequency offset is input into the corresponding mapping relationship, the local grayscale threshold downgrading factor is output, and the current local grayscale threshold is multiplied by the local grayscale threshold downgrading factor to obtain the target local grayscale threshold. If the acoustic resonance frequency offset is greater than or equal to the preset offset threshold upper limit, the current acoustic resonance frequency offset is input into the corresponding mapping relationship, and the local grayscale threshold adjustment factor is output. The current local grayscale threshold and the local grayscale threshold adjustment factor are multiplied to obtain the target local grayscale threshold.

6. The intelligent detection system for bonding quality of building exterior wall insulation layer as described in claim 1, characterized in that: The specific process for defect detection is as follows: The preprocessed image is segmented into pixels based on the adjusted image segmentation threshold, and all pixels belonging to the adhesion defects are initially marked. The defective pixels are clustered and divided into independent defective connected regions. Spatially adjacent defective pixels with the same attributes are integrated into independent defective connected regions, and the area value corresponding to each independent defective connected region is calculated one by one. Multi-level connected component area judgment thresholds are pre-set, and the area of ​​each defective connected component is compared with the corresponding judgment threshold in turn; If the measured area of ​​the connected region of the defect is less than or equal to the threshold for eliminating the smallest pseudo-defect, then the connected region is determined to be a pseudo-defect region. If the measured area of ​​the connected region is greater than the threshold for eliminating minimal false defects, the connected region is determined to be a valid defect region, and the measured area of ​​the connected region is compared with the preset micro-defect judgment threshold, specifically: If the area of ​​the effective defect connected region is between the threshold for eliminating the smallest pseudo-defect and the threshold for determining the micro-defect, then the current connected region is determined to correspond to a point-like, dispersed shallow micro-bonding defect. If the effective defect connected region area is greater than or equal to the micro-defect judgment threshold, the measured connected region area is compared with the preset large-area defect judgment threshold; if the measured connected region area is between the micro-defect judgment threshold and the large-area defect judgment threshold, it is judged as a local small-area bonding debonding or hollow defect. If the measured area of ​​the connected domain is greater than or equal to the threshold for judging large-area defects, it is judged as a continuous large-area bonding failure defect. Based on the comparison results, the identification and output of the bonding defects of the insulation layer are completed. The defect connected region area refers to the total number of pixels in the image that form the connected region of adjacent pixels that are identified as bonding defects.

7. The intelligent detection system for bonding quality of building exterior wall insulation layer as described in claim 1, characterized in that: The specific process for merging overlapping regions is as follows: Feature points in the overlapping area of ​​two adjacent consecutive inspection images are extracted, the matching similarity of all feature point pairs is calculated, and the matching similarity of each feature point pair is compared with the preset feature point matching threshold. If the matching similarity of each feature point pair is greater than or equal to the preset feature point matching threshold, it is retained and marked as a valid feature point. The feature point matching threshold is a similarity threshold used to determine whether the feature point pairs extracted from the overlapping areas of adjacent images are valid matching feature points. If the matching similarity of each feature point pair is less than the preset feature point matching threshold, then it is discarded; Based on the obtained effective feature point pairs, the horizontal and vertical offsets of pixel coordinates between adjacent images are solved to obtain the pixel coordinate offset. The pixel coordinate offset is the pixel deviation value in the horizontal and vertical dimensions of the pixel coordinates corresponding to the same physical position in the overlapping area of ​​two adjacent inspection sequence images.

8. The intelligent detection system for bonding quality of building exterior wall insulation layer as described in claim 7, characterized in that: The fusion process with overlapping regions also includes: Based on the pixel coordinate offset, the pixel coordinate translation correction and alignment transformation of the subsequent frame image are performed to eliminate pixel misalignment, ghosting and offset problems in the overlapping area of ​​adjacent images, and to achieve geometric registration of the overlapping area of ​​the preceding and following frames. After completing the geometric registration of the overlapping area of ​​the images in the preceding and following frames, weighted fusion, brightness equalization and texture transition processing are performed on the pixels in the overlapping area to achieve seamless fusion of the overlapping areas of adjacent image sequences.

9. A method for applying the intelligent detection system for bonding quality of building exterior wall insulation layer as described in any one of claims 1-8, characterized in that, include: Visible light images and infrared thermal imaging images of the exterior wall insulation area of ​​the building to be inspected are acquired simultaneously and preprocessed to obtain single-frame images and continuous inspection sequence images containing the exterior wall insulation layer. The system determines whether to perform dynamic image segmentation threshold intelligent adjustment based on the thickness of the insulation layer. If yes, adaptive precise segmentation is performed after intelligent adjustment. If no, adaptive precise segmentation is performed directly. During the adaptive precise segmentation process, the acoustic resonance frequency offset is obtained and it is determined whether to perform audio frequency offset-driven local threshold correction. If yes, defect detection is performed after correction. If no, defect detection is performed directly. Receive continuous inspection sequence images and fuse them with overlapping areas. Stitch multiple single-frame images into a complete panoramic image of the building's exterior wall facade. Map the single-frame defect detection results to the panoramic image to achieve full-area detection of bonding defects in the building's exterior wall insulation layer. The detected bonding defects are classified into multiple severity levels based on the defect distribution density in the image to complete the quality inspection of the bonding of the external wall insulation.

Images