Nondestructive testing method for hidden hollowing and stripping of historical building

Abstract

The invention provides a nondestructive detection method for hidden hollowing and stripping of a historical building, and relates to the technical field of building detection, and the method comprises the following steps: carrying out scanning detection on the surface of a to-be-detected historical building, so as to collect an acoustic vibration response signal sequence and an infrared thermal imaging sequence of the to-be-detected historical building; based on data information containing acoustic-thermal response features of a plurality of historical building materials and repairing materials, adaptive fusion and feature extraction are carried out on the acoustic vibration response signal sequence and the infrared thermal imaging sequence so as to suppress environment and surface texture interference, and a fusion feature map is output; according to the fused feature map, identifying and distinguishing spatial positions and distribution features of deep hollowing defects and shallow peeling defects under the surface of the historical building; and based on the spatial position and the distribution characteristics, generating and outputting visual defect distribution information for guiding repair. According to the method, deep fusion and collaborative enhancement of multi-physics field information are realized, and the decision-making requirement of fine repair is met.

Description

Technical Field

[0001] This application relates to the field of building inspection technology, specifically to a non-destructive testing method for concealed hollowing and peeling in historical buildings. Background Technology

[0002] Hollowing and peeling defects in historical buildings are hidden within the structure, and their accurate and non-destructive testing has always been a major challenge in the field of cultural relic preservation. Traditional methods, which mainly rely on experience-based judgment by tapping, are inefficient, highly subjective, and risk damaging fragile finishes. Modern non-destructive testing technologies, such as infrared thermal imaging, are easily affected by environmental temperature changes, sunlight and shadows, and complex textures on the building surface (such as painted decorations, brick joints, and reliefs), making it difficult to distinguish between false anomalies caused by material inhomogeneity and historical repairs and genuine defects. On the other hand, single stress wave or acoustic vibration methods are not sensitive to shallow peeling and cannot quantify the depth of defects. Existing technologies usually use these methods alone or in simple series, lacking a mechanism for deep integration and collaborative analysis. This results in a high rate of missed and false diagnoses when dealing with historical buildings with heterogeneous materials and complex stratification, making it difficult to meet the decision-making needs of refined restoration. Summary of the Invention

[0003] In view of the above-mentioned defects or deficiencies in the prior art, this application aims to provide a non-destructive testing method for concealed hollowing and peeling in historical buildings, including the following steps: The surface of the historical building to be tested is scanned to collect its acoustic vibration response signal sequence and infrared thermal imaging sequence; Based on data containing acoustic and thermal response characteristics of various historical building materials and repair materials, the acoustic vibration response signal sequence and the infrared thermal imaging sequence are adaptively fused and feature extracted to suppress environmental and surface texture interference and output a fused feature map. Based on the fused feature map, identify and distinguish the spatial location and distribution characteristics of deep hollow defects and shallow peeling defects under the surface of the historical building; Based on the spatial location and distribution characteristics, visualized defect distribution information is generated and output to guide repair.

[0004] According to the technical solution provided in this application, the step of identifying and distinguishing the spatial location and distribution characteristics of deep hollow defects and shallow peeling defects based on the fused feature map includes the following steps: Time-frequency analysis is performed on the acoustic vibration response signal sequence to extract the resonant main frequency and frequency band energy distribution of each measurement point in the preset frequency band, thus forming a power spectrum feature vector; The infrared thermal imaging sequence is subjected to time-series analysis to extract the curve of temperature change over time at each measuring point, and the time required to reach half-peak temperature rise is calculated to form the thermal conduction time constant feature. The power spectrum feature vector and the heat conduction time constant feature under the same spatial coordinates are coupled and input into the hierarchical classification model; The deep hierarchical classification model outputs a hierarchical probability prediction of the defect state at the measurement point based on the input coupling features. The defect state includes: no defect, shallow peeling, deep hollowing, and coexistence of shallow and deep defects.

[0005] According to the technical solution provided in this application, after coupling the power spectrum feature vector and the heat conduction time constant feature under the same spatial coordinates and inputting them into the hierarchical classification model, the following steps are also included: For target measurement points identified as having defects, perform defect interface location analysis, including: The target measuring point is sequentially subjected to a preheating excitation with a first power density and a main thermal excitation with a higher power density, and high frame rate infrared image sequences and acoustic vibration signals are acquired simultaneously during the two excitation phases. From the high frame rate infrared image sequence, calculate the additional temperature rise curve of the main thermal excitation stage relative to the preheating excitation stage, and extract the additional temperature rise time constant of the curve as the interface thermal resistance feature. The acoustic vibration signals of the main thermal excitation stage and the preheating excitation stage are differentially processed to obtain differential acoustic vibration signals. The energy ratio of the signals in the preset frequency band is extracted as the interface mechanical state characteristics. By combining the interface thermal resistance characteristics and interface mechanical state characteristics with the material layer information corresponding to the target measurement point in the fused feature map, the specific material interface where the defect is located is determined. The specific material interface includes: the interior of the finishing layer, the interface between the finishing layer and the mortar layer, and the interface between the mortar layer and the base layer.

[0006] According to the technical solution provided in this application, the acquisition of its acoustic vibration response signal sequence and infrared thermal imaging sequence includes the following steps: A global pre-scan of the surface of the historical building was performed to obtain visible light images and infrared preview images; Based on the infrared preview image, the initial screening area for temperature anomalies is obtained; Based on the visible light image, regions of abrupt changes in surface material, regions of historical repair marks, and regions of decorative textures are identified and marked; and combined with the initial screening region of temperature anomalies, a fusion guide map is generated to mark key feature regions. Based on the fusion guidance map, the acquisition parameters are planned, including the scanning path and the spatial density of acquisition points; Based on the acquisition parameters, the acoustic vibration response signal sequence and infrared thermal imaging sequence are acquired.

[0007] According to the technical solution provided in this application, the generation of the fusion guide map marking key feature regions includes the following steps: The visible light image and the infrared preview image are registered and feature analyzed to identify abnormal regions in visible light and abnormal regions in infrared temperature. Based on the feature coupling rule base established on typical defects and repair features of historical buildings, the spatial correspondence and feature combination pattern between the visible light anomaly region and the infrared temperature anomaly region are analyzed. Based on the analysis results, regions that simultaneously meet the infrared temperature anomaly characteristics but do not meet the typical historical repair visual characteristics are marked as high-priority defect suspicion regions and reflected in the fusion guidance diagram.

[0008] According to the technical solution provided in this application, the analysis of the spatial correspondence and feature combination pattern between the visible light anomaly region and the infrared temperature anomaly region includes the following steps: The registered visible light anomaly region and the infrared temperature anomaly region are spatially superimposed, and the overlap area between any infrared anomaly region and all visible light anomaly regions is calculated. For any of the infrared temperature anomaly regions, if the ratio of the area of ​​overlap between the infrared temperature anomaly region and any of the visible light anomaly regions to the area of ​​the infrared temperature anomaly region is greater than the first dynamic threshold, it is preliminarily determined that there is a spatial correspondence between the two regions. For regions that are initially determined to have a spatial correspondence, calculate their combination pattern features, including: the angle between the direction of the maximum temperature gradient of the infrared anomaly region and the main texture direction of the visible light anomaly region, and the average nearest distance between the boundary of the infrared anomaly region and the edge of the visible light anomaly region. If the included angle is less than the angle threshold and the average nearest distance is less than the distance threshold, then the region is determined to constitute an infrared-visual high-confidence correlation pattern, and this is used as the basis for determining the priority defect suspected region.

[0009] According to the technical solution provided in this application, before performing adaptive fusion and feature extraction on the acoustic vibration response signal sequence and the infrared thermal imaging sequence, the following steps are also included: Based on the visible light image obtained from global pre-scanning, the texture complexity index of the surface of the historical building is calculated; Determine whether the texture complexity index is lower than a preset complexity threshold; The adaptive fusion and feature extraction of the acoustic vibration response signal sequence and the infrared thermal imaging sequence includes the following steps: If so, then adaptive fusion and feature extraction are performed on the acoustic vibration response signal sequence and the infrared thermal imaging sequence.

[0010] According to the technical solution provided in this application, after determining whether the texture complexity index is lower than a preset complexity threshold, the method further includes the following steps: If not, then based on the visible light image, identify decorative texture unit categories with different visual characteristics; For each type of decorative texture unit identified, based on its statistical performance in the infrared preview image, the texture thermal properties, including emissivity correction coefficient and base temperature offset, are estimated. The texture thermal properties are mapped onto a spatial grid with the same resolution as the infrared thermal imaging sequence to generate a texture thermal interference prediction map. The actual infrared thermal imaging sequence is compared with the texture thermal interference prediction map frame by frame to obtain the purified thermal imaging sequence. Adaptive fusion and feature extraction are performed on the acoustic vibration response signal sequence and the purified thermal image sequence.

[0011] According to the technical solution provided in this application, after performing frame-by-frame difference or ratio processing on the actually acquired infrared thermal imaging sequence and the expected texture thermal interference map to obtain the purified thermal image sequence, the following steps are also included: In the purified thermal imaging sequence, a flat surface area corresponding to a known material that is uniform and defect-free in the visible light image is selected as the verification reference area. The temperature standard deviation of the verification reference area in the purified thermal imaging sequence is calculated as the first residual noise index; and its temperature fluctuation range during the entire acquisition period of the infrared thermal imaging sequence is calculated as the second residual noise index. The adaptive fusion and feature extraction of the acoustic vibration response signal sequence and the purified thermal image sequence includes the following steps: If neither the first residual noise index nor the second residual noise index exceeds the corresponding preset noise tolerance threshold, then adaptive fusion and feature extraction are performed on the acoustic vibration response signal sequence and the purified thermal image sequence.

[0012] According to the technical solution provided in this application, after calculating the first residual noise index and the second residual noise index, the following steps are also included: If the first residual noise index or the second residual noise index exceeds the corresponding preset noise tolerance threshold, the purified thermal image sequence is subtracted from the texture thermal interference expected map frame by frame to obtain a serialized compensation residual map. In the compensated residual map, pixel regions where the absolute value of the residual is continuously higher than the residual threshold are identified and located as local areas of insufficient compensation. The texture thermal properties of the insufficiently compensated local areas are optimized in a targeted manner, and the expected texture thermal interference map and the purified thermal image sequence are updated based on the optimization results.

[0013] Compared with the prior art, the beneficial effects of this application are as follows: I. Deep fusion and synergistic enhancement of multi-physics information: By synchronously or correlatedly acquiring acoustic vibration and infrared thermal imaging sequences, and adaptively fusing them based on a professional database, the mechanical vibration and thermal conduction characteristics of defects can be comprehensively utilized. This coupled analysis achieves information complementarity and cross-validation, greatly improving the detection capability and identification confidence of hidden defects, and overcoming the limitations of single detection methods.

[0014] II. Significantly Improved Accuracy and Specificity of Defect Identification: Analysis based on fused feature maps can effectively suppress common interferences such as environmental fluctuations and complex surface textures. This method can not only locate defects more accurately, but its core advancement lies in its ability to effectively distinguish between deep hollowing and shallow peeling, two types of defects with different repair process requirements. It also reduces the misjudgment of non-defect areas such as historical repairs as abnormal, and the diagnostic conclusions directly support precise repair.

[0015] Third, a complete and intelligent non-destructive diagnostic process has been constructed: from intelligent guided scanning and multi-source data fusion to in-depth defect identification and visualization output, a closed-loop methodological system has been formed. This method fully realizes non-contact and non-destructive testing, meets the stringent requirements for the protection of historical buildings, and at the same time, through process-oriented and algorithmic approaches, reduces over-reliance on the personal experience of operators and improves the objectivity and repeatability of test results.

[0016] IV. The output results are intuitive and highly practical: The final generated visualized defect distribution information (such as thermal overlay maps and 3D depth maps) can intuitively show the location, range and severity of defects. It can be directly used as a scientific basis for formulating repair plans and calculating engineering quantities, realizing a close connection from detection to engineering implementation, and improving the scientific nature and efficiency of the overall protection work. Attached Figure Description

[0017] Figure 1 A flowchart illustrating the steps of the non-destructive testing method for concealed hollowing and peeling in historical buildings provided in this application. Detailed Implementation

[0018] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0019] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0020] Example 1 As mentioned in the background section, in view of the problems in the prior art, this application proposes a non-destructive testing method for concealed hollowing and peeling in historical buildings, such as... Figure 1 As shown, it includes the following steps: S1. Scan the surface of the historical building to be tested to collect its acoustic vibration response signal sequence and infrared thermal imaging sequence; S2. Based on the data information containing the acoustic and thermal response characteristics of various historical building materials and repair materials, the acoustic vibration response signal sequence and the infrared thermal imaging sequence are adaptively fused and feature extracted to suppress environmental and surface texture interference, and a fused feature map is output. S3. Based on the fused feature map, identify and distinguish the spatial location and distribution characteristics of deep hollow defects and shallow peeling defects under the surface of the historical building; S4. Based on the spatial location and distribution characteristics, generate and output visualized defect distribution information for guiding repair.

[0021] Specifically, the first step involves scanning the surface of the historical building to be tested to simultaneously or sequentially acquire acoustic vibration response signal sequences and infrared thermal imaging sequences. Integrated or separate detection equipment can be used. Acoustic signal acquisition is typically accomplished through excitation and receiving devices. Excitation methods can be contact-type lightweight hammers (with force sensors) or non-contact airborne sound sources (such as loudspeakers), aiming to generate a transient or swept-frequency signal containing multiple frequency components. The receiving device employs a high-sensitivity accelerometer or laser vibrometer to record the vibration response of the wall surface under excitation in a dot matrix or scanning manner, forming a sequence of acoustic signals that varies with time or spatial location. Acquiring the infrared thermal imaging sequence requires an infrared thermal imager. To generate thermal contrast, thermal excitation is usually applied to the wall, such as active heating using high-power halogen lamps or infrared heating lamp arrays, or passive detection utilizing the natural cooling process after solar radiation. The thermal imager continuously captures the temperature field changes on the wall surface at a certain frame rate, thereby obtaining a time-series set of infrared images.

[0022] The second step involves adaptive fusion and feature extraction of the two types of sequences collected, based on a pre-generated database containing the acoustic and thermal response characteristics of various historical building materials and repair materials. This database is the core of the method's prior knowledge; through laboratory testing and field calibration, it systematically records the acoustic spectral characteristics (such as resonant frequency and damping ratio) and thermal response characteristics (such as heating rate and thermal diffusivity) exhibited by common historical building materials (such as blue bricks, red bricks, various mortars, plasters, and wood from different eras) and their typical repair materials under standard excitation. The adaptive fusion and feature extraction process is implemented in a computer using algorithms. First, the acoustic signal sequences are preprocessed (e.g., filtering and noise reduction) and time-frequency features (such as wavelet transform coefficients and Mel frequency cepstral coefficients) are extracted; the infrared thermal imaging sequences are subjected to time-series analysis to extract the temperature rise curve features of each pixel or region. Subsequently, fusion algorithms (e.g., feature-level fusion networks based on deep learning, or data-level fusion based on Kalman filtering) correlate and integrate the acoustic and thermal feature vectors from the same spatial location. During this process, the algorithm dynamically adjusts the fusion weights based on database information and real-time environmental parameters (such as ambient temperature and humidity) to suppress interference caused by changes in sunlight, surface shadows, and complex textures (such as painted patterns). Finally, it outputs a digital matrix that comprehensively reflects the internal structural state, namely the fusion feature map.

[0023] The third step involves identifying and distinguishing between deep voids and shallow peeling defects based on the generated fused feature map. This requires establishing or training a classification model. This model takes the data from the fused feature map as input and classifies each detected area using built-in discrimination rules or learned decision boundaries. Because deep voids (typically referring to cavities behind the structural base layer) and shallow peeling (typically referring to debonding between the surface layer and the base layer) have physical differences in acoustic vibration modes and heat conduction paths, these differences are reflected in the fused features. For example, deep voids may result in more pronounced low-frequency vibration components and slower heat conduction, while shallow peeling may cause higher-frequency local vibrations and specific surface temperature distributions. The classification model learns these subtle differences to automatically identify and label the two types of defects and their spatial locations.

[0024] The fourth step involves generating and outputting visualized defect distribution information to guide repairs based on the identified spatial location and distribution characteristics of the defects. This is typically done in the software's post-processing module. The system overlays the classification results with digital images or 3D models of the wall, using different colors, outlines, or transparency to visually indicate the areas, extent, and severity levels of deep hollowing and shallow peeling. Ultimately, it can generate thematic drawings, 3D visualization reports, or augmented reality (AR) overlay views, providing quantitative and visual scientific evidence directly for cultural relic conservation engineers to develop repair plans such as chiseling, grouting, and anchoring.

[0025] The technical principle of this implementation lies in utilizing the differences in sensitivity to interface defects of different depths and types through acoustic excitation-vibration response and thermal excitation-infrared response, overcoming the limitations of single detection methods by fusing multi-source information. Acoustic methods are sensitive to interface debonding and voids, reflecting changes in the mechanical impedance of the structure; thermal methods are sensitive to differences in material thermophysical properties and changes in heat flow paths, reflecting internal heat transfer anomalies. Fusing these two methods at the data or feature level allows for information complementarity and cross-validation, essentially performing two different perspectives on the same object, significantly improving the signal-to-noise ratio. The resulting technical benefits are multifaceted. First, it achieves truly non-destructive, non-contact, or minimally destructive testing, fully meeting the highest requirements for the protection of historical buildings. Second, this method transforms experience-based qualitative judgment into data-driven quantitative analysis, enhancing the objectivity and repeatability of the detection. Most importantly, it is the first time that a single method has been used to effectively distinguish between deep hollowing and shallow peeling, which is difficult to achieve with traditional percussion methods or single infrared methods. Its diagnostic conclusions have direct guiding value for adopting differentiated restoration processes, effectively avoiding over-restoration or under-restoration, thereby significantly improving the scientific nature, accuracy and cost-effectiveness of historical building protection work.

[0026] In a preferred embodiment, identifying and distinguishing the spatial location and distribution characteristics of deep hollow defects and shallow peeling defects based on the fused feature map includes the following steps: Time-frequency analysis is performed on the acoustic vibration response signal sequence to extract the resonant main frequency and frequency band energy distribution of each measurement point in the preset frequency band, thus forming a power spectrum feature vector; The infrared thermal imaging sequence is subjected to time-series analysis to extract the curve of temperature change over time at each measuring point, and the time required to reach half-peak temperature rise is calculated to form the thermal conduction time constant feature. The power spectrum feature vector and the heat conduction time constant feature under the same spatial coordinates are coupled and input into the hierarchical classification model; The deep hierarchical classification model outputs a hierarchical probability prediction of the defect state at the measurement point based on the input coupling features. The defect state includes: no defect, shallow peeling, deep hollowing, and coexistence of shallow and deep defects.

[0027] Specifically, firstly, a detailed time-frequency analysis is performed on the acquired acoustic vibration response signal sequence. In practice, for each pre-defined or scanned measurement point, the recorded vibration time-domain signal (usually a signal containing excitation and decay processes) is converted to the frequency domain for analysis using Fast Fourier Transform, Wavelet Transform, or modern spectral estimation methods. The core is to extract two key features: first, the resonant dominant frequency, i.e., the frequency point where the signal energy is most concentrated, reflecting the equivalent stiffness and mass of the structure at that measurement point; second, the frequency band energy distribution within a pre-defined frequency band (e.g., set to 50-1000Hz based on common defects in historical buildings), for example, dividing the frequency band into several sub-bands and calculating the energy proportion of each sub-band. These frequency and energy information together constitute a multi-dimensional power spectrum feature vector, which quantitatively describes the spectral fingerprint of the vibration at that point.

[0028] Secondly, a time-series analysis is performed on the synchronously acquired infrared thermal imaging sequences. For the same spatial measurement point, the curve showing the average temperature change over time for each pixel or small area in the thermal image sequence is extracted, i.e., the temperature rise curve. A key parameter of this curve is calculated: the time required to reach the half-peak temperature rise. Here, the half-peak temperature rise refers to the time required for the temperature at that point to rise from its initial value to half of the maximum temperature rise after the excitation begins. This time constant is closely related to the thermal diffusivity of the material; a slower thermal diffusivity (e.g., an air layer hindering heat transfer) results in a longer time to reach the half-peak temperature rise. This time constant constitutes a scalar or low-dimensional thermal conduction time constant characteristic, reflecting the thermal resistance characteristics below that point.

[0029] Then, the acoustic power spectrum feature vector from the same spatial coordinates is coupled with the thermal time constant feature. In practice, this coupling can be a simple vector concatenation, where the thermal feature is added as a new dimension to the acoustic feature vector to form a longer joint feature vector; or it can be a more complex interactive operation, such as calculating the correlation index between acoustic band energy and thermal parameters. This coupled feature vector carries information on both mechanical vibration and heat conduction at that point.

[0030] Finally, the coupled feature vector is input into a pre-trained hierarchical classification model. This model is a machine learning model, such as a support vector machine, random forest, or deep neural network, whose training data comes from the acoustic-thermal coupling features corresponding to a large number of sample points with known defect states (confirmed through drilling or other reliable methods). The model's learning objective is to establish a mapping relationship from coupling features to defect state categories (no defect, shallow peeling, deep hollowing, coexistence). In application, the model receives the coupling features of new measurement points and, through complex internal nonlinear calculations, outputs a probability value belonging to each defect state category, thereby achieving hierarchical probability prediction. For example, the model might output "shallow peeling: 85%, deep hollowing: 10%, no defect: 5%", thus providing a quantitative judgment.

[0031] In a preferred embodiment, after coupling the power spectrum feature vector and the heat conduction time constant feature under the same spatial coordinates and inputting them into the hierarchical classification model, the following steps are further included: For target measurement points identified as having defects, perform defect interface location analysis, including: The target measuring point is sequentially subjected to a preheating excitation with a first power density and a main thermal excitation with a higher power density, and high frame rate infrared image sequences and acoustic vibration signals are acquired simultaneously during the two excitation phases. From the high frame rate infrared image sequence, calculate the additional temperature rise curve of the main thermal excitation stage relative to the preheating excitation stage, and extract the additional temperature rise time constant of the curve as the interface thermal resistance feature. The acoustic vibration signals of the main thermal excitation stage and the preheating excitation stage are differentially processed to obtain differential acoustic vibration signals. The energy ratio of the signals in the preset frequency band is extracted as the interface mechanical state characteristics. By combining the interface thermal resistance characteristics and interface mechanical state characteristics with the material layer information corresponding to the target measurement point in the fused feature map, the specific material interface where the defect is located is determined. The specific material interface includes: the interior of the finishing layer, the interface between the finishing layer and the mortar layer, and the interface between the mortar layer and the base layer.

[0032] Specifically, this step is implemented on target measurement points that are determined to be highly likely to have defects (the probability of a certain defect occurring reaches 80% or higher). First, a stepped thermal excitation and simultaneous data acquisition are performed on this point. Using a controllable heat source (such as a focused infrared heating lamp), preheating excitation is first performed for a period of time at a first power density. This power density is relatively low, designed so that the heat is mainly absorbed by the surface finish layer and generates a temperature rise, but not enough to significantly affect the deeper mortar or base layer. Simultaneously, a high-frame-rate infrared thermal imager records the fine temperature field change sequence during this preheating stage, and a high-sensitivity acoustic vibration sensor records the vibration signal. Then, the heat source power is immediately increased to a higher second power density for the main thermal excitation. The heat intensity at this stage is sufficient to penetrate the finish layer, allowing heat to be transferred to the mortar layer and even the base layer. Again, a high-frame-rate thermal image sequence and acoustic vibration signal are simultaneously acquired during this stage.

[0033] Secondly, interfacial thermal resistance characteristics are extracted from the thermal data. The acquired high-frame-rate infrared image sequences are processed, and temperature rise curves for the target measurement points are plotted during the preheating and main thermal excitation stages. Then, the additional temperature rise curve for the main excitation stage relative to the preheating stage is calculated, i.e., the temperature baseline established during the preheating stage is subtracted from the main excitation temperature rise curve. The shape of this additional temperature rise curve is analyzed, and the time required to reach a certain proportion (e.g., half-value) is extracted as the additional temperature rise time constant. This time constant mainly reflects the interfacial thermal resistance encountered when heat is transferred from the finishing layer to the lower layer material. If a defect exists between the finishing layer and the mortar layer, this interfacial thermal resistance will increase abnormally, leading to a significant increase in the additional temperature rise time constant.

[0034] Simultaneously, interfacial mechanical state characteristics are extracted from acoustic data. Differential processing is performed on the acoustic vibration signals acquired during the preheating and main thermal excitation stages; that is, the signal from the preheating stage is subtracted from the signal from the main excitation stage to obtain the differential acoustic vibration signal. The purpose of this operation is to eliminate or reduce the contribution of the surface finish layer's own material properties to the vibration signal, highlighting the changes in the underlying interfacial state caused by deep thermal excitation. Then, spectral analysis is performed on the differential acoustic vibration signal to calculate the ratio of its energy to the total energy within a preset frequency band (e.g., the frequency band corresponding to interfacial shear vibration). This energy ratio reflects the integrity of the interfacial adhesion; interfacial debonding will lead to a relative increase in vibration energy in that frequency band.

[0035] Finally, a comprehensive judgment is made. The calculated interface thermal resistance characteristics (with additional temperature rise time constant) and interface mechanical state characteristics (energy ratio of a specific frequency band of the differential signal) are combined with the material layer information of the measurement point inferred from the previously fused feature map (e.g., through visible light image analysis and database comparison, it is known that this is a three-layer structure of brick base layer-mortar layer-painted finish). This information is then input into a logic judge or a small classifier. Based on the correspondence between these feature combinations and different interface defect types (e.g., defects within the finish layer may have small changes in thermal resistance but changes in acoustic energy ratio; voids at the mortar layer-base layer interface may result in significant changes in both thermal resistance and acoustic energy ratio), the specific interface where the defect is located is determined.

[0036] This implementation method elevates the precision of detection from depth ranges to specific material interfaces. This has crucial guiding significance for the restoration of historical buildings. For example, if the defect is determined to be at the interface between the finishing layer and the mortar layer, only local grouting and bonding may be necessary; while if it is determined to be at the interface between the mortar layer and the base layer, pressure grouting or structural reinforcement may be required. This precise positioning helps cultural heritage preservationists develop optimized, minimally destructive targeted restoration plans, avoiding over- or under-intervention due to interface misjudgment. It ensures the effectiveness and durability of restoration while maximizing the preservation of historical information, representing a new height in the application of non-destructive testing technology to refined conservation engineering.

[0037] In a preferred embodiment, acquiring its acoustic vibration response signal sequence and infrared thermal imaging sequence includes the following steps: A global pre-scan of the surface of the historical building was performed to obtain visible light images and infrared preview images; Based on the infrared preview image, the initial screening area for temperature anomalies is obtained; Based on the visible light image, regions of abrupt changes in surface material, regions of historical repair marks, and regions of decorative textures are identified and marked; and combined with the initial screening region of temperature anomalies, a fusion guide map is generated to mark key feature regions. Based on the fusion guidance map, the acquisition parameters are planned, including the scanning path and the spatial density of acquisition points; Based on the acquisition parameters, the acoustic vibration response signal sequence and infrared thermal imaging sequence are acquired.

[0038] Specifically, the first step involves a rapid, low-resolution global pre-scan of the target historical building's surface. During this process, operators or automated mobile platforms equipped with visible light cameras and infrared thermal imagers quickly perform a comprehensive, traversal scan of the wall surface. The visible light camera captures high-resolution color or grayscale images to record visual information about the wall surface, such as color, texture, patterns, cracks, and stains. The infrared thermal imager simultaneously captures one or more infrared preview images at a lower frame rate or resolution, reflecting the approximate temperature distribution of the wall. This pre-scan is time-efficient and aims to gather preliminary global information for subsequent analysis.

[0039] The second step involves automatically identifying initial screening areas for temperature anomalies based on the acquired infrared preview images using image processing algorithms. In practice, threshold segmentation can be performed on the infrared images (e.g., marking areas where the temperature is higher or lower than the overall average temperature by a certain threshold), or more complex background temperature field fitting and residual analysis can be conducted to quickly delineate suspected areas exhibiting obvious temperature anomalies (overheating or undercooling). These areas provide initial clues to potential defects such as hollowing, peeling, or dampness.

[0040] The third step involves using computer vision algorithms to identify and label various key surface feature regions based on the acquired visible light images. These include: regions of abrupt material changes (such as the joints between brick walls and stone carvings, and the boundaries between different colors of mortar), which are prone to generating interference signals in thermal images due to differences in the thermal properties of the materials; regions with historical repair marks (such as later-filled mortar blocks, newly replaced bricks and stones), which are not defects in themselves, but whose acoustic and thermal characteristics may differ from the original structure; and regions with decorative textures (such as complex painted patterns, and the concave and convex surfaces of reliefs), whose geometry can significantly affect the uniform distribution of heat flow and the reflection and scattering of sound waves. Through image segmentation, edge detection, texture analysis, and comparison with pre-stored pattern templates, the identification and labeling of these regions can be completed automatically or semi-automatically.

[0041] The fourth step involves fusing information from the initial screening areas of temperature anomalies in the infrared preview image and various surface feature areas from the visible light image to generate a fusion guidance map. During implementation, different colors, symbols, or layers can be overlaid on the same base map to mark these various areas. For example, red highlights can indicate high-temperature anomaly areas, blue boxes can indicate historical repair areas, and green grids can indicate complex painted areas. This map integrates thermal anomaly clues and visual interference source information, intuitively showing where the key suspicious points are and where the interference sources are likely to cause misjudgments.

[0042] The fifth step involves intelligently planning the acquisition parameters for subsequent formal scanning based on the generated fusion guidance map. The most important parameters are the spatial density of the scanning path and acquisition points. For high-priority suspected areas marked on the map (such as areas with both temperature anomalies and no obvious repairs), the planning system instructs the detection device to use a denser arrangement of measurement points, a slower scanning speed, or a longer signal acquisition time in these areas to ensure high-quality data acquisition. For large, homogeneous areas without anomalies, sparser measurement points can be used to improve efficiency. For known repaired areas, specific acquisition patterns can be planned to obtain their characteristic data for subsequent differentiation. The path planning algorithm optimizes the movement trajectory to ensure efficient coverage of all key areas.

[0043] Finally, based on the planned acquisition parameters, the detection device is controlled to perform formal and refined data acquisition, thereby obtaining the final acoustic vibration response signal sequence and infrared thermal imaging sequence.

[0044] In a preferred embodiment, generating the fusion guide map of the marked key feature regions includes the following steps: The visible light image and the infrared preview image are registered and feature analyzed to identify abnormal regions in visible light and abnormal regions in infrared temperature. Based on the feature coupling rule base established on typical defects and repair features of historical buildings, the spatial correspondence and feature combination pattern between the visible light anomaly region and the infrared temperature anomaly region are analyzed. Based on the analysis results, regions that simultaneously meet the infrared temperature anomaly characteristics but do not meet the typical historical repair visual characteristics are marked as high-priority defect suspicion regions and reflected in the fusion guidance diagram.

[0045] Specifically, the first step is to perform precise registration and feature analysis on the acquired visible light image and infrared preview image. Image registration is the foundation for subsequent spatial correlation analysis, and a feature-based registration method can be used. First, stable feature points are automatically detected and extracted in both images, for example, using SIFT or ORB algorithms to extract feature points at texture corners in the visible light image and significant temperature gradient edges in the infrared image. Then, through feature point matching and spatial transformation model calculation, the infrared preview image is accurately mapped to the coordinate system of the visible light image, ensuring pixel alignment at the same physical location in both images. After registration is completed, feature analysis is performed in parallel. On the visible light image, image segmentation algorithms and edge detection are used to identify visible light anomalous regions. Here, anomalous regions do not refer to defects, but rather to areas with abrupt changes in visual features compared to the surrounding homogeneous background. These mainly include: material boundaries where color, brightness, or texture changes abruptly; historical patch blocks with regular geometric shapes, clear edges, and material textures that are significantly different from the surrounding substrate; and decorative texture patterns formed by complex painting or engraving. On the registered infrared preview image, infrared temperature anomaly regions are identified by temperature threshold segmentation or cluster analysis, which are continuous sets of pixels whose surface temperature is significantly higher or lower than the average temperature of the surrounding area.

[0046] The second step is to analyze the spatial relationship between the two types of anomalous regions based on a pre-built feature coupling rule base. This rule base is a digital representation of domain knowledge, based on numerous historical building cases, summarizing the performance patterns of different types of targets in visible and infrared images. For example, one rule might describe: a real, active hollow defect typically appears as a relatively blurred temperature difference anomalous area in an infrared image, while in the corresponding high-definition visible light image, this area often lacks regular geometric boundaries, obvious color differences between old and new materials, and complete coverage of decorative patterns. Another rule might describe: a stable, dense historical cement repair block may appear as a clearly defined temperature difference anomalous area in an infrared image due to differences in material thermal capacity, but in a visible light image, it will definitely present as a regular geometric shape (such as a rectangle or square) and a color and texture different from the original wall surface.

[0047] The third step involves decision-making and labeling based on the above analysis results. If an infrared temperature anomaly region meets two conditions simultaneously: first, it exhibits obvious temperature difference anomaly characteristics; second, the visual characteristics of its corresponding visible light region do not conform to the characteristic patterns of typical historical repairs or known harmless material mutations defined in the rule base, then this region is identified as a high-priority defect suspicion area. This means that the heating or cooling phenomenon in this region cannot be explained by surface-visible, known harmless factors, thus greatly increasing the possibility of hidden defects inside. Ultimately, all such regions are marked on the fusion guidance map in a conspicuous manner (e.g., flashing red outline, highlighted fill), while historical repair areas may be marked as reference information (e.g., blue semi-transparent cover). This map thus achieves the function of extracting high-value suspicious points from massive image data.

[0048] Furthermore, the analysis of the spatial correspondence and feature combination patterns between the visible light anomaly region and the infrared temperature anomaly region includes the following steps: The registered visible light anomaly region and the infrared temperature anomaly region are spatially superimposed, and the overlap area between any infrared anomaly region and all visible light anomaly regions is calculated. For any of the infrared temperature anomaly regions, if the ratio of the area of ​​overlap between the infrared temperature anomaly region and any of the visible light anomaly regions to the area of ​​the infrared temperature anomaly region is greater than the first dynamic threshold, it is preliminarily determined that there is a spatial correspondence between the two regions. For regions that are initially determined to have a spatial correspondence, calculate their combination pattern features, including: the angle between the direction of the maximum temperature gradient of the infrared anomaly region and the main texture direction of the visible light anomaly region, and the average nearest distance between the boundary of the infrared anomaly region and the edge of the visible light anomaly region. If the included angle is less than the angle threshold and the average nearest distance is less than the distance threshold, then the region is determined to constitute an infrared-visual high-confidence correlation pattern, and this is used as the basis for determining the priority defect suspected region.

[0049] Specifically, firstly, basic spatial overlay and area calculation are performed. In the digital coordinate system after image registration, the algorithm performs a Boolean AND operation on each infrared temperature anomaly region defined by a set of pixels and all visible light anomaly regions to calculate the pixel area of ​​their overlapping portions. For each infrared anomaly region Ri, the algorithm iterates through all visible light anomaly regions Vj, recording the overlap area A_overlap(i,j) with each Vj.

[0050] Secondly, a preliminary spatial correspondence determination based on a dynamic threshold is performed. For an infrared region Ri, the percentage of its overlapping area with each visible light region Vj is calculated, i.e., P(i,j) = A_overlap(i,j) / Area(Ri). Here, the first dynamic threshold is not a fixed value; its setting considers the size of the infrared region itself and the signal-to-noise ratio. For example, for a small infrared region with potentially blurred boundaries, it needs to be highly correlated with a visible light region to be considered associated, so the threshold may be set relatively high (e.g., 70%); for a large infrared region with clear boundaries, the threshold can be appropriately reduced (e.g., 50%). If there exists a Vj such that P(i,j) is greater than the currently set dynamic threshold, then a preliminary spatial correspondence is determined between the infrared region Ri and the visible light region Vj, and they are recorded as a region pair (Ri, Vj) to be further investigated.

[0051] Then, for each initially identified region pair, in-depth combined pattern feature calculations are performed. The first feature is the directional angle. For the infrared region Ri, by calculating the two-dimensional gradient of its internal temperature field, the direction with the largest gradient magnitude is found. This direction is usually perpendicular to the isotherms and may indicate the orientation of the internal heat source (defect) or the main path of heat flow. For the visible light region Vj, by analyzing the grayscale changes of its internal pixels, such as using Gabor filter banks or principal component analysis, the main direction of its texture is extracted, i.e., the direction of the most significant extension of the texture pattern. The angle θ between these two direction vectors is calculated. The second feature is the boundary distance. The precise contour boundary pixel sets of Ri and Vj are extracted separately. The shortest Euclidean distance from each pixel on the boundary of Ri to the boundary of Vj is calculated, and then the average of this shortest distance for all pixels is obtained to obtain the average nearest distance d. This value reflects the degree of fit between the two regions in shape and position.

[0052] Finally, a high-confidence correlation pattern is determined. An angle threshold (e.g., 30 degrees) and a distance threshold (e.g., 5 pixels) are preset. For a region pair (Ri, Vj), it is only considered to constitute an infrared-visual high-confidence correlation pattern if its calculated angle θ is less than the angle threshold and its average nearest distance d is less than the distance threshold. This means that not only do the two regions largely overlap in area, but the dominant direction of thermal diffusion exhibited by the infrared anomaly is roughly aligned with the dominant direction of the visible surface texture, and their boundary contours are also very close in space. Only region pairs that satisfy such strict geometric constraints are considered to have a highly reliable correspondence, and the existence of this correlation pattern is used as the core criterion for marking Ri as a high-priority defect suspicion region.

[0053] The technical principle of this embodiment is that a thermal anomaly zone caused by internal hollowing may have its shape and heat flow direction controlled by the geometry of internal defects. If this defect happens to be located under a surface decorative texture, or is caused by unevenness in the surface material, then the direction of the surface texture and the morphology of the thermal anomaly zone may be intrinsically related (e.g., peeling along brick joints). Simultaneously, anomalies of different modes caused by the same physical entity should have highly overlapping spatial contours. By quantitatively calculating directional consistency and boundary fit, it is possible to very strictly distinguish which are true, physically related multimodal anomalies and which are merely accidental spatial overlaps. The introduction of a dynamic threshold allows the algorithm to adapt to anomaly regions of different scales and qualities, avoiding the problem of large areas being misjudged due to slight overlap or small areas being missed due to registration errors. The dual constraints of direction and distance constitute a very strong filter, effectively eliminating a large number of false associations that overlap in area but are unrelated in shape. For example, even if a circular infrared hotspot and a long strip of decorative ribbon partially overlap, they will be excluded due to the large difference in direction and the large boundary distance. This makes the high-confidence associated regions that are ultimately selected highly defect-oriented, compressing the scope of the investigation to the maximum extent, and allowing subsequent fine-grained inspections to focus on the most suspicious and likely to reveal the real problems, thus achieving the optimal balance between detection accuracy and efficiency as a whole.

[0054] In a preferred embodiment, before performing adaptive fusion and feature extraction on the acoustic vibration response signal sequence and the infrared thermal imaging sequence, the method further includes the following steps: Based on the visible light image obtained from global pre-scanning, the texture complexity index of the surface of the historical building is calculated; Determine whether the texture complexity index is lower than a preset complexity threshold; The adaptive fusion and feature extraction of the acoustic vibration response signal sequence and the infrared thermal imaging sequence includes the following steps: If so, then adaptive fusion and feature extraction are performed on the acoustic vibration response signal sequence and the infrared thermal imaging sequence.

[0055] Specifically, an upfront intelligent decision-making branch is introduced before the core data fusion process. Its core idea is to select different data processing strategies based on the visual complexity of the surface of the object being detected, thereby optimizing resources and improving process efficiency.

[0056] The first step is to quantitatively calculate the texture complexity index of the wall surface based on high-resolution visible light images acquired through global pre-scanning. Texture complexity here is a mathematical measure used to quantify the richness of surface patterns, details, and irregularities. Several mature image texture analysis methods are available for implementation. A common method is to use the gray-level co-occurrence matrix (GLCM), which calculates the joint probability statistics of gray levels for pixel pairs in a specific direction and distance, extracting features such as contrast, correlation, energy, and homogeneity, and synthesizing a comprehensive complexity score. Another more intuitive method is to calculate the edge density of the image. This involves first detecting all edges in the image using operators such as Canny or Sobel, and then counting the number of edge pixels per unit area; the denser and more complex the edges, the higher the value. The information entropy of the image can also be calculated; a higher entropy value indicates a more random distribution of pixel gray values, a greater amount of information, and a more complex texture. In practical applications, one or more features can be selected and combined, and through weighted or principal component analysis, a scalar index between 0 and 1 or within a certain calibrated range can be formed to represent the texture complexity of the entire image or a specific region within the image.

[0057] The second step is to compare the calculated texture complexity index with a preset complexity threshold. This threshold needs to be determined through experiments and experience. For example, on a large number of historical building samples, the image complexity values ​​of those with simple textures such as plain gray walls and flat brick walls are statistically analyzed, as well as the image complexity values ​​of those with complex textures such as intricate paintings, complex reliefs, and haphazard stone masonry. A boundary value that can better distinguish between these two types of cases is selected as the threshold. The judgment logic is a simple two-branch approach: if the calculated index is lower than the threshold, the building surface is determined to be a simple texture or a homogeneous surface.

[0058] The third step is the fusion process following the branch decision. When the determination is yes, meaning the surface texture is simple, the system will employ a standard or basic data fusion and feature extraction process. This means that the acoustic vibration response signal sequence and infrared thermal imaging sequence collected from this surface will be directly processed using an adaptive fusion algorithm. In this case, because the surface visual texture is simple and uniform, its interference with the infrared thermal image is minimal, and its impact on sound wave scattering is also weak. Therefore, the standard algorithm is sufficient to effectively suppress environmental interference and extract features related to internal defects, without needing to activate an additional, computationally expensive dedicated processing module. This direct processing method is the most efficient.

[0059] This implementation significantly improves the overall efficiency and practicality of the detection method when dealing with diverse detection objects. It achieves intelligent allocation of computing resources. For a significant proportion of historical buildings with simple textures (such as the plain walls of many residences, warehouses, and temples), the system automatically selects the fast track, shortening the overall data processing time and enabling faster preliminary results from on-site inspections, thus improving operational efficiency. It enhances the applicability and economy of the method. This makes the detection system not only suitable for top-tier historical buildings with ornate decorations, but also highly effective for a large number of general historical buildings with simple textures that also require protection, broadening the application scenarios of the technology.

[0060] In a preferred embodiment, after determining whether the texture complexity index is lower than a preset complexity threshold, the method further includes the following steps: If not, then based on the visible light image, identify decorative texture unit categories with different visual characteristics; For each type of decorative texture unit identified, based on its statistical performance in the infrared preview image, the texture thermal properties, including emissivity correction coefficient and base temperature offset, are estimated. The texture thermal properties are mapped onto a spatial grid with the same resolution as the infrared thermal imaging sequence to generate a texture thermal interference prediction map. The actual infrared thermal imaging sequence is compared with the texture thermal interference prediction map frame by frame to obtain the purified thermal imaging sequence. Adaptive fusion and feature extraction are performed on the acoustic vibration response signal sequence and the purified thermal image sequence.

[0061] Specifically, when the system determines that the texture complexity index is not lower than a preset threshold, i.e., the surface has a complex decorative texture, this compensation process is initiated. The first step is to perform a detailed deconstruction of the surface decoration based on the visible light image. Using image segmentation algorithms, the complex decorative pattern is decomposed into different decorative texture unit categories. For example, for a painted painting, the algorithm might segment it into red oil paint areas, gold leaf areas, dark outline areas, and exposed plaster base areas. Segmentation can be based on features such as color, texture, and edges, using K-means clustering, watershed algorithms, or semantic segmentation models based on deep learning. The goal is to identify consecutive pixel regions with similar visual features (mainly color and surface microstructure) and assign a unique label to each such region.

[0062] The second step is to predict the key thermal properties of each identified texture unit. This requires combining information from the infrared preview image. Under the premise of image registration, for each texture unit category, all pixel locations belonging to that category are found in the infrared preview image, and the temperature values ​​of these pixels are extracted and statistically analyzed. Two main parameters are predicted: First, the emissivity correction coefficient. Surfaces of different colors and materials have different infrared emissivity; even if the actual temperature is the same, the brightness (apparent temperature) displayed on the thermal imager will differ. By comparing the statistical differences in the apparent temperature of different texture unit areas in the infrared preview image under the same environment (assuming they have the same internal structure and similar actual temperatures), the emissivity ratio between each category can be relatively estimated and normalized into a correction coefficient. Second, the baseline temperature offset. Due to the different absorption rates and heat capacities of different colors and materials, their equilibrium temperatures inherently differ even under the same environmental conditions. By analyzing the offset of the average temperature of each type of texture unit in the infrared preview image relative to a certain reference area (such as a large area of ​​plaster), this inherent temperature baseline difference can be estimated. These two parameters together constitute the thermal properties of this type of texture unit, which are used to predict its performance in thermal imaging.

[0063] The third step is to generate a texture thermal interference prediction map. Based on the spatial grid and time series planned during the actual scan, each pixel is assigned a previously estimated emissivity correction factor and a base temperature offset according to its texture unit category, thereby constructing a simulated image that is completely consistent with the actual acquired infrared thermal imaging sequence in terms of spatiotemporal resolution. This prediction map simulates the appearance of the infrared image that should be presented under ideal conditions without internal defects, solely due to differences in surface decorative textures. It is essentially a background map driven by texture categories, containing systematic differences in emissivity and temperature baselines.

[0064] The fourth step is thermal feature purification. The actual infrared thermal imaging sequence, which includes texture interference and real defect signals, is compared with the generated texture thermal interference prediction map using pixel-by-pixel and frame-by-frame calculations. These calculations can be differential processing: subtracting the prediction value from the actual value to directly deduct systematic deviations caused by the texture; or ratio processing: dividing the actual value by the prediction value, etc., to correct for the influence of emissivity. After this step, in the purified thermal image sequence, fixed pattern interference caused by the surface decorative texture itself is greatly suppressed, while dynamic or abnormal signals caused by internal defects that do not conform to the texture prediction are relatively highlighted.

[0065] Finally, the purified thermal image sequence and the acoustic vibration response signal sequence are fed into a standard adaptive fusion and feature extraction module for further analysis.

[0066] This implementation solves the problem of complex decorative textures obscuring infrared thermal images, allowing weak thermal signals emitted by internal defects hidden by exquisite paintings or reliefs to be revealed, greatly improving the detection capability of infrared technology in such scenarios. By pre-subtracting systematic texture interference, the difficulty of subsequent fusion algorithms in distinguishing between real and false anomalies is reduced, improving the overall system's accuracy and reliability in identifying defects, especially shallow and small-sized defects.

[0067] In a preferred embodiment, after performing frame-by-frame difference or ratio processing on the actually acquired infrared thermal imaging sequence and the texture thermal interference expected map to obtain the purified thermal image sequence, the method further includes the following steps: In the purified thermal imaging sequence, a flat surface area corresponding to a known material that is uniform and defect-free in the visible light image is selected as the verification reference area. The temperature standard deviation of the verification reference area in the purified thermal imaging sequence is calculated as the first residual noise index; and its temperature fluctuation range during the entire acquisition period of the infrared thermal imaging sequence is calculated as the second residual noise index. The adaptive fusion and feature extraction of the acoustic vibration response signal sequence and the purified thermal image sequence includes the following steps: If neither the first residual noise index nor the second residual noise index exceeds the corresponding preset noise tolerance threshold, then adaptive fusion and feature extraction are performed on the acoustic vibration response signal sequence and the purified thermal image sequence.

[0068] Specifically, the first step is to select a verification benchmark area. This step relies on a thorough understanding or simultaneous analysis of the building surface prior to the initial assessment. Ideally, operators or algorithms need to pre-identify one or more areas in the global visible light image that meet the following stringent conditions: First, the area appears to have a uniform and singular material under visible light, such as a large area of ​​undecorated plain gray wall or a flat brick surface of consistent material, without complex paintings, carvings, or obvious stains. Second, based on existing building survey reports or through pre-scan guided analysis, there is a high degree of certainty that the internal structure of the area is intact and free from defects such as hollowness or peeling. Finally, the surface of the area is physically smooth, without significant unevenness. In the data processing system, the spatial coordinate range of this area is recorded and labeled; it does not participate in defect judgment but is specifically used as an internal standard for evaluating system performance. In subsequent purified thermal image sequences, the algorithm will extract the temperature data time series corresponding to these spatial locations based on these coordinates for analysis.

[0069] The second step involves calculating two key residual noise metrics. The first is the temperature standard deviation. For the verification baseline area, the standard deviation of the temperature values ​​of all pixels within that area is calculated in each frame of the purified thermal imaging sequence. This standard deviation measures the residual temperature inhomogeneity within this theoretically homogeneous area after texture compensation at any given moment. Ideally, perfect compensation should result in a completely uniform temperature across a single frame, with a standard deviation close to zero. A larger standard deviation indicates insufficient compensation, and localized temperature differences caused by surface textures still exist. Typically, the average or median of the standard deviations across all frames in the entire sequence is calculated as the first residual noise metric. The second metric is the temperature fluctuation range. The temperature curve of a representative point (e.g., the center point) within the verification baseline area is extracted over the entire infrared acquisition period (potentially several minutes to tens of minutes), and the difference between its highest and lowest temperatures is calculated. This fluctuation range primarily reflects whether, after compensation, the area still exhibits a slow-changing, systematic temperature drift over time. This drift may stem from inaccurate estimation of the base temperature offset of the texture unit, failing to fully compensate for the differences in the dynamic response of different materials to changes in ambient temperature. This difference is the second residual noise indicator.

[0070] The third step is decision-making based on quantitative indicators. The system presets two noise tolerance thresholds, which are upper limits of allowable error determined based on extensive experimental and engineering experience. For example, the first threshold (corresponding to the standard deviation) might be set at 0.1°C, and the second threshold (corresponding to the fluctuation range) might be set at 0.5°C. The decision logic is: only when the calculated first and second residual noise indicators are simultaneously lower than their respective tolerance thresholds does the system determine that the current texture thermal compensation is sufficient and successful. At this point, the quality of the purified thermal image sequence is considered acceptable, and it can safely enter the subsequent core process, namely, adaptive fusion and feature extraction with the acoustic vibration response signal sequence for defect identification. This ensures strict requirements for data quality.

[0071] In a preferred embodiment, after calculating the first residual noise index and the second residual noise index, the method further includes the following steps: If the first residual noise index or the second residual noise index exceeds the corresponding preset noise tolerance threshold, the purified thermal image sequence is subtracted from the texture thermal interference expected map frame by frame to obtain a serialized compensation residual map. In the compensated residual map, pixel regions where the absolute value of the residual is continuously higher than the residual threshold are identified and located as local areas of insufficient compensation. The texture thermal properties of the insufficiently compensated local areas are optimized in a targeted manner, and the expected texture thermal interference map and the purified thermal image sequence are updated based on the optimization results.

[0072] Specifically, the first step is to automatically initiate a diagnostic procedure when the system detects any noise index exceeding the limit. First, it performs a reverse calculation to generate a compensated residual map. Specifically, it subtracts the previously obtained purified thermal image sequence from the texture thermal interference prediction map used to generate it, frame by frame. The objects of this operation are the data that has already undergone one round of compensation and purification, and the original prediction map. The result of the subtraction generates a completely new, serialized compensated residual map. In this residual map, the value of each pixel and each frame represents the residual error after subtracting the model's predicted value from the actual observed value. Theoretically, if the compensation is perfect, the residual map should be close to a zero-value noise field. However, in areas of insufficient compensation, the model underestimates or overestimates the actual thermal performance, and these areas will exhibit persistent and significant positive or negative residual values.

[0073] The second step is to automatically locate problem areas from the residual map. The system sets a residual threshold. The algorithm analyzes the compensated residual map frame by frame, identifying pixel clusters where the absolute value of the residual (whether positive or negative) consistently exceeds the threshold. "Consistent" is defined by whether the residual at that pixel location exceeds the threshold across multiple frames in the sequence (e.g., more than 70% of the frames), thus excluding transient noise interference. These identified connected pixel regions are marked as locally undercompensated areas. They clearly indicate which specific parts of the wall surface show significant discrepancies between the previous texture thermal property prediction model and the actual infrared thermal behavior.

[0074] The third step involves targeted optimization of the texture thermal properties of the identified problem areas. This optimization is not a global recalculation, but rather a precise, targeted adjustment. During implementation, the system first maps each locally undercompensated area back to the original visible light image and high-resolution infrared preview image to trace the source of the problem. It analyzes which texture unit the area belongs to under visible light (e.g., a deep red painted area) and retrieves the globally estimated properties (emissivity correction coefficient and base temperature offset) for that texture unit. Then, the algorithm analyzes the infrared statistical performance of the problem area itself: in the infrared preview image, it calculates the average temperature, temperature distribution variance, etc., of the area and compares them with the statistical values ​​of other normal areas of the same texture unit (i.e., areas not marked as undercompensated). Based on the pattern of differences, different optimization strategies are implemented: if the difference is mainly manifested as an overall shift in temperature level (different mean values ​​but similar distribution shapes), then only the base temperature shift of this type of texture unit may be finely adjusted; if the infrared statistical characteristics of this area are found to be very different from other areas of the same type, then the system can determine that the surface material at this location has special characteristics (such as differences in pigment composition, different aging levels, or slight contamination), thereby separating it from the original texture unit category, treating it as a new, independent subcategory, and re-predicting a set of exclusive texture thermal properties based on its own data.

[0075] The fourth step is to update the system based on the optimization results. According to the optimization strategy described above, the system generates an updated texture thermal property parameter table. Using the new parameters, an improved texture thermal interference prediction map is regenerated. Subsequently, this new prediction map is used to re-perform the compensation and purification operation on the original acquired infrared thermal imaging sequence, resulting in a new, higher-quality purified thermal imaging sequence. Theoretically, after verification, the residual noise index of this new sequence should be reduced. The system can be programmed with a set number of iterations to repeatedly execute the verification-diagnosis-optimization-update process until the quality meets the standard or the maximum number of iterations is reached.

[0076] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. The above descriptions are only preferred embodiments of this application. It should be noted that due to the limitations of written expression, while there are objectively infinite specific structures, those skilled in the art can make several improvements, modifications, or changes without departing from the principles of this invention, and can also combine the above technical features in an appropriate manner. These improvements, modifications, changes, or combinations, or the direct application of the inventive concept and technical solution to other situations without modification, should all be considered within the scope of protection of this application.

Claims

1. A non-destructive testing method for concealed hollowing and peeling in historical buildings, characterized in that, Includes the following steps: The surface of the historical building to be tested is scanned to collect its acoustic vibration response signal sequence and infrared thermal imaging sequence; Based on data containing acoustic and thermal response characteristics of various historical building materials and repair materials, the acoustic vibration response signal sequence and the infrared thermal imaging sequence are adaptively fused and feature extracted to suppress environmental and surface texture interference and output a fused feature map. Based on the fused feature map, identify and distinguish the spatial location and distribution characteristics of deep hollow defects and shallow peeling defects under the surface of the historical building; Based on the spatial location and distribution characteristics, visualized defect distribution information is generated and output to guide repair.

2. The method for non-destructive testing of concealed hollowness and peeling in historical buildings according to claim 1, characterized in that: The step of identifying and distinguishing the spatial location and distribution characteristics of deep hollow defects and shallow peeling defects based on the fused feature map includes the following steps: Time-frequency analysis is performed on the acoustic vibration response signal sequence to extract the resonant main frequency and frequency band energy distribution of each measurement point in the preset frequency band, thus forming a power spectrum feature vector; The infrared thermal imaging sequence is subjected to time-series analysis to extract the curve of temperature change over time at each measuring point, and the time required to reach half-peak temperature rise is calculated to form the thermal conduction time constant feature. The power spectrum feature vector and the heat conduction time constant feature under the same spatial coordinates are coupled and input into the hierarchical classification model; The deep hierarchical classification model outputs a hierarchical probability prediction of the defect state at the measurement point based on the input coupling features. The defect state includes: no defect, shallow peeling, deep hollowing, and coexistence of shallow and deep defects.

3. The method for non-destructive testing of concealed hollowness and peeling in historical buildings according to claim 2, characterized in that: After coupling the power spectrum feature vector and the heat conduction time constant feature under the same spatial coordinates and inputting them into the hierarchical classification model, the following steps are also included: For target measurement points identified as having defects, perform defect interface location analysis, including: The target measuring point is sequentially subjected to a preheating excitation with a first power density and a main thermal excitation with a higher power density, and high frame rate infrared image sequences and acoustic vibration signals are acquired simultaneously during the two excitation phases. From the high frame rate infrared image sequence, calculate the additional temperature rise curve of the main thermal excitation stage relative to the preheating excitation stage, and extract the additional temperature rise time constant of the curve as the interface thermal resistance feature. The acoustic vibration signals of the main thermal excitation stage and the preheating excitation stage are differentially processed to obtain differential acoustic vibration signals. The energy ratio of the signals in the preset frequency band is extracted as the interface mechanical state characteristics. By combining the interface thermal resistance characteristics and interface mechanical state characteristics with the material layer information corresponding to the target measurement point in the fused feature map, the specific material interface where the defect is located is determined. The specific material interface includes: the interior of the finishing layer, the interface between the finishing layer and the mortar layer, and the interface between the mortar layer and the base layer.

4. The method for non-destructive testing of concealed hollowness and peeling in historical buildings according to claim 1, characterized in that: The acquisition of its acoustic vibration response signal sequence and infrared thermal imaging sequence includes the following steps: A global pre-scan of the surface of the historical building was performed to obtain visible light images and infrared preview images; Based on the infrared preview image, the initial screening area for temperature anomalies is obtained; Based on the visible light image, regions of abrupt changes in surface material, regions of historical repair marks, and regions of decorative textures are identified and marked; and combined with the initial screening region of temperature anomalies, a fusion guide map is generated to mark key feature regions. Based on the fusion guidance map, the acquisition parameters are planned, including the scanning path and the spatial density of acquisition points; Based on the acquisition parameters, the acoustic vibration response signal sequence and infrared thermal imaging sequence are acquired.

5. The method for non-destructive testing of concealed hollowness and peeling in historical buildings according to claim 4, characterized in that: The process of generating a fusion guide map of the marked key feature regions includes the following steps: The visible light image and the infrared preview image are registered and feature analyzed to identify abnormal regions in visible light and abnormal regions in infrared temperature. Based on the feature coupling rule base established on typical defects and repair features of historical buildings, the spatial correspondence and feature combination pattern between the visible light anomaly region and the infrared temperature anomaly region are analyzed. Based on the analysis results, regions that simultaneously meet the infrared temperature anomaly characteristics but do not meet the typical historical repair visual characteristics are marked as high-priority defect suspicion regions and reflected in the fusion guidance diagram.

6. The method for non-destructive testing of concealed hollowness and peeling in historical buildings according to claim 5, characterized in that: The analysis of the spatial correspondence and feature combination patterns between the visible light anomaly region and the infrared temperature anomaly region includes the following steps: The registered visible light anomaly region and the infrared temperature anomaly region are spatially superimposed, and the overlap area between any infrared anomaly region and all visible light anomaly regions is calculated. For any of the infrared temperature anomaly regions, if the ratio of the area of ​​overlap between the infrared temperature anomaly region and any of the visible light anomaly regions to the area of ​​the infrared temperature anomaly region is greater than the first dynamic threshold, it is preliminarily determined that there is a spatial correspondence between the two regions. For regions that are initially determined to have a spatial correspondence, calculate their combination pattern features, including: the angle between the direction of the maximum temperature gradient of the infrared anomaly region and the main texture direction of the visible light anomaly region, and the average nearest distance between the boundary of the infrared anomaly region and the edge of the visible light anomaly region. If the included angle is less than the angle threshold and the average nearest distance is less than the distance threshold, then the region is determined to constitute an infrared-visual high-confidence correlation pattern, and this is used as the basis for determining the priority defect suspected region.

7. The method for non-destructive testing of concealed hollowness and peeling in historical buildings according to claim 1, characterized in that: Before performing adaptive fusion and feature extraction on the acoustic vibration response signal sequence and the infrared thermal imaging sequence, the following steps are also included: Based on the visible light image obtained from global pre-scanning, the texture complexity index of the surface of the historical building is calculated; Determine whether the texture complexity index is lower than a preset complexity threshold; The adaptive fusion and feature extraction of the acoustic vibration response signal sequence and the infrared thermal imaging sequence includes the following steps: If so, then adaptive fusion and feature extraction are performed on the acoustic vibration response signal sequence and the infrared thermal imaging sequence.

8. The method for non-destructive testing of concealed hollowness and peeling in historical buildings according to claim 7, characterized in that: After determining whether the texture complexity index is lower than a preset complexity threshold, Includes the following steps: If not, then based on the visible light image, identify decorative texture unit categories with different visual characteristics; For each type of decorative texture unit identified, based on its statistical performance in the infrared preview image, the texture thermal properties, including emissivity correction coefficient and base temperature offset, are estimated. The texture thermal properties are mapped onto a spatial grid with the same resolution as the infrared thermal imaging sequence to generate a texture thermal interference prediction map. The actual infrared thermal imaging sequence is compared with the texture thermal interference prediction map frame by frame to obtain the purified thermal imaging sequence. Adaptive fusion and feature extraction are performed on the acoustic vibration response signal sequence and the purified thermal image sequence.

9. The method for non-destructive testing of concealed hollowness and peeling in historical buildings according to claim 8, characterized in that: After performing frame-by-frame difference or ratio processing on the actually acquired infrared thermal imaging sequence and the expected texture thermal interference map to obtain the purified thermal image sequence, the following steps are also included: In the purified thermal imaging sequence, a flat surface area corresponding to a known material that is uniform and defect-free in the visible light image is selected as the verification reference area. The temperature standard deviation of the verification reference area in the purified thermal imaging sequence is calculated as the first residual noise index; and its temperature fluctuation range during the entire acquisition period of the infrared thermal imaging sequence is calculated as the second residual noise index. The adaptive fusion and feature extraction of the acoustic vibration response signal sequence and the purified thermal image sequence includes the following steps: If neither the first residual noise index nor the second residual noise index exceeds the corresponding preset noise tolerance threshold, then adaptive fusion and feature extraction are performed on the acoustic vibration response signal sequence and the purified thermal image sequence.

10. The method for non-destructive testing of concealed hollowing and peeling in historical buildings according to claim 9, characterized in that: After calculating the first residual noise index and the second residual noise index, the following steps are also included: If the first residual noise index or the second residual noise index exceeds the corresponding preset noise tolerance threshold, the purified thermal image sequence is subtracted from the texture thermal interference expected map frame by frame to obtain a serialized compensation residual map. In the compensated residual map, pixel regions where the absolute value of the residual is continuously higher than the residual threshold are identified and located as local areas of insufficient compensation. The texture thermal properties of the insufficiently compensated local areas are optimized in a targeted manner, and the expected texture thermal interference map and the purified thermal image sequence are updated based on the optimization results.

Images